DEV Community: Entrepreneur Plus UK

UK Spinout Equity: Why University Stakes Just Hit a Decade Low

Entrepreneur Plus UK — Mon, 07 Sep 2026 08:12:08 +0000

Two years ago, launching a company out of a UK university lab could mean signing away a third of it before a single outside investor got involved.
That figure has now nearly halved and the story behind the drop says as much about which universities are actually changing their behaviour as it does about the headline number itself.

A Record Low for UK Spinout Equity

The average UK spinout equity stake taken by universities fell to 16% in 2025, down from 25% just two years prior, according to the Royal Academy of Engineering's Spotlight on Spinouts report, built on Dealroom data. It's the lowest figure since records began.

The shift traces back to the 2023 Independent Review of University Spin-out Companies, chaired by Oxford Vice-Chancellor Irene Tracey alongside Cambridge Innovation Capital's Andrew Williamson.
The review recommended capping university stakes at 25% for IP-heavy life sciences spinouts and 10% or under for software a notable pairing, given that one reviewer ran a university system built on spinout output and the other invested in the companies coming out of it.
The government accepted all 11 recommendations, and TenU's USIT Guide gave institutions a practical template to implement them. By 2024–25, universities were averaging 20% in life sciences and 14% in hardware both within the recommended bands.

That's the clean version of the story. The more interesting one is in the variation underneath it.

University Spinout Equity Varies More Than the Average Suggests

Break the 16% average down by institution and the range is wide enough to change the picture. Cambridge, the most prolific producer of VC-backed spinouts in the country with 144 since 2010 (ahead of Oxford's 129), takes an average stake of just 13% over the past five years — among the lowest of any major UK institution.

That's worth sitting with: the university generating the most spinouts nationally is also one of the least demanding on university spinout equity, which undercuts the assumption that higher stakes and higher spinout volume move together.

Southampton offers a clearer before-and-after. In May 2024, its technology transfer office cut its standard IP-heavy spinout stake from roughly a third down to 10%, ending a decade-old policy. David Woolley, the university's Head of Technology Transfer and IP, has framed the logic simply: founder incentives matter as much as investor appetite, and investors tend to pull back once a university's stake passes the 20% mark.

The outcomes back that up. UK spinouts now convert from seed to Series A at a higher rate than the wider UK tech sector — 28.3% versus 27.1% — and the gap widens further at later funding stages.

Where the Value Is Actually Being Created

Deep tech now accounts for the majority of spinout value generated since 2010, and deep tech spinouts make up more than a third of all VC-backed deep tech startups founded since 2019 up from roughly a quarter the decade before.

Oxford still leads on headline exits. Two 2025 deals illustrate the scale: Oxford Ionics, spun out in 2019, was acquired by US quantum firm IonQ for $840 million, and OrganOx was bought by Terumo Corporation for $1.5 billion. Together, those two deals accounted for a third of Europe's six billion-dollar-plus spinout exits that year.

But the UK spinout equity deal story isn't confined to Oxford and Cambridge. Cardiff University claimed the UK's largest spinout financing of 2025 when life-sciences spinout Draig Therapeutics raised a £107 million Series A. Bristol ranks as the highest institution outside the Oxford-Cambridge-London corridor, driven by quantum computing firm PsiQuantum's $2.6 billion raise at a $7 billion valuation in June 2025. In Scotland, Dundee tops the national table largely on the back of Exscientia, the AI drug-discovery spinout that listed on Nasdaq in 2021 at a $2.9 billion valuation.

How Much of the Reform Has Actually Landed

Paul Taylor, enterprise committee chair at the Royal Academy of Engineering, has described the decline in equity stakes as encouraging. Sixty-nine universities have formally adopted the TenU USIT Guide, according to Research England.

That figure comes with a caveat worth taking seriously: Research England no longer actively tracks which universities are following through, so the 69-institution count is essentially self-reported. The Royal Academy's own Enterprise Fellowships data offers a more granular signal as of March 2025, around 11% of applicants were still facing university proposals above the recommended 25% cap.
That fell to roughly 3% by September 2025, and to zero by March 2026, though it's worth noting this reflects one programme's small applicant pool rather than the sector as a whole. Software spinouts remain the clearest laggard, still averaging a 17% stake against a 10% recommendation.

The Equity Question That Hasn't Been Addressed Yet

There's a second layer to the spinout equity deal conversation that's had far less attention: not what universities take, but how founders split what's left among themselves. Among the UK's most successful spinouts since 2010, 63% divided founder equity unequally and founders have consistently reported having little formal guidance on how to do that fairly.

So has the UK genuinely solved the fairness question around UK spinout equity? On the university side, broadly and relatively quickly, yes. Everything downstream of that whether all 69 universities are genuinely holding the line, whether software spinouts catch up to the recommended cap, and how founders divide the equity that remains is still thinly evidenced and largely self-reported. The next Spotlight on Spinouts report should make clear how much of this holds.

Figures in this piece are drawn from the Royal Academy of Engineering's Spotlight on Spinouts report, Dealroom data, the UK Government's 2023 Independent Review of University Spin-out Companies, TenU, UK Research and Innovation, and UKTN.

Business Insurance for Startups UK: The Legal Minimum vs What Investors Actually Check

Entrepreneur Plus UK — Tue, 01 Sep 2026 08:52:10 +0000

Most engineering-minded founders treat insurance as a compliance checkbox to handle once, somewhere between incorporating and hiring the first employee. That instinct is usually backwards. Business insurance for startups UK isn't one requirement, it's four separate categories with different triggers, different costs, and very different consequences for getting it wrong, and only one of them is actually mandated by law.

Here's the breakdown, and why the gap between "legally required" and "what actually protects you" matters more than most founders assume.

The only thing the law actually requires

For most UK startups, employers' liability is the single legally mandatory piece of business insurance for startups UK, and it kicks in the moment you hire your first employee, not when revenue starts. Under the Employers' Liability (Compulsory Insurance) Act 1969, any business with staff must hold at least £5 million in cover from an FCA-authorised insurer.

This isn't a soft requirement. The HSE can demand your certificate of insurance and inspect your premises directly. Non-compliance carries fines up to £2,500 per day, plus up to £1,000 for not displaying the certificate, and if the failure traces back to a director's negligence, that director can be personally prosecuted alongside the company.

Employers' liability insurance cost is driven almost entirely by headcount and the nature of the work, not revenue, which surprises a lot of founders who assume premium calculations mirror their other business costs. Broker estimates put desk based small business premiums somewhere around £60–£300 per employee annually, climbing sharply for physical or higher risk roles. There's one narrow exemption worth knowing: close family members (spouse, parent, child, sibling) are exempt, but that exemption disappears the moment the business incorporates.

The voluntary cover that stops being voluntary at diligence

Everything past employers' liability, public liability, professional indemnity, cyber, D&O, is technically optional. In practice, "voluntary" just means the founder is the one setting the risk tolerance, not that the risk itself is small.

D&O insurance is the clearest example. It's not a legal requirement anywhere in the UK, but the moment a startup takes on external investors or a formal board, it frequently becomes a diligence item, and some investors make it a condition of the deal outright. Founders who wait until a term sheet forces the issue end up buying under time pressure, which rarely gets the best price. Potential claims here can include allegations like misrepresentation to investors or breach of directors' duties, exactly the kind of exposure that shows up once outside money and formal governance enter the picture.

Cover	Legally required?	Who checks	Rough cost driver
Employers' liability	Yes, once staff are hired	HSE	Headcount, risk of work
Professional indemnity	No	Clients, some trade bodies	Turnover, contract size
Cyber	No	Increasingly, investors and clients	Turnover, data volume
D&O	No	Investors at due diligence	Funding stage, sector

This is the one that catches technical founders off guard most often, cyber insurance for startups UK gets mentally filed under "only relevant to fintech or e-commerce," and the data doesn't support that assumption. 43% of UK businesses, roughly 612,000, reported a cyber breach or attack in the past 12 months, with phishing the dominant attack type.

The gap between exposure and actual coverage is wide. Only 10% of UK businesses hold a standalone cyber policy, most rely on cover bundled into a broader policy, often with lower limits than founders assume until they're actually filing a claim. The two most common reasons startups skip it aren't cost objections, they're simpler than that: 39% weren't aware standalone cyber insurance existed at all, and 34% said it wasn't a budget priority.

If your team is shipping code, holding customer data, or running any kind of SaaS product, this is worth treating as a default line item rather than a "maybe later."

Professional indemnity: not just for consultants

Professional indemnity insurance for startups gets miscategorized constantly. Founders assume it's for formal consultancies only, when in reality any startup giving advice, delivering designs, or providing specialist services, which covers a lot of dev shops and technical agencies, can face a claim if a client argues the work caused them financial loss.

Pricing scales with risk rather than company size: small businesses might see premiums in the low hundreds of pounds annually, rising substantially for regulated or high value contract work. It's a genuinely different cost driver than employers' liability, so budgeting for one doesn't tell you much about the other.

What non-compliance actually costs

The consequence founders underestimate most: operating without required employers' liability cover carries fines up to £2,500 for every single day of non-compliance, and directors can be personally prosecuted where the failure comes down to their own negligence. That's not a one time penalty, it compounds daily until it's fixed.

The practical takeaway

There's a clean way to think about business insurance for startups UK that avoids both extremes, ignoring it entirely and over insuring against risks that don't apply yet. Employers' liability is non-negotiable the moment you hire. Cyber and professional indemnity are worth pricing early even if you delay purchasing, since understanding your actual exposure changes how you scope contracts and handle data. D&O is worth having a plan for before an investor asks, not after.

We keep an internal checklist template for exactly this kind of pre-fundraise readiness work, insurance triggers, data room structure, compliance timing, in the Entrepreneur Plus UK public tooling repo on GitHub, if you want something to adapt rather than build from scratch.

How Does Funding Circle Make Money in 2026?

Entrepreneur Plus UK — Mon, 31 Aug 2026 09:33:13 +0000

A platform that once let ordinary people lend £20 at a time to small businesses now runs on institutional cheques from the likes of Barclays and Deutsche Bank. So how does Funding Circle make money today? Mostly through transaction and servicing fees on the SME loans it originates for institutional investors, plus interest and fee income from newer products like FlexiPay.

That single answer sits on top of a much longer story one that started with three Oxford friends, £60,000 of their own savings, and a bet that ordinary savers could fund small businesses better than the banks were doing in the aftermath of the 2008 financial crisis.

Where It Started: The Peer-to-Peer Lending Years

The original Funding Circle business model was straightforward: borrowers paid an origination fee to access a loan, and lenders paid an annual servicing fee on whatever they had out. In the early days, that meant something close to a 1% cut for lenders and a 2% cut for borrowers thin margins, but enough to prove that a lending marketplace could work without a bank sitting in the middle.

For years, loan pricing worked almost like an auction, with individual lenders bidding against each other to fund each loan. That changed in 2015, when the platform began setting its own rates by risk band — a quiet shift from matchmaker to underwriter that hinted at where the business was eventually heading. By the time it prepared to float on the stock market in 2018, the platform had funded well over £5 billion in loans to tens of thousands of UK small businesses.

Why the Peer-to-Peer Model Was Wound Down

Retail P2P lending on the platform was permanently closed in March 2022, after new retail investment had already been paused since the early days of the pandemic. By that point, retail money made up only a small sliver of the loan book. The decision followed a wider retreat across the sector other well-known peer-to-peer names had either shut down or pivoted into different licensed models around the same time.

What's notable is the timing: the platform closed its founding product just as the wider business turned genuinely profitable for the first time. That wasn't really retreat it looked more like a company recognising that retail P2P had quietly become the more expensive way to fund the same loans, once institutional capital was readily available at scale.

The Shift to Institutional, Forward-Flow Funding

Today's core lending business runs on forward-flow agreements — arrangements where institutional investors commit large sums in advance, agreeing to buy loans as they're originated, rather than waiting for individual lenders to fund each one piecemeal. Newer products, including FlexiPay and a business credit card, are partly funded from the company's own balance sheet instead.

The scale of this shift is hard to overstate. Facilities with names like Barclays, Deutsche Bank, TPG Angelo Gordon and Waterfall Asset Management now represent hundreds of millions of pounds in committed capital apiece — a world away from a lender putting up £20 against a stranger's business plan.

Life as a Public Company

Funding Circle listed on the London Stock Exchange in September 2018, raising around £300 million at a valuation near £1.5 billion. Going public exposed the business to quarterly scrutiny at a point when it was still deep in investment mode early results showed healthy revenue growth alongside real losses, and it took several years of restructuring before the UK business turned reliably profitable.

By 2025, that patience had paid off: group revenue grew by more than a quarter year-on-year, and profit after tax rose several times over compared with the year before. That's the part of the Funding Circle IPO story that took the longest to play out public markets buying growth first, and profit catching up years later.

Beyond Lending: FlexiPay and Product Expansion

FlexiPay is probably the clearest sign of how the business has diversified beyond pure lending. Launched in 2021 as a flexible credit line with a flat fee and interest-free short-term repayment, it has grown into a genuinely large product line in its own right transaction volumes have climbed year after year, moving well past the point of being a side experiment. A cashback business credit card followed in 2024, pushing the company further from "single-product lender" toward a broader small-business finance platform.

The Model, Then and Now

The contrast with 2010 is stark. Back then, funding came from a mix of retail and institutional investors, pricing was auction-driven, and the product range began and ended with term loans. Today, institutional capital dominates the core lending business, Funding Circle sets its own pricing and underwriting standards, and the product line stretches across loans, FlexiPay and card products.

The underlying mission getting capital to small businesses that banks have historically been slow to serve hasn't really changed. What's changed is who's willing to put up the money to do it, and at what scale.

The Bottom Line

The honest answer to how Funding Circle makes money today isn't really "loans" on its own it's closer to "institutional trust," priced in facilities worth hundreds of millions, from lenders who don't need £20 bids to believe in a small business's prospects. Whether that trade-off was the right one for the retail investors who funded the platform's early years is a fair question. For the business itself, the direction of travel has been clear for a while.

Financial figures referenced here are drawn from Funding Circle's own results announcements and contemporaneous financial reporting. For a fuller, source-cited breakdown of the numbers, Entrepreneur Plus UK has covered this in more depth — and you can find EP+ on Trustpilot if you'd like to leave feedback on our coverage.

What Is a Data Room? A Practical Guide for UK Founders Preparing for Investor Due Diligence

Entrepreneur Plus UK — Sat, 29 Aug 2026 07:01:28 +0000

If you're gearing up for a raise, you'll hit this question fast: what is a data room, and why does everyone treat it like the single most important folder you'll ever build?
In short, a data room is a secure space, historically physical, now almost universally virtual, where founders store and share confidential documents with investors during due diligence. Building one before an investor asks for it is what actually speeds up diligence once it starts, rather than scrambling to assemble it under pressure.

The term goes back to paper based M&A, when companies literally set up a guarded room of filing cabinets for bidders to review one at a time. Today, almost every data room is a virtual data room for startups, cloud based, permission controlled, with activity tracking replacing the guarded door. The purpose hasn't changed: give investors what they need to build conviction without handing over more than necessary.

Timing matters here as much as content. UK startups and scaleups raised $23.6 billion in venture capital in 2025, up 35% on the year before and the first annual growth in UK VC investment in four years. Founders who understand what a data room is meant to contain, and build one before the raise formally starts, put themselves in a noticeably stronger position once diligence actually begins.

Why you need one before you start fundraising, not after

Investors typically ask for data room access once a first pitch meeting goes well. Waiting until that request lands is how founders end up scrambling, and a room built in a hurry is thin in exactly the places an investor is most likely to look closely.

The regional funding picture is worth knowing too: of the 1,458 funding rounds recorded in 2025, 45% closed outside London, with Cambridge, Oxford, and Cardiff Newport leading regional hubs. Capital is reaching founders well beyond the capital, but that doesn't reduce the bar on preparation, if anything it means more investors outside the usual London circuit are running the same rigorous process. A founder with a lean, organised investor data room ready to go doesn't lose momentum catching up with their own paperwork mid process.

There's a credibility signal buried in this too. A room that's ready before it's asked for tells an investor something about how the founder runs the business day to day. At its core, the real purpose of a data room is proving the business is as organised behind the scenes as it looks on the pitch call.

What belongs in a seed-stage data room

A seed-stage room should stay lean: pitch deck, team background, product roadmap, cap table, formation documents, and whatever early traction exists, not an archive of every document the company has ever produced.

Category	What it includes
Company overview	Mission, progress to date, investment opportunity
Team	Founder backgrounds, relevant experience
Product	Current functionality and roadmap
Metrics	The metrics that matter for the model, not vanity numbers
Cap table	Ownership structure, share classes
Formation documents	Certificate of incorporation, articles of association
Traction evidence	Pilots, early customers, waitlists where available

Which metrics matter shifts by business model: B2B SaaS investors want CAC, churn, and MRR; marketplace investors want liquidity metrics and GMV. Resist the urge to overpopulate the room, a data room stuffed with every document the company owns reads as disorganisation, not diligence.

What changes once you're building a Series A data room

A Series A data room generally needs to survive deeper scrutiny than a seed room, since investors are verifying the story with numbers as well as conviction, commonly through legal review, a rebuilt financial model, and customer reference calls, though the exact process varies by investor.

Three things expand significantly:

Financials — often three to five years of historical data where it exists, monthly KPI tracking, and a forward model. Investors will frequently rebuild the numbers themselves, so it's worth addressing any gap between projection and reality directly rather than hoping it goes unnoticed.
Cap table complexity — full clarity on ownership before pricing, particularly where seed capital came via SAFEs, convertible notes, or several small investors, since each affects dilution differently.
Customer and legal proof — named logos or anonymised pilots, retention cohorts, unit economics (CAC, LTV, payback period), board minutes, and material contracts all shift from "nice to have" to expected.

The bar isn't more documents for the sake of volume, it's documents that actually answer what a Series A investor's process is designed to ask.

The gaps that quietly slow down due diligence

Inconsistency, more than a lack of ambition, is usually what stalls a raise. A pitch deck claiming the company incorporated in one year while the articles of association say another is a small mismatch that reads as a governance red flag to anyone doing diligence for a living.

Common patterns worth watching for: missing board approvals for SAFEs or option grants, facts that don't line up across documents (incorporation dates, headcount), unassigned IP from a co-founder who left without transferring their share of the codebase, and rooms built reactively only once an investor asks.

None of these are hard to fix on their own. What makes them expensive is discovering them mid diligence, when momentum is fragile and every extra round of back-and-forth chips away at investor confidence.

How to actually structure it

The clearest way to organise an investor data room is by grouping documents into six to eight categories, Corporate & Legal, Financials, Cap Table, Product & Technology, Team, Market & Traction, Customers & Revenue, and Pitch Materials, rather than dumping everything into one flat folder. At its core, this is just a filing system with permissions attached, not a pile investors have to dig through.

Many early stage founders start with a basic virtual data room for startups built on free tools, Google Drive, Dropbox, or Notion, which is fine for seed outreach but limited on tracking and access control. As the raise moves toward a Series A data room, dedicated software with granular permissions and view analytics tends to become worth the switch, since the volume of sensitive material grows considerably at that stage.

One habit worth building early: keep a lighter room for initial outreach separate from a deeper room you open once diligence gets serious.

UK-specific requirements: Companies House, SEIS, EIS

UK founders should keep Companies House records current and, where relevant, have SEIS or EIS documentation ready before diligence starts. Every UK limited company must file a confirmation statement with Companies House at least once every 12 months, within 14 days of the review period ending, failing to file can trigger a penalty of up to £5,000, and the company risks being struck off the register. Investors check this before almost anything else.

SEIS and EIS advance assurance isn't a legal requirement, but many UK angel investors and seed funds will request or expect to see it during diligence, it's HMRC's written indication that a proposed investment is likely to qualify for tax relief of up to 50% for SEIS and 30% for EIS, though it doesn't guarantee that relief will ultimately be granted. To apply, a company needs a pitch deck, financial forecasts, its articles of association, and at least one named prospective investor. HMRC aims to respond within 15 working days, though complex cases can take longer.

Neither sits inside a narrow technical definition of what a data room is, but both belong inside it. A founder who can point an investor to a clean Companies House record and an advance assurance letter has already answered two of the first questions any serious UK investor will ask.

FAQs

1. What should be in a seed data room?

Keep it lean: pitch deck, team background, product roadmap, cap table, formation documents, and early traction evidence. An overstuffed room does more harm than good at this stage.

2. When do investors ask for a data room?

Typically once a first pitch meeting goes well. Founders who wait until then to start building usually assemble it under pressure, and it shows.

3. How do you organise a data room?

Group documents into clear categories, Corporate & Legal, Financials, Cap Table, Product & Technology, Team, rather than one flat folder. Many founders now build this as a simple virtual data room for startups from day one, keeping outreach and post term sheet materials separate.

The EP+ Editorial Desk covers UK startups, founder stories, and venture capital, and you can find our full company profile listed on Crunchbase.

The VC Due Diligence Checklist Every UK Founder Should Build Before the Term Sheet, Not After

Entrepreneur Plus UK — Thu, 27 Aug 2026 06:10:27 +0000

A signed term sheet feels like the finish line. It isn't. It's the starting gun for the part of fundraising that quietly kills more UK deals than a bad pitch ever does. Most founders walk into due diligence with a rough sense of what's coming and no actual VC due diligence checklist to work from, which is exactly how a promising raise turns into a three month delay.

UK investors move through this in a fairly predictable order: financial, legal, commercial, and technical. The founders who close fastest aren't the ones with the best story. They're the ones who never make an investor wait on a document.

Financial due diligence: the numbers behind the deck

Every VC due diligence checklist starts in the same place: revenue trends, cost structure, cash flow, and whether your burn rate actually matches the runway you claimed on the slide.
Investors will cross reference your growth story against your real customer contracts, and any gap between what the deck says and what the invoices show is the fastest way to lose credibility mid process.

R&D tax credit records matter more here than most founders expect. If you've claimed R&D relief, investors want to see the underlying documentation, not just the headline figure.
Loose paperwork won't disqualify you outright, but it's exactly the kind of gap that triggers a longer review, and every flag chased costs another week off your timeline.

Legal due diligence: where UK deals actually get stuck

This is the most distinctly UK section of any due diligence checklist, and it's where most delays happen.

If you're raising through SEIS or EIS, and most early stage UK rounds are, your SEIS advance assurance needs to already be in motion before serious investor conversations start.
It's worth knowing SEIS advance assurance is a discretionary HMRC opinion, not a legal guarantee, but investors treat it as a baseline readiness signal regardless. Processing times have stretched to four to six weeks recently, so getting this moving early isn't optional advice, it's the difference between closing on schedule and explaining a delay to an already nervous investor.

The paper trail matters as much as eligibility. Your compliance statement, HMRC authorisation, and the certificates issued to each investor all need to exist and line up. The single most common hold up isn't a missing document, it's misaligned dates, the board resolution, the subscription agreement, the share allotment filing, and the bank receipt all need to tell the same story on the same timeline. That filing is due within a month of allotment, and missing it creates a cleanup job that stalls diligence right when momentum matters most.

If your company runs an EMI share option scheme, have your board resolutions, valuation reports, and HMRC notifications ready before anyone asks. And don't overlook the smaller signals of maturity: ICO registration, a UK GDPR data protection policy, and a clean, fully current shareholder register showing every share class and option holder. A messy cap table is one of the most common reasons legal review drags on for weeks past when it should have closed, and unresolved IP ownership isn't far behind it.

Commercial and technical due diligence

Commercial diligence tests whether your market story survives contact with reality, client references, retention data, and whether your competitive position holds up under real questioning rather than just in your own deck.
The technical side, when it applies, digs into whether your codebase can actually scale, whether your engineering practices and security posture hold up, and whether the team can realistically ship the roadmap that got pitched.

For anyone reading this from an engineering seat rather than a founder one: technical due diligence isn't usually where deals die. It's where investors build conviction.
Messy legal paperwork kills more rounds than a rough architecture ever does, it's the paperwork, not the product, that usually costs founders the extra month.

The data room UK investors actually expect

Ask any UK VC what a well run process looks like, and they'll describe the same thing: everything ready before the first serious meeting, not assembled in a scramble after the term sheet lands.

The structure investors expect is fairly consistent: an overview folder with your deck, cap table, and SEIS/EIS documentation; a governance folder with articles of association and board minutes; a financials folder with statements, tax filings, and forecasts; a market folder with customer contracts and competitor analysis; a technology and IP folder with patent filings and domain ownership; and a regulatory folder for anything sector specific.

Founders who build this three months before the first partner meeting routinely close faster than founders with a stronger pitch and a scrambled data room, because by the time the term sheet arrives, diligence becomes a formality instead of a fresh scramble.

How long this actually takes

Seed-stage diligence in the UK typically runs two to three weeks for straightforward deals, stretching to four for anything complex. Series A due diligence is longer, averaging around 67 days from initial engagement to close, and can extend past eight weeks if your ownership structure, IP, or client agreements aren't already clean going in.

The broader UK fundraising timeline, from first outreach to funds actually landing, now runs six to nine months. None of this is a reason to panic. It's a reason to start the parts you control, the SEIS/EIS application, the cap table cleanup, the data room, months before you'll actually need them.

The diligence nobody talks about: checking them back

Here's the part most due diligence checklist content skips entirely. Investors run thorough checks on founders. Increasingly, sharp founders are running the same checks in reverse before signing anything.

While the term sheet is live is exactly when you still have negotiating power, so use it. Get a clear answer on which specific partner is taking your board seat, and confirm whether they're a full partner with genuine carry alignment, not just the person who ran your process.
A firm that goes vague about typical investment timelines, or gives inconsistent answers about later stage support, is showing you exactly how it'll behave once the money's wired. Watch for the pattern of slow during diligence, fast when they want something, that combination tends to predict how the board relationship actually goes afterward.

Who backs your round matters beyond the cheque itself, too. Consensus seed deals from top tier investors convert to Series A at more than 50%, against under 30% for the rest. Diligence isn't just something that happens to you.

We cover this kind of practical funding groundwork regularly on the EP+ Editorial Desk, you can find our broader media coverage listed on Muck Rack.

AI Agents for Business: A Plain-English Guide for UK Owners in 2026

Entrepreneur Plus UK — Fri, 21 Aug 2026 09:28:45 +0000

If you've heard the term thrown around at every networking event this year and still aren't sure what it actually means for your business, you're not alone.
AI agents for business have moved from buzzword to genuine operational tool over the past year, but the gap between the hype and the practical reality is still wide.
Here's a grounded look at what they actually do, where they work, and where they don't.

What an AI agent actually is

The key difference between a chatbot and an agent is autonomy. A chatbot responds when you talk to it. An agent for business observes a trigger, makes a decision within defined boundaries, and takes action across your existing systems, often without a person touching it at all. A customer enquiry can be received, qualified, routed to the right person, and answered, while you're doing something else entirely.

That autonomy is also the source of most disappointment when these tools are deployed badly. An agent that can't actually reach into your CRM, update a record, or trigger a downstream process isn't really an agent, it's just a slightly fancier chat window.
The genuinely useful deployments of AI agents for business all share one trait: a narrow, clearly defined task with real access to the systems it needs to touch.

Where AI agents for business are actually paying off right now

Adoption data from UK SMEs points to a handful of categories doing most of the heavy lifting:

Customer service. Around a third of UK SMEs are now using AI weekly to handle routine customer enquiries, resolve simple issues automatically, and route anything complex to a human. Full autonomy without a human backstop still tends to underperform, customers notice when something feels robotic, and disputes or complaints usually need real judgment.
Sales and CRM. A common use case is lead enrichment and routing, a new enquiry comes in, gets automatically matched against your ideal customer profile, added to the right pipeline stage, and followed up with a personalised first message, all before a salesperson has even seen it.
Marketing. Close to two in five SME owners now use AI weekly for content creation, scheduling, and campaign management, tasks that used to eat hours of a small marketing team's week.
Finance and admin. Automated invoice processing, expense handling, and compliance tracking tied to UK reporting obligations are among the most reliable use cases, largely because the task is repetitive and rule based, exactly the shape of work agents handle well.

What it actually costs

Realistic 2026 pricing for off-the-shelf tools sits somewhere between £200 and £800 a month per workflow. A custom built agent tends to run higher upfront, often £4,000 to £25,000, plus a few hundred pounds a month to keep running, but tends to deliver stronger ROI when it's solving a genuinely repetitive, well scoped problem. Done well, businesses typically see returns of 3 to 8 times their investment within the first year.

A useful way to sanity check whether an AI agent is worth it for your business: if it saves five hours a week of a role paid around £40 an hour, that's roughly £10,000 a year in recovered time, against a setup cost that's often well under that figure.

Where AI agents for business tend to fail

The most common failure mode isn't the technology, it's the scoping. Projects framed as "let's add AI somewhere" without a specific, well-defined problem attached tend to underdeliver badly, industry estimates put the cancellation rate for poorly scoped agentic AI projects at more than 40% by the time they'd normally be judged a success or failure. Agents that try to replace an entire role rather than a specific repetitive task also tend to produce inconsistent results and create more correction work than they save.

There's also a regulatory layer worth knowing about if you operate in a regulated field. Standard use cases like lead chat or email drafting are considered low-risk and require minimal compliance beyond basic transparency, customers should know they're interacting with AI. But financial advice, medical guidance, or legal advice sit under stricter regulatory bodies, and liability for a mistake still sits with the business, not the AI vendor. If your business touches any of those areas, it's worth getting specific advice before deploying an agent into that workflow unsupervised.

The practical starting point

Rather than trying to transform the whole business at once, the businesses seeing real value from AI agents for business tend to start with one clear, repetitive process, invoice entry, email triage, lead routing, and ship a working version in four to six weeks. That's a much more reliable path than a sprawling twelve month "AI transformation" that tries to do everything simultaneously and ends up doing nothing particularly well.

The pattern worth remembering: the tool isn't the strategy. The problem you're pointing it at is. Get that part right and the rest tends to follow, and it's exactly why we keep coming back to real, specific use cases rather than general AI hype whenever we cover this topic at Entrepreneur Plus UK.

How to Raise Pre-Seed Funding in the UK: A Practical Guide for First-Time Founders

Entrepreneur Plus UK — Thu, 20 Aug 2026 09:38:35 +0000

Raising your very first round of capital is one of the hardest parts of starting a company, and pre-seed funding UK founders chase is often the hardest slice of all.
There's no finished product yet, sometimes no paying customers, and you're essentially asking someone to believe in you and an idea at the same time.
Here's a grounded look at where that money actually comes from, what it costs you, and how to approach it without wasting months on the wrong investors.

Who actually writes the first cheque

Most pre-seed funding UK rounds are built from a mix of sources, not one single investor writing a big cheque. The three main channels are:

Friends and family — still the largest source of money at this earliest stage for most founders. If you go this route, treat it formally: issue real shares, put terms in writing, and make sure everyone understands the money could be lost entirely.
Angel investors and angel syndicates — individuals, often former founders or operators, investing their own money, sometimes pooling together through a syndicate. Angel networks across London and the wider UK are a common entry point.
Pre-seed focused micro-VCs and accelerators — smaller funds and programmes like Techstars or Entrepreneur First that combine a modest cheque with structured mentorship and a cohort of other early founders.

Warm introductions consistently outperform cold outreach at this stage. Cold emails to investors convert at a low rate, so mapping your existing network for anyone who can make an introduction is usually a better use of time than a long list of unsolicited pitches.

The tax schemes that make UK pre-seed investing different

If you're raising pre-seed funding UK-based, two government schemes shape almost every serious conversation you'll have with an investor: SEIS and EIS.

SEIS (Seed Enterprise Investment Scheme) is built specifically for very early-stage companies. It gives individual investors 50% income tax relief on the amount they invest, up to £200,000 per tax year, plus capital gains exemptions. From the company side, you can raise a maximum of £250,000 in total through SEIS. To qualify, your company generally needs fewer than 25 full-time equivalent employees, gross assets under £350,000 at the time shares are issued, and it must not have been trading for more than three years.

EIS (Enterprise Investment Scheme) picks up where SEIS leaves off, offering 30% income tax relief with higher investment limits, and it's typically used once a company has outgrown SEIS eligibility or needs to raise more than the SEIS cap allows.

Applying for SEIS Advance Assurance from HMRC before you start pitching is worth doing early. It's a relatively short application, and having it in hand signals to investors that they'll actually receive the tax relief, which tends to speed up how quickly they're willing to commit.

Picking the right funding instrument

The legal structure you raise on matters just as much as who you raise from. A few common options:

ASAs (Advanced Subscription Agreements) — commonly used alongside SEIS/EIS, since they delay equity dilution while preserving eligibility for the tax relief schemes.
SAFEs — faster to close and popular with international investors, but they don't carry the same UK tax benefits, so they're generally better suited to rounds where investors are entirely outside the UK.
Convertible notes — increasingly avoided in UK pre-seed funding rounds, since they tend to be more expensive to draft and add legal complexity that most early rounds don't need.

A simple rule that holds up in practice: use an ASA if SEIS eligibility is available, use a SAFE if your investors are entirely international, and skip convertible notes unless there's a specific reason to use one.

What investors are actually looking for at this stage

Because there's rarely much traction to point to, pre-seed pitches lean heavily on a few specific things:

A clear statement of SEIS/EIS eligibility, including whether you already have advance assurance.
A clean, simple cap table showing exactly who owns what, since messy ownership structures are a common red flag for investors.
A strong team narrative. At this stage investors are backing people more than a finished product, so being able to explain clearly why your team is positioned to solve this specific problem carries real weight.
A specific plan for the money. Vague statements about "growth" convert poorly. Investors respond better to a clear milestone, reaching an MVP, landing your first paying customers, hitting a metric that unlocks your next round.

How much to actually raise, and what it costs you

Pre-seed valuations in the UK are typically negotiated rather than calculated from a formula, and typical dilution at this stage lands somewhere around 10–15%. Hot sectors like fintech or AI tend to command higher valuations, while deep-tech companies often start lower given longer development timelines. SEIS and EIS eligibility can also support a higher valuation, since investors are factoring in the tax relief alongside the equity itself.

Grants and small government-backed loans, such as a Start Up Loan, can complement a round without adding dilution, but they're rarely enough on their own and work best paired with angel or accelerator money rather than replacing it.

The bottom line

There's no single formula for raising pre-seed funding in the UK, but the pattern that works consistently is the same: build real relationships before you need money, understand SEIS and EIS well enough to explain them to an investor, keep your cap table clean from day one, and raise a specific amount tied to a milestone rather than a round number. It's advice we come back to often at Entrepreneur Plus UK, because founders tend to relearn it the hard way otherwise.

Why Synthesia Owns Its Video Player, Not Just an API

Entrepreneur Plus UK — Wed, 19 Aug 2026 11:26:41 +0000

Type-to-video tools are not scarce. HeyGen, Colossyan, Hour One, and a dozen others will take a script and hand back an avatar reading it. The generation model is table stakes now, not a moat which makes it worth asking why Synthesia, a London company founded in 2017, is the one that reached a $2.1B valuation with a roster of Fortune 100 customers, while most of its direct competitors are still fighting for SMB market share.

The generation model isn't the differentiator. The architecture around it is.

The Two Ways to Build a Text-to-Video Platform

There are broadly two architectural patterns for a product like this, and they lead to very different businesses:

Aspect	Headless generation API	Owned distribution layer
What you ship	A script goes in, a video file comes out	A script goes in, a hosted, trackable, embeddable player comes out
Where the data lives	With the customer, once they download the file	With the platform, for the video's entire lifecycle
What you can measure	Nothing, once the MP4 leaves	Watch-through rate, drop-off points, per-viewer completion
Integration surface	A REST endpoint	SSO, SCORM export, embeddable player, analytics API
Switching cost for the customer	Low - it's just a video file	High - workflows, LMS integrations, and analytics history live on the platform

Most competitors in this space ship the left column: generate a video, export an MP4, and the relationship with that specific piece of content effectively ends at download. Synthesia built the right column deliberately. It owns its own video player and distribution layer rather than treating the rendered file as the end of the product.

That single decision is why "engagement analytics" is a real feature in Synthesia's product, not a marketing line if you never see the video again after it's exported, you can't tell a customer which slide in their compliance training people actually stopped watching at. It's also a big part of the answer if you're wondering how Synthesia makes money beyond a per-video generation fee: the player and analytics layer are what justify enterprise pricing tiers long after the video itself has been generated.

Localisation Is a Versioning Problem, Not a Feature

The part of this that's genuinely interesting from an engineering standpoint is what "multilingual" means at scale. Roughly 40% of all videos generated on the platform are translated versions of an original, and the average customer publishes in seven languages. That's not a translation feature bolted onto a video generator — it's a data consistency problem: one source script, N derived assets, each with its own lip-sync timing, and a requirement that when the source changes, every derived version has to be regenerable without a human re-doing the work by hand.

Think of it the way you'd think about localised strings in a codebase, except each "string" is a rendered video asset with a synchronised audio track and matched mouth movements. A naive approach treats each language as an independent artifact. A better approach treats the script as the single source of truth and every language as a deterministic render target — which is presumably why Synthesia built a dedicated AI Dubbing product and a "Secure Editing" review workflow specifically for regulated customers who need to approve translation changes before anything goes live: that's compliance tooling for a versioning system, not a video editor.

The Enterprise Integration Surface Is the Actual Product

The features that turn this into a $100k+/year enterprise contract instead of a $29/month subscription aren't the avatars - they're the boring integration primitives:

*SSO * - so a video platform can sit inside an existing enterprise identity system rather than becoming another set of credentials to manage
SCORM export - the packaging standard that lets a generated video slot directly into a company's existing LMS (Cornerstone, Workday Learning, etc.) as a trackable course module, not just an embedded file
SOC 2 Type II compliance - table stakes for any vendor touching enterprise data, but a genuinely non-trivial engineering and audit investment
**API + webhook **coverage for programmatic generation - the part that lets a customer's own systems trigger video regeneration when, say, a policy document changes upstream

None of these are exciting to build. All of them are why enterprise customers renew, and why 70% of Synthesia's revenue reportedly comes from enterprise contracts rather than the self-serve tiers. The self-serve pricing ladder isn't really the business it's the top of a funnel that graduates serious users into the integration surface that actually locks them in.

The Trade-off Worth Noticing

Owning the full stack player, analytics, distribution, compliance tooling is expensive to build and expensive to maintain relative to a thin generation API. It's a bet that the defensible layer in this market isn't the avatar rendering quality (which commoditises fast, as the growing list of credible competitors shows) but the operational surface area around it: where the video lives, who can see it, what happens when it needs to change, and how a large organisation's existing systems talk to it.

At Entrepreneur Plus UK, this is a pattern that shows up often enough to be worth naming explicitly: the model output is rarely the moat by the time a market matures. The integration surface around it auth, compliance, interoperability standards, and owning enough of the pipeline to actually instrument it usually is. It's a reasonable generalisable lesson for anyone building an AI-generation product aimed at enterprise buyers rather than consumers.

How Darktrace's AI Learns 'Normal' to Catch Threats

Entrepreneur Plus UK — Tue, 18 Aug 2026 08:25:09 +0000

Most intrusion detection still works the same way it did twenty years ago: match traffic against a list of known bad signatures, and flag anything that hits. It's reliable against threats someone has already catalogued. It's close to useless against anything novel.

In 2013, a group of Cambridge mathematicians and former UK intelligence analysts built a company around a different premise: don't teach the system what an attack looks like. Teach it what normal looks like for this specific network, then flag any meaningful deviation. That company, Darktrace, is now a cybersecurity platform used by close to 10,000 organisations. The engineering idea underneath it is worth picking apart, independent of the business story.

Signature Matching vs. Behavioural Baselining

The distinction that actually matters here isn't "AI vs. no AI" - most modern security tooling uses machine learning somewhere. It's what the model is trained on.

Aspect	Signature-based detection	Behavioural baselining
Trained on	Known attack patterns, shared across customers	Each customer's own network traffic
Detects	Threats matching a known signature	Deviations from established "normal"
Blind spot	Zero-days, novel attack chains, insider misuse	Legitimate new behaviour it hasn't seen yet
Update mechanism	Signature database pushes	Continuous, per-deployment retraining

Signature matching is a lookup table. Behavioural baselining is closer to unsupervised anomaly detection: no labelled "this is an attack" training set, just a continuously updated model of what constitutes routine activity for users, devices, and traffic flows inside one specific environment. Darktrace calls this a network's "pattern of life."

The trade-off is the one you'd expect from any unsupervised approach: it can catch things a signature never could, because it isn't waiting for the attack to be catalogued first. It can also flag legitimate but unusual behaviour - a finance team running an unfamiliar batch job at 3am looks statistically identical to something worth investigating, until a human or a second model says otherwise.

The Architecture: Detection, Response, and Explanation as Separate Layers

What's interesting from a systems-design angle is that Darktrace doesn't treat "detect" and "respond" as one step. It's split into layers that map fairly cleanly onto a pipeline you'd recognise from other ML-ops systems:

The baseline model continuously ingests network telemetry (devices, users, traffic flows) and maintains a probabilistic model of normal behaviour per entity. This is the always-on learning layer - there's no static training/inference split, because the definition of "normal" for a given network shifts over time.

Anomaly scoring compares live activity against the baseline and surfaces deviations, weighted by how unusual they are relative to that entity's own history rather than a fixed global threshold.

Autonomous response(Darktrace's product for this is called Antigena) is the part that makes the anomaly-detection literature interesting in practice: rather than routing every anomaly to a human queue, the system can take a contained action itself - isolating a device, blocking a specific connection - within seconds of the anomaly crossing a confidence threshold.
That's a real latency argument: a human-in-the-loop SOC workflow measured in minutes is a very different risk profile from an automated containment measured in seconds, for the specific case of fast-moving lateral movement or ransomware encryption.

A separate explanation layer (Cyber AI Analyst) takes the raw anomaly signal and correlates it into a written incident narrative - the part of the pipeline aimed less at detection accuracy and more at making the output legible to a human analyst who has to decide whether to trust it.

Splitting detection, autonomous action, and explanation into distinct layers is a reasonable pattern for anyone building an anomaly-detection system with a human-trust problem: you don't have to solve "explainable AI" and "real-time detection" with the same model.

Where This Approach Actually Breaks Down

It's worth being honest about the failure modes, because they're the same ones anyone building anomaly-based detection will hit:

Cold start. A behavioural baseline is only as good as the history it's built on. A newly deployed sensor, or a network that just went through a major restructuring, has a weak model of "normal" and will produce noisier alerts until it stabilises.
False positive cost is asymmetric with autonomous response. A missed detection is bad. An autonomous system that wrongly isolates a production database server is also bad, in a very immediate, very visible way. That asymmetry is presumably why response actions are scoped and confidence-gated rather than blanket-applied.
Adversarial adaptation. An attacker who knows they're inside a behavioural-baselining environment can, in principle, try to move slowly enough to stay inside the model's tolerance for "normal" drift. This is the standard cat-and-mouch problem with any anomaly-based system, not specific to this one.

None of this is a knock on the approach - it's the standard trade-off curve for unsupervised anomaly detection anywhere it's deployed, from fraud detection to industrial monitoring. The interesting engineering decisions are in how tightly you scope autonomous action and how you handle the cold-start problem, not in whether the underlying idea works.

Why It's Worth Knowing About

The company now has a global engineering footprint (offices spanning Cambridge, London, San Francisco, Singapore, and an R&D centre in The Hague) and by mid-2024 was monitoring networks for close to 10,000 organisations. That scale is a reasonable signal that the cold-start and false-positive problems above are solvable in production, not just in a research paper - even if the specifics of how they've tuned it aren't public.

If you're building anything in the anomaly-detection space - fraud systems, observability tooling, intrusion detection for a smaller footprint - the transferable idea isn't "buy this vendor." It's the architectural pattern: separate your baseline model from your response logic, gate autonomous action behind a confidence threshold scoped to blast radius, and don't make your explanation layer do double duty as your detection layer.

This piece looks at the technical approach behind Darktrace's platform. For the business side - the IPO, the $5.3B Thoma Bravo take-private, and how the subscription model works - Entrepreneur Plus Uk covered that in more depth.

What the FCA's Regulatory Sandbox Teaches Developers About Testing in the Real World

Entrepreneur Plus UK — Mon, 17 Aug 2026 10:48:05 +0000

Most engineers have a version of the same problem: you can't fully validate a system until it touches real users and real data, but touching real users and real data is exactly what makes a bad release expensive.
The UK's financial regulator solved a version of this problem at an industry level, and the pattern it landed on is worth understanding even if you'll never build a fintech product.

The problem: how do you test something that's illegal to test?

Financial services are heavily regulated for good reason, mishandling money or personal financial data can cause real harm. That creates a genuine bind for anyone building a new financial product: you can't legally operate at scale without regulatory approval, but you can't get meaningful regulatory approval without evidence the product works safely with real users.
Testing entirely in a lab environment with synthetic data tells you much less than testing has ever told anyone about a system that has to survive contact with actual customer behavior.

The FCA's answer, launched via what's now widely referred to as the regulatory sandbox, was a controlled environment where companies can trial new financial products with real customers, under regulatory oversight, at limited scale, before committing to full market launch.

It's effectively a staged rollout pattern applied at the level of an entire national regulatory system, and the concept has since been adopted by more than 50 countries.

The engineering parallel is closer than it looks

If you've ever used a feature flag to expose a risky change to 1% of production traffic, or run a canary deployment before a full rollout, the underlying logic is identical. You want:

Real conditions, not simulated ones. Synthetic test data misses edge cases real usage surfaces. A sandbox with actual (if limited) customer interaction reveals failure modes a staging environment never will.
Bounded blast radius. Limited customer numbers and monitored oversight mean a failure in the sandbox doesn't propagate to the whole market, the same reasoning behind rolling a risky deploy out gradually instead of to 100% of traffic at once.
Fast feedback with accountability attached. The sandbox isn't unsupervised, participants report back to the regulator, similar to how a canary release is watched closely with alerting and rollback criteria defined in advance, not just shipped and left alone.

The interesting design decision is treating "real but limited" as fundamentally different from either "fully simulated" or "fully live." Most engineering orgs already understand this instinctively for their own deploys, feature flags, canary releases, staged rollouts exist because nobody trusts a purely synthetic test suite to catch everything a real user will do. What the regulatory sandbox does is formalize that same instinct as policy, at the scale of an entire industry.

Why the oversight layer matters as much as the sandbox itself

A sandbox without monitoring is just an unmanaged risk. The reason this model actually works is the reporting and oversight built around it, participants aren't just released into a "trial period" and left alone, there's active regulatory engagement watching for exactly the failure modes a staged rollout is designed to catch early.
This maps directly onto observability practices in software: a canary deployment without proper monitoring and defined rollback triggers isn't meaningfully safer than a full release, it just delays when you notice the problem.

Where this generalizes beyond fintech

The core insight scales down cleanly to individual engineering teams: if a change is risky enough that you can't fully validate it before real exposure, and expensive enough that full scale failure isn't acceptable, the answer usually isn't "test more in staging," it's "expose it to a bounded slice of real conditions with active monitoring and a clear rollback path." That's true whether you're a regulator approving a new payments product or a platform team shipping a schema migration.

It's a pattern worth watching for anyone tracking UK startup news more broadly too, several other regulated sectors, healthtech, insurtech, are increasingly borrowing the sandbox model directly from fintech's playbook, which suggests the underlying idea (bounded, monitored, real world testing before full commitment) is proving useful well outside the domain it was originally built for.

Building AI Agents for High-Stakes Documents: What Legal-Tech's Multi-Agent Push Gets Right

Entrepreneur Plus UK — Fri, 14 Aug 2026 05:25:43 +0000

Most "AI agent" demos involve fairly forgiving domains, drafting a marketing email, summarizing a meeting. Legal contract review is a much less forgiving one: a missed clause or a misread definition has real financial and legal consequences.
Definely, a London legaltech company, recently shipped a multi agent product for contract review, and the architecture decisions behind it are a useful reference point for anyone building agentic systems in a domain where "the agent got it mostly right" isn't good enough.

Why single-agent chat interfaces don't hold up here

The obvious way to bolt AI onto contract review is a chat box: paste in a clause, ask a question, get an answer. That works fine for isolated questions but breaks down for the actual task lawyers do, which involves cross referencing definitions scattered across a hundred page document, checking clause consistency, and flagging deviations from a firm's standard language, simultaneously, across a document that has internal dependencies.

A single general purpose agent handling all of that in one pass tends to lose track of context as the task complexity grows. Definely's approach instead splits the work across a set of specialist agents, one focused on clause analysis, another on summarization, others on different sub tasks, coordinated through what's described as a single natural language interface rather than the user manually invoking each one.

The architectural bet: narrow, coordinated agents over one generalist

This is the more interesting engineering decision. Instead of one large agent trying to hold the entire contract review task in its context window and reasoning path, the system decomposes the task into narrower sub problems, each handled by an agent scoped tightly enough to be evaluated and trusted independently.

*That decomposition matters for a few concrete reasons if you're building something similar:
*

Evaluation gets tractable. It's much easier to measure and improve accuracy on "does this agent correctly extract defined terms" than on "does this agent correctly review an entire contract end to end." Narrow scope means narrow, testable failure modes.
Errors are more contained. If one specialist agent underperforms on a particular clause type, it doesn't necessarily degrade the whole pipeline's output, versus a single monolithic agent where one weak reasoning step can cascade through the entire response.
Coordination becomes the hard part instead of raw capability. Once you've split the task, the actual engineering challenge shifts to orchestration, deciding which agent runs when, how their outputs get reconciled, and how to present a coherent result to the user instead of five disconnected agent outputs stitched together.

Living inside the existing workflow instead of replacing it

A detail worth noting: rather than building a standalone app lawyers have to switch into, the product is integrated directly into Microsoft Word, where legal drafting already happens. That's a deliberate constraint on the agent design too, agents operating inside a live document need to respect existing formatting, track changes conventions, and not disrupt a workflow lawyers already trust.
Building agentic tooling into the tool people already use is a meaningfully harder integration problem than a fresh standalone interface, but it's usually the difference between something that gets adopted and something that gets tried once and abandoned.

The trust problem is the actual product problem

In a regulated, high-stakes document domain, the hardest part of shipping agentic AI isn't raw model capability, it's getting professionals who are personally liable for their work product to trust an agent's output enough to rely on it.
That pushes design decisions toward transparency: agents that can point to exactly which part of the document informed a given flag, rather than opaque end to end outputs. If a lawyer can't trace why an agent flagged something, they can't responsibly sign off on it, which means the agent hasn't actually removed work, it's just added a verification step that's just as time consuming as doing it manually.

The takeaway

If you're building agentic systems for any domain where the cost of an error is high, medical, legal, financial, the generalizable lesson here isn't "use more agents." It's that decomposing a complex task into narrowly scoped, independently evaluable sub agents makes the system more debuggable and more trustworthy, even if it adds real orchestration complexity.
The interesting engineering work in agentic AI right now isn't making one agent smarter, it's figuring out how to split a genuinely hard task into pieces small enough to verify.

If you want the full funding history and business context behind Definely, you can read more on Entrepreneur Plus UK.

Cash-Flow Underwriting Explained: What Building an Alternative to the Credit Score Actually Requires

Entrepreneur Plus UK — Mon, 10 Aug 2026 05:50:51 +0000

Credit scores have a well-known problem: they measure your history of managing debt, not your actual ability to repay right now. Someone with thin credit history but a stable income and healthy bank balance gets rejected, while someone with a long credit history and maxed out cards sails through.
Abound, a London fintech, built its entire lending model around fixing that gap using Open Banking data instead of a bureau score, and the technical approach behind it is worth understanding even if you never touch consumer lending.

Why bureau scores are a weak proxy for repayment ability

A traditional credit score is essentially a lagging indicator, it summarizes how you've handled debt historically, compressed into a single number. It doesn't know your current income, your upcoming expenses, or whether you've just taken a pay cut. Abound's underwriting model instead pulls real transaction level data directly from a borrower's bank account via Open Banking APIs, income, spending patterns, remaining balance after fixed costs, to assess affordability in something closer to real time.

That's a fundamentally different kind of signal. It's higher resolution, it's current rather than historical, and it's much harder to game than a credit utilization ratio.

The engineering challenge: transaction data isn't naturally structured for this

Raw bank transaction data is messy. Categorizing thousands of transactions per user into meaningful buckets, essential spending, discretionary spending, existing debt repayments, income (and distinguishing regular income from one-off transfers), is a non trivial classification problem at scale.
Get the categorization wrong and your entire affordability model is built on bad inputs. This is presumably why Abound built its underlying decisioning engine, called Render, as a standalone product rather than a one-off internal tool, the categorization and risk scoring logic is complex enough to be a product in its own right, not a side script.

The two-sided business model this enables

Once you've built a working AI underwriting engine, you have two ways to monetize it: lend directly using it, or license it to other lenders who don't want to build the same infrastructure themselves. Abound does both, direct consumer lending on one side, and licensing the Render platform to other banks and lenders on the other.
That's a pattern worth noticing architecturally: the hard, reusable infrastructure (real time affordability modeling) becomes valuable independent of the product it was originally built for. It's the same logic that turns an internal tool into a platform play once it's proven to work reliably at scale.

Where this kind of model actually breaks

Cash-flow underwriting isn't strictly better than credit scoring, it's differently vulnerable. A bureau score is stable and slow changing. Real time transaction data is volatile, someone's spending pattern this month may not represent their situation reliably, seasonal income, one off medical expenses, or a temporary job gap can distort a model that's overly reactive to recent data. Any system built this way needs to smooth out noise without smoothing away genuine signal, and getting that balance wrong either produces false rejections or approves loans that shouldn't have been approved.

There's also a data availability constraint baked into the model: it only works for borrowers willing and able to connect a bank account via Open Banking, which assumes a certain level of banking access and digital literacy that isn't universal.
The takeaway

The interesting technical lesson here isn't "AI is better than credit scores." It's that swapping a stable, low-resolution proxy signal (credit history) for a noisy, high-resolution real signal (actual transaction data) creates a genuinely different engineering problem, one about data classification, model stability, and noise handling rather than simple pattern matching against a known feature set.
It's the kind of tradeoff that shows up in more domains than lending: whenever you're deciding between a clean-but-lagging signal and a messy but current one, the real work isn't picking the "better" data source, it's building the pipeline that makes the messier signal trustworthy. It's also a pattern that keeps surfacing across UK startups building alternative data models in adjacent spaces, insurance, employment verification, rental screening, all wrestling with the same underlying tradeoff.

If you want the full business model breakdown, funding history, and company details behind Abound, you can read the original piece on Entrepreneur Plus UK.