Even on the most conservative count, the average company runs well over 80 SaaS applications. BetterCloud's 2025 State of SaaS puts it at 106, down from a 2022 peak of 130, while Zylo's 2026 SaaS Management Index, which discovers spend instead of surveying IT, puts the average portfolio at 305. The two methodologies disagree by a factor of three, so treat any single number here with suspicion. What they agree on is the waste: Zylo found organizations leave 36% of their SaaS licenses unused, worth an average of $19.8M a year against $55.7M of annual spend. Companies are also paying more for the same tools, with total SaaS spend up 8% year over year while app counts fell 0.07%.
For SMEs and post-PMF founders, the question is no longer "which SaaS should we add" but "which ones can we kill, and which ones should we rebuild from scratch?" This playbook answers that — vendor-neutral, with real cost math, the decision framework we use on every NodeSparks engagement, and the failure modes nobody warns you about.
TL;DR — Replacing SaaS does not mean cloning HubSpot. It means rebuilding the specific single-purpose workflows you actually use — outreach, scheduling, content engagement, basic CRMs — as custom systems you own. Above ~$300/month with a stable 6-month-old workflow, custom typically pays back inside 18 months. Below that, stay on the SaaS. The full decision is captured in the SaaS Replacement Matrix.
Key Takeaways
- Above ~$300/month with a stable, 6-month-old workflow, a custom replacement typically pays back inside 18 months; below that, staying on the SaaS is the cheaper call.
- App-count estimates vary by method: 106 average (BetterCloud, 2025) versus 305 (Zylo, 2026). Both clear 80.
- Median SaaS spend is $9,455 per employee per year, and 36% of licenses sit unused (Zylo, 2026).
- Companies in the 501-2,500 employee band waste an average of $9.5M a year on unused licenses (Zylo, 2026).
- The real win is a fixed-floor cost plus owning the repo, the data, and the workflow, instead of a per-seat bill that scales with headcount forever.
What does "replace your SaaS" actually mean?
Replacing SaaS does not mean writing a clone of HubSpot in a weekend. It means identifying the single-purpose tools in your stack — schedulers, outreach senders, scraping subscriptions, basic CRMs, content engagement platforms — and rebuilding the specific workflow you actually use as a custom system you own.
The replacement system typically consists of:
- A small backend (Python or TypeScript) running on cheap infrastructure — Render, Railway, fly.io, or a $20/month VPS
- One or two LLM API calls per workflow step, billed per token (OpenAI, Anthropic, or Gemini)
- A database — usually Postgres on Neon or Supabase free tier, sometimes just SQLite
- A thin UI for the operator — a Next.js dashboard or a Slack bot for the people who use it
- A cron schedule or webhook trigger
That is the entire stack for replacing a $200-$1,500/month subscription. The system runs on your infrastructure, holds your data, exposes your APIs, and costs the API bill — typically 5-25% of the SaaS price.
How big is the SaaS sprawl problem in 2026?
The numbers from neutral SaaS-management vendors are less consistent than they look, because the two main publishers measure on incompatible bases. BetterCloud surveys IT teams. Zylo discovers spend from financial systems. That is why their app counts differ by roughly 3x, and why anyone quoting a single tidy range is blending two things that do not blend:
| Metric | Figure | Source |
|---|---|---|
| Average SaaS apps in use (IT survey basis) | 106, down from 112 | BetterCloud, 2025 |
| Average SaaS portfolio (spend-discovery basis) | 305 average / 240 median | Zylo, 2026 |
| Median SaaS spend per employee per year | $9,455 | Zylo, 2026 |
| Licenses left unused | 36% | Zylo, 2026 |
| Annual waste, 501-2,500 employee band | $9.5M | Zylo, 2026 |
| Share of SaaS spend controlled by business units, not IT | 81% | Zylo, 2026 |
| AI-powered apps per organization | 27, about 22% of the stack | BetterCloud, 2026 |
One caution if you go looking at these reports yourself: Zylo reported an average of $4,830 per employee in its 2025 index and a median of $9,455 in 2026. That is a change of metric, not a doubling of spend, and it gets misquoted as growth constantly.
| Tools with feature overlap (paying twice) | 30-40% of spend | BetterCloud |
Read those rows together: roughly one third of every dollar a typical SME spends on SaaS pays for unused seats or duplicated features. That is the budget the replacement playbook recovers.
The build vs buy decision framework
Not every SaaS should be replaced. The decision turns on three variables: workflow specificity, integration depth, and vendor moat. We use this matrix on every prospect call.
| Variable | Stay on SaaS | Replace with custom |
|---|---|---|
| Workflow specificity | Complex, multi-team, evolving | Single-purpose, stable for 6+ months |
| Integration depth | Embedded in 5+ external systems | Talks to 1-3 APIs |
| Vendor moat | Network effect (Stripe, Slack), regulatory (Gusto) | Thin wrapper around an API or database |
| Data volume | Tens of millions of records, real-time | Thousands to low millions, batch OK |
| Compliance burden | Vendor-attested SOC 2 / HIPAA required | Your existing infra already covers it |
| Total cost of ownership over 3 years | Lower than replacement build + hosting | Higher than replacement build + hosting |
If three or more rows on the right apply to a tool, it is a replacement candidate. If three or more on the left apply, leave it alone — the math will not work.
What categories almost always fail the SaaS test
These are the categories where, in our experience, custom replacement wins inside 18 months:
- Outbound outreach platforms (Apollo, Lemlist, Instantly) — thin UIs over SMTP and a contact database (build your own outreach agent)
- Scheduling tools (Calendly Premium tiers, Chili Piper) — calendar API + a form
- Content scheduling and engagement (Buffer, Hootsuite) — social APIs + cron
- Basic CRMs for small teams (HubSpot Starter, Pipedrive) — Postgres + a Kanban view
- Scraping subscriptions (Phantombuster, Apify) — Playwright + a queue
- Analytics dashboards on top of your data (Mixpanel for small teams) — SQL + Metabase
- AI-wrapper tools that just call an LLM — direct API call costs 1/10th the price (the AI-wrapper SaaS trap)
What categories almost always fail the build test
- Stripe and payment processors — regulated, embedded, network effect
- Email infrastructure (Postmark, SendGrid) — deliverability is the product
- Auth providers (Clerk, Auth0) — security boundary, not worth owning
- Full-suite ERPs and payroll — compliance is the product
- Slack, Linear, Notion — collaboration tools where your team is the moat
What does it cost to replace a $300/month SaaS?
Here is the cost math worked end to end for a typical replacement. Numbers come from real NodeSparks engagements; we anchor them to public API pricing so you can verify each line.
Worked example 1 — replacing a $300/month outreach platform (Apollo Basic + Lemlist)
| Line item | Cost |
|---|---|
| One-time build (4-6 weeks) | $6,500 |
| Hosting (Render starter tier) | $7/month |
| OpenAI API for personalization (~5,000 emails/month at GPT-4o-mini) | $12/month |
| Email infrastructure (Postmark or SES) | $25/month |
| Lead enrichment API (one of Apollo Basic, Hunter, or proxycurl, pay-as-you-go) | $30-$60/month |
| Database (Neon free tier covers it) | $0 |
| Total ongoing cost | $74-$104/month |
Break-even: month 24 against $300/month SaaS. Lifetime savings (year 3 onwards): $2,400-$2,700/year. The system is yours forever.
Worked example 2 — replacing $400/month of LinkedIn + content tools (Hootsuite + Phantombuster + a content calendar SaaS)
| Line item | Cost |
|---|---|
| One-time build (3-4 weeks) | $5,000 |
| VPS (Hetzner, $8/month) | $8/month |
| LinkedIn engagement runs via Playwright + your own residential proxy | $20/month |
| OpenAI API for content generation/editing | $15/month |
| Storage (S3 or backblaze, low usage) | $3/month |
| Total ongoing cost | $46/month |
Break-even: month 14 against $400/month SaaS. After year 1, you save the full subscription cost. The system runs every weekday, posts to two LinkedIn accounts, drafts 3 weekly newsletters, and feeds a Slack channel with engagement opportunities.
Worked example 3 — replacing $1,200/month of stack consolidation (HubSpot Starter + Calendly Premium + a reporting tool)
| Line item | Cost |
|---|---|
| One-time build (6-8 weeks) | $14,000 |
| Hosting (Railway or Render) | $25/month |
| OpenAI API for write-ups, summaries, lead scoring | $40/month |
| Postmark for transactional email | $15/month |
| Database (Neon Pro for backup retention) | $20/month |
| Total ongoing cost | $100/month |
Break-even: month 12-13 against $1,200/month. Year 2 savings: ~$13,200. The system holds the contact database, runs the booking flow, sends the follow-ups, generates the weekly leadership report, and routes inbound to the right person.
The pattern across every engagement: one-time cost of 6-12 months of the SaaS subscription, then 5-25% of the SaaS cost ongoing. Break-even sits between month 12 and month 36. After that, the savings compound and the asset is yours.
When should you NOT build?
This is the section every "replace your SaaS" pitch skips. The cases where SaaS still wins:
When the SaaS is the system of record for a regulated process
If the tool is the audit trail your accountant or your DPO depends on, do not rebuild it. Payroll, accounting, e-signature with legal holdup — the vendor's compliance posture is the product. You will not save enough to justify owning the audit risk.
When you are pre-product-market-fit
If you are still figuring out the workflow, SaaS is the right answer. You need to swap tools every quarter. Custom builds reward stability — the workflow has to be set for 6+ months before the math works. Building too early just locks you into your wrong guess.
When the vendor genuinely has a network effect
Stripe and Slack are not subscriptions you replace. Stripe is plugged into the global financial system; Slack is plugged into your team. The product is the network, not the feature set.
When the SaaS provides ongoing capability you cannot maintain
If the tool only works because the vendor maintains 50 integrations updated weekly (Zapier, Make, n8n cloud), and you only need a few of them — the math might still favor SaaS. The honest comparison: how many integrations do you actually use? If it is fewer than five, custom usually wins. If it is fifteen, stay on Make or n8n.
When the volume is too low to amortize
If you spend $40/month on the tool, the build cost will never amortize. The replacement playbook starts paying off above roughly $150-$200/month, depending on complexity.
How do you actually run the replacement?
The workflow we use on every engagement, mapped step by step.
Step 1 — audit the actual stack (week 0)
Pull the company credit card statement and list every recurring SaaS charge. Add the tools paid via PayPal, personal cards, and team-purchased subscriptions Finance does not see. The number is always at least 30% higher than what IT thinks. This is the input to the decision matrix.
Step 2 — score every tool on the matrix (week 0-1)
For each tool, walk down the six rows of the matrix above. Mark each row "stay" or "replace". Three or more "replace" votes promotes the tool to a candidate. Sort candidates by monthly spend, descending.
Step 3 — pick one tool, not five (week 1)
The most common mistake: trying to consolidate the whole stack in one build. Pick the single highest-spend candidate where you also have a clear, stable workflow. Ship that one. Measure the savings. Then move to the next.
Step 4 — write the workflow before you write the code (week 1-2)
Every SaaS replacement starts with a workflow doc. What triggers the tool? What does it do? What does it output? What systems does it touch? If the doc fits in two pages, the build is straightforward. If it does not, the SaaS is doing more than you think — go back to the matrix.
Step 5 — build in weekly increments (weeks 2-6)
Ship something runnable at the end of every week. Week 1: data ingestion. Week 2: the LLM/workflow logic. Week 3: the output and human-in-the-loop UI. Week 4: integration with existing tools. Week 5: shadow-run alongside the SaaS. Week 6: cancel the SaaS.
Step 6 — shadow-run for two weeks before cancelling
Never cancel the SaaS the day the custom build ships. Run them in parallel for two weeks. Compare outputs. Find the edge case the workflow doc missed. Then cancel.
Step 7 — own the maintenance (month 1 onwards)
Allocate one engineer (or one ops lead with engineering taste) for 1-3 hours per month for the first six months. Document the system. Put it in your runbook. Make sure two people on the team know how to restart it. After six months, the system runs on its own and costs you the API bill.
Common pitfalls (and how we avoid them)
Pitfall 1 — replacing the tool, not the workflow
The SaaS often does seven things; you use three of them. Replacing all seven in a custom build wastes 60% of the budget. Build only what you actually use. If a feature has not been touched in three months, it does not need to ship in the replacement.
Pitfall 2 — picking the wrong layer to replace
Outreach platforms are made of three layers: data, personalization, sending. If your only pain is data quality, do not rebuild the sender. Replace the data layer with a scraper + enrichment API and keep the SaaS for sending. Layered replacement is cheaper and faster.
Pitfall 3 — over-engineering the LLM layer
Half the AI features SaaS vendors charge $200/month for are a single GPT-4o-mini call costing $0.0001. Do not build a vector database, fine-tuning pipeline, and agent orchestration framework when one prompt and one API call solves the problem. Start naive. Optimize when traffic justifies it.
Pitfall 4 — skipping the human-in-the-loop UI
The custom build needs an interface for the operator who runs it. A Slack bot, a tiny Next.js dashboard, even a Google Sheet. Every replacement we have shipped without one has failed to stick — the team reverts to the SaaS within 60 days because the new system is invisible.
Pitfall 5 — building before negotiating the renewal
Sometimes the cheapest move is calling your account manager and asking for a 40% discount. Many SaaS vendors will discount up to 50% to retain a client at renewal time, especially below $1,000/month MRR. Negotiate first. If they refuse and the matrix says replace, replace.
What does owning the tool mean, legally and technically?
Three things are true after a NodeSparks-style replacement, and they should be true after any custom replacement:
- The repo lives on your GitHub organization. The full source code, with infrastructure config, database migrations, and deployment scripts. Not "we host it for you" — the repo is yours.
- The data lives on your infrastructure. Whatever cloud account you already use, with your existing security posture, your DPAs, your backups. No vendor sees your contacts, leads, or revenue numbers.
- The system runs without us. Documentation, runbooks, monitoring dashboards, on-call playbook. If we vanished tomorrow, the system would still run, and any competent engineer could maintain it.
That is what "own your stack" actually means. The opposite of vendor lock-in is not "we host it for you on better terms" — it is "you can host it yourself, modify it yourself, and audit it yourself, with no permission required."
What is the best first SaaS to replace?
For SMEs we work with, the answer is almost always one of three:
- Outreach platform — highest spend, simplest workflow, biggest immediate savings
- Reporting tool — most painful to maintain manually, biggest time savings
- Content scheduling + engagement stack — most overlapping, biggest consolidation win
These three categories account for roughly 70% of the SaaS replacement work we do. Pick the one with the highest monthly bill and the most stable workflow. Ship it in three to six weeks. Apply the savings to the next replacement.
The two-lane positioning of NodeSparks — replace your SaaS, automate your daily ops — exists because every SME owner we have talked to is paying for both: tools they do not own, and manual work no tool solves cleanly. The playbook above attacks the first half. The strategic pillars on the home page cover both.
Cited data and frameworks
The frameworks in this pillar are also published as standalone canonical pages, easier to cite directly:
- The SaaS Replacement Matrix — the six-row build-vs-buy decision matrix used throughout this essay
- The 5-Question Build-vs-Buy Rubric — the short-form intake version
- Automation Platform Pricing 2026 — vendor-neutral cost data for the "cost trajectory" row of the matrix
- AI SDR & Outreach Tool Pricing 2026 — same shape for sales tooling
All four published under CC BY 4.0 — free to cite and adapt with attribution.
Where to go next
This pillar is the entry point. The cluster covers individual replacements in depth, with the same cost math and the same workflow:
- Apollo.io alternative: build your own outreach agent
- Zapier vs n8n vs Make vs custom code: the honest comparison
- n8n alternatives — and when to skip n8n entirely
- The AI-wrapper SaaS trap: when $75/month is robbery
- What is an AI SDR? The build-vs-buy read
- Replace Bill.com with a Slack-native invoice agent
- Clay alternative: a custom enrichment agent you own
- Calendly alternative: a custom scheduling agent
- DIY cold email stack: replace Lemlist, Clay, and Calendly
- The Ops Automation Playbook: which manual work to automate first
New posts ship weekly. If you want to talk through your specific stack and which tools are replacement candidates, book a 30-minute call.
You will leave the call with a one-page replacement plan whether or not you hire us.
Sources (retrieved 2026-08-11):
- Zylo, 2026 SaaS Management Index: https://zylo.com/news/2026-saas-management-index
- Zylo, How Much Is Wasted on SaaS Spend: https://zylo.com/blog/how-much-wasted-on-saas-spend
- BetterCloud, 2025 State of SaaS Trends: https://www.bettercloud.com/monitor/2025-state-of-saas-trends/
- BetterCloud, 2026 State of SaaS Report: https://www.bettercloud.com/monitor/the-2026-state-of-saas-report/
Originally published on nodesparks.com.
Top comments (0)