An indie hacker's journey building BillSnip: from frustrated consumer to shipping a DeepSeek-powered bill negotiation agent that actually works.
It started with a $47 internet bill.
I was sitting at my desk, staring at my Comcast invoice, wondering why my "promotional" rate had magically doubled after 12 months. I'd been a loyal customer for three years. I'd never been late on a payment. And yet, here I was — paying $20 more per month than my neighbor for the exact same plan.
So I did what any reasonable person would do: I called Comcast. I spent 45 minutes on hold, navigated through four different departments, and eventually convinced a retention agent to drop my bill by $15/month. Fifteen dollars. For 45 minutes of my life and a mild case of repetitive strain injury from pressing "1" on my phone keypad.
That's when it hit me: this problem is solvable with software.
Not just solvable — it's perfect for an AI agent. The negotiation scripts are predictable. The retention playbooks are well-known. The leverage points (competitor pricing, loyalty discounts, bundling options) are finite and discoverable. A human doing this manually is slow, emotional, and inconsistent. An AI agent could be fast, patient, and data-driven.
That weekend, I started building what would become BillSnip.
The Concept: An AI Agent That Fights Your Bills So You Don't Have To
BillSnip is pretty simple in theory: you upload your bills, and an AI agent analyzes them, identifies where you're overpaying, and then automatically negotiates with your service providers on your behalf.
The magic isn't just in the analysis — anyone can tell you that your internet bill went up. The magic is in the action. BillSnip's AI agent drafts negotiation emails, sends them to your providers' retention departments, follows up, counter-offers, and keeps pushing until it either secures a lower rate or exhausts all reasonable options.
And here's the part I'm most proud of: we pay you 60% of whatever we save you. If BillSnip saves you $200 a year, we keep $80 and you get $120 back. No savings? You pay nothing. The incentive alignment is clean — we only make money when you save money.
The Tech Stack: What's Under the Hood
Let me get into the technical weeds, because I know the DEV community cares about this stuff.
Frontend: Next.js with a Focus on Simplicity
The frontend is built on Next.js with the App Router. I chose Next.js because I wanted SSR for SEO (this is a consumer product, discoverability matters) and the API routes meant I could keep the backend API simple during the early days.
The UI is deliberately minimal. Upload a bill (PDF or screenshot), see the analysis dashboard, track active negotiations. That's it. I'm not trying to win design awards — I'm trying to get people from "uploaded bill" to "money saved" in as few clicks as possible.
Backend: FastAPI + DeepSeek
The backend is a FastAPI service running on a modest VPS. FastAPI gives me async capabilities out of the box, which is critical because the negotiation workflow is inherently asynchronous — we send an email, wait 2-3 days for a response, analyze the reply, draft a counter-offer, repeat.
The brain of the operation is DeepSeek. I evaluated several LLM options, and DeepSeek won on two axes that mattered most for this use case:
Cost efficiency: BillSnip processes hundreds of bills and generates dozens of negotiation emails daily. At DeepSeek's pricing, the LLM cost per negotiation is roughly $0.03. That's critical when your business model depends on keeping per-user costs low.
Instruction following: Negotiation emails need to be precise. You can't afford the model hallucinating a competitor price or promising a cancellation that didn't happen. DeepSeek's adherence to structured prompts and its ability to maintain context across multi-turn email threads has been remarkably reliable.
Each negotiation session is essentially a multi-turn conversation where the AI agent:
- Reads the provider's initial response
- Identifies their position (firm on price? willing to offer credits? likely to match competitor?)
- Crafts a response that applies the right pressure point
- Knows when to push harder and when to accept the best available offer
Data Layer: PostgreSQL
PostgreSQL handles everything persistent — user accounts, bill records, negotiation histories, provider contact databases, and the credit/transaction ledger. I use JSONB columns extensively for storing the structured analysis results (bill line items, identified savings opportunities, provider response classifications).
The credit system runs on a simple ledger table. Every action (upload, analysis, negotiation round) costs credits. The transaction log is append-only, which makes auditing and dispute resolution straightforward.
The Business Model: Pre-Paid Credits, Transparent Pricing
I deliberately avoided the subscription trap. Most SaaS products in the bill management space charge $10-15/month whether they save you money or not. That creates a misalignment — you're paying for the promise of savings, not actual savings.
BillSnip works differently. You buy credits, and credits get consumed as the AI agent works:
- $10 → 100 credits
- $25 → 300 credits (best value for most users)
- $50 → 650 credits (power users and people with many bills)
Each bill analysis costs a few credits. Each negotiation round costs a few more. If we don't save you anything, you've only spent a handful of credits on the analysis — not a monthly subscription you're resentfully paying.
Membership Tiers: The More You Use, The Cheaper It Gets
On top of the credit system, there's a membership tier structure that rewards active users:
- FREE: Sign up, get 100 credits + 3 free bill analyses. No credit card required.
- SILVER: After your first credit purchase.
- GOLD: As you use more, you unlock better per-credit rates.
- PLATINUM: Our heaviest users get the absolute best pricing.
It's a simple progression, but the psychology works. People who start with the free tier and actually see results naturally want to negotiate more bills. The tier system makes that progression feel rewarding rather than punitive.
What the First Users Are Saying
I launched the beta two months ago with a simple landing page and a ProductHunt post. No ads. No influencer partnerships. Just the product and a handful of early believers.
The results have been encouraging:
- Average savings per successfully negotiated bill: $18-24/month
- Success rate on negotiations: ~65% (higher than I honestly expected)
- Most common wins: Internet/cable bills, phone plans, insurance premiums
- Average time to first savings: 5-8 days (the agent sends the initial email within hours, but provider response times vary)
One user, a freelancer named Marcus, uploaded six bills on day one. BillSnip negotiated his internet from $89 to $55, his phone plan from $85 to $60, and his car insurance from $145 to $118. That's $86/month in savings — $1,032/year. After BillSnip's 40% cut, Marcus still pockets $619.20 annually. For uploading six PDFs.
Why This Works (The Uncomfortable Truth About Bill Negotiation)
Here's something that might surprise you: most service providers expect to be negotiated with.
Cable companies have retention departments staffed by people whose entire job is to keep you from leaving. They have scripts, discount tiers, and promotional offers they're authorized to apply. They know that roughly 30-40% of customers who call to complain will accept the first offer. Another 20% will push back and get a better deal. The remaining 40% will either accept their current rate or cancel.
The problem isn't that companies won't give you a discount. The problem is that most people don't negotiate. They accept the rate on the bill because it feels like too much hassle to fight.
BillSnip eliminates the hassle. The AI agent doesn't get frustrated. It doesn't feel awkward asking for a discount. It doesn't give up after one try. It just... keeps going, methodically, until it either wins or reaches the limit of what's reasonable.
Lessons Learned Building This Solo
Building BillSnip as a solo developer has been one of the most educational experiences of my career. A few things I've learned the hard way:
Email deliverability is a hidden boss fight. When your AI agent sends negotiation emails to Comcast, AT&T, and State Farm, those emails need to land in the right inbox — not spam. Setting up proper SPF, DKIM, and DMARC records, warming up sending domains, and crafting emails that look professional (not spammy) took me longer than building the actual negotiation logic.
Provider responses are wildly inconsistent. Some companies reply in hours with structured responses. Others take a week and send back a wall of legal text. I had to build a robust parsing layer that can extract the key information — "we can offer you $X off for Y months" — from emails that range from three sentences to three pages.
Users trust matters more than I thought. People are uploading their actual utility bills, phone statements, and insurance documents to my website. That's sensitive financial data. I spent an entire week just hardening the security — encryption at rest, minimal data retention, clear privacy policies. If you're building anything that touches financial data, don't skimp on this.
The "last mile" of automation is always manual. About 10% of negotiations hit a point where the provider says something like "we need to speak with the account holder directly." For those cases, I've built a human-in-the-loop fallback where the user gets notified and can either jump into the conversation or let the agent try a different approach.
The Roadmap: What's Next
We're just getting started. Here's what's on the horizon:
Phone negotiation: The email workflow is proven. We're building voice-capable negotiation where the agent can actually call providers and negotiate in real-time using synthesized speech. Early tests are promising.
Proactive monitoring: Instead of waiting for you to upload bills, BillSnip will monitor your accounts and flag rate increases automatically — then negotiate before you even notice.
Business accounts: Small businesses have even more bills than consumers, and the per-bill savings are often larger. A dedicated business tier is in the works.
API access: For the developers and fintech builders in the audience — we're exploring an API that lets other applications integrate BillSnip's negotiation engine.
Try It Free
If you've read this far, you're probably thinking about that one bill that makes your eye twitch every month. Good. That's exactly who we built this for.
Sign up at billsnip.publiflow.vip — you'll get 100 free credits and 3 complimentary bill analyses. No credit card, no commitment. Upload a bill, see what the AI finds, and decide from there.
The worst case? You spend 10 minutes uploading a PDF and learn something interesting about your bill. The best case? You save hundreds of dollars a year for almost zero effort.
I built BillSnip because I was tired of overpaying and I had the technical skills to do something about it. If you're in the same boat, I'd love for you to try it.
Tech stack recap: Next.js · FastAPI · DeepSeek · PostgreSQL · Vercel · Railway
Built with too much coffee and a genuine hatred for Comcast's retention hold music.
Questions, feedback, or just want to talk about bill negotiation strategies? Drop a comment — I read every one.
Top comments (0)