We Turned Down a $200k Project Last Month — Here's Why

A developer at Autor turned down a $200k project from a Series B fintech company to build their customer support automation platform, citing scope creep, maintenance burden, and the risk of diluting their healthcare focus. The decision was driven by the realization that their competitive moat lies in operational knowledge of healthcare voice AI, not generalist AI development.

Last month a Series B fintech company offered us $200k to build their entire customer support automation platform. We said no. It was the hardest business decision I've made this year, and I'm still not 100% sure it was right. Their VP of Engineering found us through a referral. They'd seen what we built with Loquent — our production voice AI platform handling thousands of automated calls per month for healthcare and dental clients — and wanted something similar for their support operation. The scope was clear: voice AI for inbound support, chatbot for their web app, automated ticket routing, and integration with their existing Zendesk and Salesforce stack. On paper, this was a dream project. $200k budget. Reasonable 16-week timeline. A technical team on their side that actually understood AI limitations. They weren't asking us to "add AI" to a product that didn't need it — they had a real problem with support costs eating 40% of their gross margin, and they'd done the math on what automation could save them. I spent three days putting together a technical proposal. We mapped out the architecture: Twilio for voice, Anthropic Claude for conversation handling, Pinecone for their knowledge base retrieval, NestJS for the backend services. I knew exactly how we'd build it because we'd already solved most of these problems with Loquent. Then I killed the deal. It wasn't one thing. It was a pattern I'd been watching for six months across our client engagements. The first red flag was scope. This wasn't a voice AI project with some chat bolted on. This was three distinct products: a voice agent, a chatbot, and a ticket routing system. Each one of those is a 6-8 week build if you're doing it right. At $150/hr with our team of four senior engineers, $200k buys you roughly 1,333 hours. Split across three products over 16 weeks, that's about 83 hours per person per product. That's tight. Not impossible, but tight enough that quality would suffer. The second flag was maintenance. We've been running Loquent in production for over a year now. I've written about the 3am calls, the edge cases that break your confidence in AI, the 18% of calls that still need human transfer. What I haven't written about much is the ongoing engineering cost of keeping a production voice AI system reliable. Every month we spend 60-80 hours on Loquent maintenance: prompt tuning, handling new edge cases, updating integrations when Twilio or our LLM providers ship breaking changes, monitoring call quality metrics. If we built three products for this client, we'd be signing up for 150-200 hours of monthly maintenance work. That's essentially one full-time senior engineer dedicated to a single client. At our size — four senior engineers, no offshore, no handoffs — that's 25% of our entire capacity locked up indefinitely. The third flag was the one that actually made the decision. This project would pull us away from healthcare. When I started Autor in 2021, we were a generalist AI development studio. We'd build anything: recommendation engines, NLP pipelines, computer vision systems, chatbots. We were good at all of it. We shipped 50+ products across 10+ countries. But "good at everything" is a terrible positioning for a six-person studio competing against agencies with 200 engineers and enterprise consultancies with thousands. The shift happened when we built Loquent. For the first time, we weren't just building AI — we were operating it. We understood the difference between a demo that handles 10 calls and a system that handles 10,000. We learned things about healthcare voice AI that nobody else had learned because nobody else was running it at our scale in Canada with PHIPA compliance baked in from day one. That knowledge is our moat. Not our code. Not our tech stack. The operational knowledge of what breaks at 2am when a patient calls about a prescription refill and your AI confidently gives the wrong pharmacy number. Taking the fintech project would have meant three things for our positioning: First, it would dilute our healthcare expertise. Four months of building fintech support tools means four months not deepening our understanding of healthcare voice AI. In a market moving this fast, four months of distraction is significant. Second, it would muddy our story. Right now when someone asks "who builds voice AI for healthcare in Canada?" there's a short list, and we're on it. If we start building fintech support platforms, we become "the AI agency that does everything" again. We already learned that lesson once. Third — and this is the part that surprised me — it would actually hurt our revenue long-term. We ran the numbers. The $200k project would generate roughly $200k over 16 weeks, then maybe $30-40k/year in maintenance revenue. But our healthcare pipeline had two deals in it totaling $280k, both specifically seeking a team with production healthcare voice AI experience. If we took the fintech work, we'd likely lose both of those deals because we wouldn't have the capacity to start them on time. So the math was: $200k now + $30-40k/year recurring, versus $280k in 8 weeks + the compounding value of deeper healthcare positioning. Telling the fintech VP was uncomfortable. They'd spent time with us, shared their architecture docs, introduced us to their CTO. I didn't want to waste their time or damage the relationship. I was honest about it. I told them exactly what I'm telling you: we're a healthcare voice AI company now, and taking their project would make us worse at the thing we're best at. I offered to introduce them to two other studios I trust — one in Toronto, one in Vancouver — who could handle their scope without the positioning conflict. Their response surprised me. The VP said something like: "That's actually why I wanted to hire you. Most agencies would have said yes to $200k without thinking twice." He asked if we'd consider a smaller advisory engagement — 20 hours to help their internal team with the voice AI architecture, since that was where our expertise was most relevant. We said yes to that. That advisory engagement turned into a $15k engagement, which isn't $200k, but it's $15k of pure expertise delivery with zero maintenance tail. And the referral relationship is intact — they've already sent one healthcare company our way. I'd be lying if I said this was an easy call. $200k is a lot of money for a studio our size. There are months where payroll feels tight, where a big contract would solve problems I think about at 3am. And there's a real risk in over-indexing on positioning. If healthcare voice AI hits a regulatory wall, or if a bigger player commoditizes what we do, we'll wish we had diversified. I think about that. I don't have a clean answer for it. What I do know is that in the 18 months since we started focusing on healthcare, our close rate went from 15% to 40%, our average project value went up 60%, and we stopped competing on price. When a dental clinic in Ontario needs a voice AI system that handles patient calls 24/7 and complies with PHIPA, they don't get five proposals. They get maybe two. We want to be one of those two. Scope math doesn't lie. Before you say yes to any project, divide the budget by your hourly rate, then by the number of distinct products. If each product gets less than 400 engineering hours, you're going to cut corners. You might not notice on delivery day, but you'll notice in month three of maintenance. Maintenance is the real cost. Every product you ship is a recurring obligation. At a small studio, three or four maintenance-heavy products will consume your entire capacity for new work. Calculate the maintenance tail before you sign. "Good at everything" is positioning for nobody. This is especially true for studios under 20 people. You can't outbuild a 200-person agency on breadth. You can absolutely outbuild them on depth in one domain. Be honest when you say no. Don't make up excuses. Tell the client exactly why you're declining. The best referral sources are people who respect your honesty, even when it costs them time. Run the 12-month math, not the 12-week math. The $200k looked better than our pipeline on a 16-week horizon. On a 12-month horizon, staying focused was worth roughly 40% more in revenue — and that's before you factor in the positioning value. If you're building something similar, we'd love to hear about it. Reach out at hello@autor.ca or visit autor.ca.