The first version of an AI workflow is usually cheap. Someone builds it in a few days, it mostly works, and everyone's impressed. Then it needs fixing. Then a new edge case appears. Then a second tool gets added to cover what the first one can't do.
A year later, the company is spending real money every month on something nobody decided to fund at that level. The cheapest AI decision in month one can become the most expensive one by month twelve.
The four costs nobody adds up
- Developer labor. Not the build, the upkeep. Hours spent patching prompts, debugging, fixing integrations and handling edge cases. In a fragile setup, this is often the biggest cost and the least visible.
- Token and API spend. Sending far more text to the model than the task needs, on every call, adds up fast at volume.
- Software overlap. Three AI tools with overlapping features, each with its own subscription, its own data and its own owner.
- Waiting. Every month spent maintaining a workaround is a month the process you actually wanted to change stays the same.
Compare it to the process, not the price
Build versus buy usually gets framed as a software price comparison. That's the wrong frame. The real question is what the setup costs compared to the business process it's supposed to change.
Picture a team that built an AI assistant for customer support. It costs a modest amount in tokens, but one developer spends a third of every week keeping it running, and support headcount hasn't changed. When they add it up, the "cheap" tool costs more than a purpose-built product would, and it hasn't moved the outcome it was built for.
That isn't a failure of AI. It's a failure to model the economics before the setup became permanent.
Watch for the warning signs
You may be funding a leak if fixes are a regular line on the engineering agenda, nobody can say what the AI setup costs in total each month, you're paying for more than one tool that does roughly the same job, or the process it was meant to change still has the same headcount and cycle time.
What to do this week
- Add up monthly developer hours spent maintaining AI workflows, not building new ones.
- Pull last month's API and AI software bills into one number.
- List every AI tool you pay for and mark the overlaps.
- Name the process this setup is meant to transform, and its current monthly cost.
- Project 12 months forward and decide: keep, rebuild, replace or stop.
Build vs. Buy & Token Leak Calculator
The Build vs. Buy & Token Leak Calculator models the next 12 months of developer labor, API and token spend, AI software subscriptions and overlap against the business process the setup is meant to transform, so you can see the real cost and return before you commit another dollar.
If the current setup survives 12 more months, know what you'll have spent and what you'll actually own.
Open Build vs. Buy & Token Leak Calculator →Get the next lesson in your inbox
Short, practical lessons like this one for founders running $1M–$20M companies, plus every new free tool before anyone else.
If this is the call in front of you, let's talk it through.
A focused 30-minute conversation. No pitch. Just a direct discussion about the decision and whether I can help.
Book a call with Danny →
Danny Feldman is a 3× founder and the advisor behind Feldman Advisory. He founded and scaled Stellar to $100M in annual revenue across 15 states, raised $55M+ as a founder, and used AI to double revenue before exiting to private equity. Earlier, he recruited and trained 55,000+ sales reps at Cutco and grew a division from 1 to 58 locations. Operator to operator, he now helps founders of $1M–$20M companies get out of the middle of the business, build the leadership team, and raise capital. Book a 30-minute call →