How our AI cost diagnosis helped a HealthTech company add $28M to its valuation in a few months.

The costs were spread across teams, so no one could see the full amount.

The HealthTech company's leaders were preparing for a funding round, but the cost of running the AI their product depended on was spread across teams and budgets. A model-provider invoice showed one expense, cloud infrastructure appeared in another budget, and the work needed to prepare inputs, review outputs, and maintain supporting tools sat elsewhere.

Each expense could look reasonable on its own while the combined cost reduced what the company kept from its revenue. Leadership needed to understand that total before deciding what to change, and this company was one of five growth-stage HealthTech businesses whose AI costs we reviewed.

We traced the cost across the whole workflow.

We spent ~3 days investigating inside each company, following the work across the teams responsible for it through read-only diagnostics approved by each CTO and security review.

We examined infrastructure activity and bills, then worked with finance, technology, and engineering teams to understand what each expense supported, including the work around AI such as preparing inputs and maintaining the tools needed to deliver an answer.

That brought costs usually discussed separately into the same view, allowing us to assess where a change could reduce expense while keeping the capabilities the business still needed.

We gave leadership a recommendation it could act on.

We put the findings, their effect on costs and margins, and a recommended next step into a one-page executive report so leadership could see what needed attention and decide how to respond. The full analysis sat behind it for finance and technical leaders who wanted to examine the evidence.

One company used that report to act with its own team: we supplied the diagnosis and recommendation, and they carried out the changes. We'd also agreed to return and measure operating costs again so the company could judge what those changes achieved, rather than stopping at a projected saving.

Costs fell before the company closed its round at a higher valuation.

Three months after the company's changes, we repeated the review of system activity and the survey, finding that AI operating costs had fallen while the CTO reported no drop in output quality. The physician-review program remained in place.

Leadership later reported that margins and earnings were the biggest changes in the business, and alongside continued growth, those improvements gave the company a stronger operating picture to take to investors. Within a few months, it moved from a $32M valuation offer to a funding round closed at ~$60M.

That $28M increase belongs to this company, with our contribution being the diagnosis that made the cost problem visible, the recommended response, and the follow-up measurement after its own team acted.

We corrected two mistakes in our own report.

A later review found two problems in our original analysis: we'd used an unsupported hourly rate in part of the calculation, and we'd treated clinical review too broadly as a cost that could be removed without first separating its purpose from how the work was organized.

We told all five companies and sent a corrected page. The company with the valuation result hadn't cut physician review as part of its changes.

That changed how we assess costs, because a reduction is only useful if it accounts for the work the business must keep doing. Every rate now needs a traceable basis, and every proposed reduction needs an explanation of what the work protects or makes possible.

A useful diagnosis didn't require us to do the build.

We offered these diagnostics free in the hope they would lead to implementation work, but none of the five became a paid engagement. The company with the documented valuation result could act with its own team because our report gave its leaders a problem they could see and a recommendation they could carry out.

That was a lesson for our commercial approach: the diagnosis had value even when the client didn't need us to build another system.

How we traced costs and measured the change.

The technical companion explains the diagnostic method, cost definitions, follow-up measurement, and corrections to the original report.

Their own team could act once leadership could see the problem.

The company's AI costs were spread across teams and budgets, which made it hard for leadership to see their combined effect on margins. We brought those costs together and recommended a response its own team could carry out, then returned to measure the lower operating costs after the changes.

If you’re trying to grow your own business, start with your goal, look at what it'll take to achieve it, and then identify what's keeping you from taking action. Just remember that every business is different and that the lack of a clear view of its total AI costs was just their bottleneck and it might not be yours.