New·Argos now detects usage deviation across 100+ model endpoints See how →
Home / Case studies / 43 pilots into one governed portfolio
Case study

43 pilots into one governed portfolio

A US national health plan had more AI activity than any company should. None of it produced results anyone could measure.

$7.5M+annual savings unlocked
43 → 14pilots consolidated
37%measured contribution to FY24 objectives
11 → 1cost centres into one AI budget

US national health plan · 14 months

The situation

Forty-three teams, forty-three ideas about what AI should do.

Different teams had spun up pilots with different tools and vendors. Some were machine-learning projects from 2021 nobody had looked at in a year. Others were generative experiments two people started without telling IT. Forty-three separate projects, each consuming budget and people, most without a shared metric. The core problem was not ambition — nobody owned the whole picture.

What we found

The real cost of fragmentation is what you never measure.

No central view of AI spend; budget spread across eleven cost centres. Most pilots had no measurable business KPI. Multiple teams working the same problems without knowing it. No process for deciding what to build next or what to stop.

How we did it

We started by understanding what existed, not by building.

Month 1: every initiative mapped — what problem, how success is defined, who owns it. Most couldn't answer all three. Month 2: stop 19, merge 14 into 6, accelerate 10. Months 3–6: a central portfolio function, a shared KPI framework and a prioritisation committee with real authority. Months 7–14: highest-priority use cases into production, each with a baseline, a target and an owner.

The outcome

Numbers that moved because we made sure they could be measured.

$7.5M+ in annual savings across the rationalised portfolio, a 37% measured contribution to FY24 business objectives, and 100% of active use cases with a KPI, baseline and named owner.

What we learned

Four things worth carrying forward.

Stop before you build

The fastest path to AI outcomes is rarely a new project. It is stopping the ones that aren't working.

Measurement is not a reporting task

The baseline has to exist before the work starts. That is why we could show $7.5M with confidence.

Governance is not bureaucracy

It means clarity: who decides, how, and what happens when a use case stops delivering.

The portfolio view is the unlock

No single use case produced $7.5M. Running the right ten things well instead of forty-three badly did.

Let's talk

Tell us the number you need to move.

A 45-minute working session with an operator who has run the kind of work you are describing. You will get an honest read on where your programme stands and what it would take to move it.