For years, I’ve worked at the intersection of health system and payer economics, negotiating contracts from both sides of the table, designing value-based arrangements, advising on ACO strategy, and working on health plan products spanning Medicare Advantage quality programs and tiered networks.

Two weeks ago I joined Translucent, which builds AI for health system finance and managed care teams. The experience has changed how I think about a problem I’ve seen for years.

Many financial decisions in a health system start as a hunch.

“I think this payer is underpaying us.” 

“I think the shift into that HMO is hurting our yield.” 

“I think a billing issue is driving these denials.” 

Every one of these is a reasonable instinct from someone who knows their business. But health systems are complex organizations, and the people asked to confirm or kill those hunches often sit under one umbrella: “finance”. The work itself demands expertise in revenue cycle, contract negotiation, physician productivity, and payer policy. These specialists don’t always sit on a health system’s executive leadership team, so the CFO is left to make the call. They decide which hunches are worth chasing from data that lives in three systems and a dozen spreadsheets. When every answer takes that much effort, teams prioritize the opportunities someone already has visibility into, with no way to know whether those are the only opportunities or even the most valuable ones.

The biggest opportunities may be the ones no one has a hunch about.

Everyone is pitching AI that answers your questions. That is useful, but it has a ceiling: you can only ask about what is already on your radar. The underpayment from a recent payer policy change nobody expected never makes the list, because no one had a reason to go looking for it.

What Translucent does differently is send agents that understand the domain to go dig, including in places no one had a hunch about. In contract negotiation, that means finding that a payer has started applying a site-of-service differential or a multiple-procedure reduction that isn’t in your contract, or that a rate increase you fought for is being clawed back by “lesser of” language because your chargemaster hasn’t kept pace with your negotiated rates. The output is not a dashboard. It is a finding, with the evidence, that a managed care director can take into a meeting.

Scenario modeling finally moves at the speed of the negotiation.

The industry is changing fast enough that a single-point forecast is close to useless. What happens if oncology volume grows 8% a year for three years? What happens to the contracts we have pegged to a percentage of Medicare if our DSH or IME adjustments change, and does the definitions section even pick those up? What if we take Aetna’s rates and lay them over our Blue Cross volume? Most systems have a contract modeling tool that can reprice a claims file, but building and maintaining the term models and running each what-if still falls to a small team. The answer managed care leaders need often arrives after the question has stopped mattering.

When that modeling becomes fast, it changes how a CFO or managed care leader negotiates. You can build a real playbook: an opening ask, a target, and a walkaway number. In many instances, a health system CFO's walkaway number for any given payer is built on experienced intuition rather than defensible analysis their team can take into a negotiation. When systems don't have reliable cost data by service line, they often lean on cost-to-charge ratios as a proxy, which makes it hard to know whether a proposed rate actually covers the cost of care.Getting it right means reconciling what the contract says against what claims actually paid and modeling the counterproposal on top, and by hand that can take weeks. That delay is why health systems have signed contracts without fully modeling how the terms would interact with real volumes. It is rarely a knowledge problem. More often it is a modeling-capacity problem, and that part is now solvable.

It gets smarter the longer it works with you.

This is the part that worried me most about AI in this space, and the part Translucent has thought hardest about. A lot of answers in healthcare finance are right on paper and wrong in practice. A generic model will tell you Medicaid or traditional Medicare is your worst payer and suggest you renegotiate. A naive benchmark will let a children’s hospital’s base rate inflate a system-wide peer comparison. These are not data errors; they are domain errors, and an agent that surfaces opportunities is only worth having if it knows the difference.

Some of that judgment is industry knowledge, which is where my job comes in. But a lot of it is specific to one health system, and it lives in a few people’s heads: why that carve-out exists, which payer always disputes the same DRG, which service line the board is watching this year. When they leave, the knowledge leaves with them. Translucent works the same way those people do. Every time a user corrects a finding, explains a carve-out, or notes why a rate looks the way it does, that context stays and gets used across revenue cycle, contracting, and finance. That is how a well-run team works, and it is how most finance organizations, split across departments and turnover, do not.

Two weeks in, the honest summary is this: I watched our agents do in an afternoon what would have taken me a week to get back to a client as a consultant. They surfaced findings no one asked me to look for, considering more variables than I could ever build into a revenue model by hand.If you run managed care, revenue cycle, or finance at a health system and any of this sounds familiar, I’d like to show you what it looks like on your own contracts and claims.

‍