For deal teams

Know what the AI in this deal will actually cost you. Before IC.

A written diligence read on the target's AI: what the technology actually does, what the claims will survive, what integration will really cost, and what the next buyer will ask that this one didn't.

Book a call Ways to work with me

Five business days from data room access. Fixed scope, fixed date, agreed in writing before any work starts. Built to land before the investment committee, not after. Your QoE provider costs more and will not answer any of this.

The gates I have been through

Delivered inside their regulated programs
Johnson & JohnsonGileadModernaAbbVieAmerican Red Cross

Their clinical trials, run from inside the technology vendor that delivered them.

Cleared their review, and bought
NHSU.S. Department of Veterans Affairs

An FDA-regulated wearable, through two of the most demanding enterprise buyers there are. Security, privacy, clinical, and procurement all had a veto. None of them used it.

Four functions hold a veto over AI in regulated healthcare: regulatory, quality, security, and privacy. I have held accountability in all four. More than 30 FDA-regulated products through those gates as an operator, not an advisor: AliveCor, LabCorp, Vineti, Korio, Clip Health, endpoint Clinical.

The gap in the process

Commercial and QoE diligence will not find this.

Your commercial diligence tests the market. Your quality of earnings tests the numbers. Neither is built to test whether the AI in the deck is a product, a feature, or a demo, or what it will cost to own once you have it.

That gap is where AI-era deals go wrong. The model works. The pilot data is real. And the operating cost, the claims exposure, and the integration bill all sit outside what anyone in the process was asked to look at.

What kills the AI in a deal is rarely the model. It is the gate nobody priced.

The questions I ask that the CIM does not answer

Six of them. Which ones matter depends entirely on the deal.

“What does the change-of-control clause in the top ten customer contracts actually allow?”
Deal models routinely assume a low single-digit invocation rate, and I have watched that assumption go untested straight through to close. In healthcare services, where the contract holder is often a health system with its own procurement politics, invocation is a bargaining event rather than an exit. The customers who invoke mostly do not leave. They renegotiate. If nobody has read those ten clauses and stress-tested the rate, the deal model contains a number somebody guessed.
“What share of COGS is foundation model API fees?”
A lot of AI health companies are built on pricing that does not yet reflect the cost of serving them. A 30% increase takes a business from 70 to 80% gross margin down to 50 to 60%. Same EBITDA multiple, materially less equity value.
“What is the liability cap in the AI vendor contracts?”
The standard is twelve months of fees paid. On a $500K clinical AI tool, that caps vendor recovery at $500K, against patient harm litigation that routinely clears $10M. Sign that without negotiating it and the portfolio company is self-insuring.
“Has the target ever been audited at the standard of the planned acquirer?”
A company that has never been audited to the standard of its acquirer is not one remediation project away from integration. It is one remediation project away from being ready to start integrating. That gap is a precondition, not a line item, and I have never once seen it in a model.
“Do three years of compliance committee minutes contain the word ‘identified’ next to overpayment language?”
Under the sixty-day rule, “identified” starts the clock. Not “quantified,” not “confirmed.” The memo written to flag a problem becomes the exhibit that proves the clock started and nobody stopped it.
“How many EMR instances does the platform run?”
Single-instance is now table stakes for the next buyer rather than a premium. Multiple instances get priced as a consolidation project and handed back to you as a lower multiple.
Primary engagement
Pre-IC Diligence Sprint

A written read on the AI in the target, separating what is product, what is feature, what is roadmap, and what is demo. The claims exposure. The integration economics, including the preconditions nobody priced. The contract and regulatory risks that hit the model.

Written for the IC memo. Short enough to be read, specific enough to change the number.
Book a call

What you get in five business days

  • What the AI is, and what it is not
  • Claims exposure and evidence gaps
  • Integration economics and preconditions
  • Contract risks that hit the model
  • Regulatory exposure that becomes a value problem
  • What the next buyer will ask that you did not
Arvita Tripati

I have been on the other side of the data room.

For nearly two decades I built the things a buyer inspects. Quality systems. Design histories. Clinical protocols. Regulatory submissions. More than 30 regulated products through FDA and CE pathways.

I have defended FDA inspections and come out without a finding. I have sat in the room while a company assembled its story for a buyer, and watched which parts of it were true.

Which is why I know where to look.

None of that is in the CIM. All of it is knowable in five days if someone knows what to ask for.

Before any work starts, you see the plan.

A 30-minute callThe deal, the timeline, what the AI is claimed to do, what worries you. You ask anything.
A written scopeWhat I review, what you receive, the delivery date, the fee. You read it on your own time.
You decideIf it fits, we start on data room access.
Book a 30-minute call

Why not the alternatives

  • Your existing providers. Commercial tests the market, QoE tests the numbers, legal tests the contracts. None is built to test whether the AI is real, or what it costs to operate once you own it.
  • A technical DD firm. They review the code, the architecture, the security posture. Useful. It will not tell you whether the claims survive a health system's legal review, or which customer contracts are about to be renegotiated.
  • Skipping it. The AI questions do not go away. They arrive at exit, asked by a buyer with more time than you had, and by then the answer costs a multiple rather than a fee.

I am not here to confirm the thesis

  • I will not diligence a company I have advised
  • I will not give a deal team the paper it needs to reach a yes it has already decided on
  • If what I find changes the price, good. If it kills the deal, better now than at exit.

They hired me to find what nobody else would.

“Arvita crushes challenging problems and creates order where there is uncertainty. I hired her, and she became an integral member of the leadership team. I hired her again to join me on the founding team at Korio.”

Chuck Harris
Former CEO, endpoint Clinical (acquired by LabCorp)
COO & Co-founder, Korio

“I love Arvita's discipline and the structure of her thinking. Her work was thorough, well-organized, and genuinely useful.”

Drew Bennett
Innovation Partnerships
University of Michigan

Ways to work with me

Fixed scope, fixed date, both agreed in writing before any work begins. No hourly billing, no staffing pyramid, no change orders.

Pre-IC Diligence Sprint
5 business daysfrom data room access

For deal teams with an AI-enabled target and a committee date.

  • A written read on what the AI is, and what it is not
  • Claims exposure and evidence gaps
  • Integration economics, including the preconditions
  • Contract and regulatory risks that hit the model
  • What the next buyer will ask that you did not
The decision this enables: what are we actually underwriting, and at what price?
Book a call
Post-Close AI Operating Blueprint
2 weeks to diagnosis30 days to blueprint

For a company you now own, where AI was in the thesis and value has not shown up. Ranked initiatives, implementation economics, adoption constraints, decision rights, board measures, and a 90-day execution plan.

The decision this enables: what does management scale, fix, defer, or stop?
Book a call
Fractional Chief AI Officer
6 to 12 monthsI make the decisions, not just recommend them

The title is the shorthand. The substance is decision rights. Every other engagement on this page ends in a document that somebody else still has to sign. This one does not. You give me authority over a defined set of calls, and I make them.

Which calls are mine is the scope, and we agree it before I start. The list below is where those decisions usually sit. We cut it, add to it, and set the escalation path. Then it is signed, and the accountability is real.
  • The review threshold. What a human still has to touch, and when that falls again
  • What gets funded, what gets deferred, and what stops
  • The evidence, claims, and controls a regulated product has to clear
  • The escalation path when regulatory, quality, security, or clinical says no
  • Board and IC reporting on what the AI is actually doing to the business
The decision this enables: the one nobody will make, because their name goes on it.
Book a call
AI Value-Creation Advisory
Monthly retainertypically 6 to 12 months

Senior judgment while the company executes. Monthly executive sessions, business-case pressure testing, vendor decisions, roadmap reprioritization, adoption reviews, board and IC preparation, and early warning on product, regulatory, or operating risk.

Book a call

Frequently asked

Five business days from data room access. If your IC is sooner than that, say so on the call. I will tell you whether it is worth starting.

Data room access, the CIM, and 30 to 60 minutes each with two or three people at the target where that is possible. If management access is off the table, tell me before we start. It changes what I can tell you, and you should know that going in rather than reading it as a caveat in the memo.

Then you walk into IC able to say so, with the evidence, instead of hoping. Confidence is the product. It is worth as much when the answer is yes.

Not at the practice level, and it is worth being direct about why. I will not diligence a company I have advised, and I will not advise a company I have diligenced, and I run that check before any engagement starts. Beyond that, the readiness work is why the diligence work is good. I know what a company does to make itself look ready for a buyer, because I am often the person preparing it. That is not a conflict. It is the whole reason I know where to look.

The AI questions do not go away. They just get asked later, by someone with more leverage.

Better to ask them now, while the price is still a fee.

Book a call

Already own the company? Read the portfolio page  ·  Running one? Read the CEO page