Elena Petrova Executive mandate AI Transformation Seat · USD 8,000 / month
90-day AI Transformation mandate: Hollerith Insurance Services
Elena Petrova leads adoption. The mandate has a stop condition, and the first deliverable is a list of what not to do.
Hollerith has nine AI pilots running in seven departments, none in production, and a board that has been told twice that transformation is underway. This mandate stops most of them. It also says, in clause 5, the circumstances in which it should stop itself.
From 12 April, Elena owns the AI programme, which pilots continue, the deployment gate and spend to USD 50,000. On the evidence, six of the nine current pilots should stop: three duplicate each other, two have no owner, and one is being used on claims data in a way that has not been through the company's own approval process.
Document control
- Document number
- MAN-HOL-01
- Version
- 1.0
- Effective
- 12 April 2027
- Owner
- Elena Petrova, Fractional Head of AI Transformation
- Approved by
- Hollerith executive committee, minute 2027-19
- Next review
- 12 July 2027
- Classification
- Confidential. Executive committee
Why
What the diagnostic found
| Observed | Evidence |
|---|---|
| Nine pilots, seven departments, nothing in production | Confirmed with each department lead |
| Three pilots do substantially the same thing | Document summarisation, bought separately three times |
| Two have no named owner | The person who started them has left in both cases |
| No deployment gate exists | Nothing defines what a pilot must show to go live |
| One pilot processes claims data | Outside the company's own data approval process |
| Two pilots are genuinely good | Both in underwriting support, both with a real owner and a real measure |
The claims data pilot is not a hypothetical concern. It was set up in good faith by a capable team who did not know an approval process existed, which is itself the finding.
Authority
Decision rights
| Elena decides | Executive decides | Board or regulator matter | |
|---|---|---|---|
| Which pilots continue, pause or stop | Yes | ||
| The deployment gate and its criteria | Yes | ||
| Tooling, vendors and architecture | Yes | ||
| Programme spend to USD 50,000 | Yes | ||
| Capability building and training | Yes | ||
| Pausing anything she considers unsafe | Yes, immediately | ||
| Putting anything into production | Recommends | Decides | |
| Anything touching customer or claims data | Data owner and DPO | ||
| Anything affecting an underwriting or claims decision | Regulatory question. Board and compliance | ||
| Headcount changes arising from automation | Executive, with People | ||
| Public or board statements about AI adoption | CEO |
The gate
What a pilot must show to go live
- A named business owner who is accountable after go-live, not the person who built it
- A measure agreed before the pilot, and a result against it
- A human review step for every output that reaches a customer or a decision
- Data approval from the data owner, in writing, naming what data and for what
- A stated failure mode and what happens when it occurs
- A cost per use, and a comparison against doing it the current way
- A way to turn it off that someone other than the builder can operate
Plan
What the 90 days does
-
The claims data pilot is paused
Same day. Not stopped, paused, pending the data owner's review. Nothing about this implies the team did anything wrong.
-
All nine pilots inventoried against the gate
Owner, measure, data, cost, failure mode.
-
Recommendation to stop six
With the reason for each, to the executive. Three duplicates consolidate into one.
-
The two good pilots go to the production decision
Executive decides. Elena recommends.
-
Capability, not tooling
The reason nine pilots started independently is that nobody knew what anyone else was doing. A register and a monthly forum fixes more than a platform would.
-
Review
Fewer pilots, one or two things in production, a gate that outlives the mandate.
Stop
When this mandate should end early
| Measure | Today | What good looks like by 12 July |
|---|---|---|
| Pilots running | 9 | 3, each with an owner and a measure |
| Things in production | 0 | 1 or 2, through the gate |
| Deployment gate | None | Written, used, and it outlives the mandate |
| Pilots with a named accountable owner | 7 of 9 | All of them |
| Data approvals in writing | Unknown, at least one missing | All of them |
| Cost saved or revenue gained | Claimed, never measured | Measured on whatever reaches production |
Fewer pilots and one production system is a better 90 day outcome than nine pilots and a strategy document, and the executive has agreed that in advance so that week four is not a surprise.
Document control
Revision history
| Version | Date | Author | Change |
|---|---|---|---|
| 1.0 | 2 April 2027 | Elena Petrova | Agreed at executive committee, minute 2027-19. Effective 12 April 2027. |
| 0.3 | 29 March 2027 | Elena Petrova | Draft. Stop condition added at clause 5 at my request after the regulated-decision question came up in discussion. |
| 0.2 | 25 March 2027 | Elena Petrova | Draft. Immediate pause authority added after the claims data pilot was identified. |
| 0.1 | 20 March 2027 | Elena Petrova | First draft from the AI adoption diagnostic of 16 March. |