The AI 100-day plan: what a portfolio company should ship in its first quarter.
One process, one owner, one number: what should be true on day 100, how to get there, and what to leave out.
An AI 100-day plan is a short, dated plan for one portfolio company that ends with one AI use case live in production and measured against a signed business case. It is not a strategy, a tool rollout or a list of pilots. If nothing is running with real work by day 100, the plan has failed, however good the slides are.
Most post-close AI plans go wrong in the same way. They try to cover the whole company, they start with a platform decision, and they produce a roadmap in month three. The better plan is narrower: one process, one owner, one number.
What should be true on day 100.
- One use case is live with real volume, not a demo or a sandbox.
- A named person inside the company owns it and wants to keep it.
- Adoption, hours saved and error rate are measured weekly against a baseline taken before the build.
- The CFO has signed the business case and agrees with how the saving is counted.
- A ranked list of the next two or three use cases exists, with rough ranges.
Days 1 to 14: choose and prove the case.
Pick the process before the technology. The best first candidates share four traits: high volume, clear rules, a measurable unit of work, and a manager who is tired of the backlog. Document intake, data checks, reconciliations and first drafts of standard reports are common choices.
In the same two weeks, take the baseline. Count the volume, the minutes per item, the error and rework rate and the fully loaded cost. Write a one-page business case with a range, the assumptions and the one dependency most likely to break it, which is usually data access. Have the CEO and CFO sign it. A case nobody signed is a case nobody will defend in month six.
Days 15 to 50: build with the company’s own team.
Build the first version with the people who do the work and the engineers who will maintain it. A build handed over by outsiders rarely survives the first exception it was not designed for.
Run it in shadow mode first. The system does the work in parallel, people keep doing it the old way, and the team compares the two every day. Shadow mode finds the exceptions, shows the reviewers that the output can be trusted, and produces the accuracy figure the business case needs.
Settle three things during the build: who reviews the output and how often, what happens when the model is wrong, and what data may and may not be sent to an outside vendor.
Days 51 to 100: go live and measure.
Move real volume across in stages, starting with the simplest cases. Keep a person in the review loop until the error rate has held below the agreed level for several weeks.
Report four numbers every week: the share of volume going through the new process, hours saved, error rate against the baseline, and the running cost. Translate hours into money only in the way the CFO agreed at the start. In most mid-market companies the first benefit is capacity, meaning growth without adding back-office headcount, and it should be reported that way.
What to leave out of the first 100 days.
- A company-wide AI strategy. It is easier to write after one thing works.
- A platform or vendor selection exercise beyond what the first use case needs.
- More than one build at a time, unless the company has two separate teams ready.
- Headcount reduction targets. Agree any people plan with management before staff hear about it.
- Customer-facing AI features, unless the product team already ships them well.
Who owns what.
The operating partner sets the target and removes blockers. The CEO sponsors it and picks the process owner. The process owner decides what good output looks like. An engineering lead owns the build and the running cost. The CFO owns the measurement. If any one of those five seats is empty, fill it before day one.
After day 100.
The plan should end with a decision, not a celebration: scale this use case, start the next one, or stop. Across a portfolio, the more valuable result is a design that can be reused. A process that works in one company can usually be deployed in the next similar company much faster, because the hard questions have already been answered.
If you want a plan like this written before close, it is the last section of an AI diligence memo. If the deal has already closed, an AI Value Sprint runs on the same timetable.