The two failure modes
We see two versions of AI in 100-day plans.
The first ignores it. The plan covers finance upgrades, pricing, procurement, the usual levers, and AI gets a line in year two under "digital". By the time anyone returns to it, half the hold period is gone.
The second overcorrects. AI becomes its own workstream with its own steering committee, its own consultants, and a tool selection exercise that runs for a quarter. Nothing ships. The management team learns that "AI" means meetings.
Both approaches treat AI as a separate initiative, rather than a way to deliver work already in the plan.
What the first 100 days are actually for
The first hundred days are for two jobs: stabilise the business and identify where the value will come from. AI has a role in both, but a supporting one.
In the stabilisation phase, the highest-value AI work is often basic data plumbing. If management accounts take fifteen days to close, an automated reporting pipeline pays for itself before the first board meeting. If sales leads sit in an inbox for two days before anyone responds, an automated triage and response system is a week of build for a permanent uplift in conversion.
During diagnosis, AI can speed up a diligence-grade review of workflows, systems and data quality. Processing documents, interviews and system exports with AI cuts the work from months to weeks and produces a sharper value creation plan.
The sequencing that works
A practical sequence has three stages:
Days 1 to 30: instrument the business. Map where time and margin actually go before starting a build. Identify the three processes that are most manual, most repetitive and closest to revenue or cost. Look for high-volume, rules-based work currently done by people who would rather be doing something else.
Days 30 to 60: ship the first build. Choose one project for speed of payback rather than strategic elegance. Automated lead response, document intake, or reporting automation are the usual candidates. Within the first two months of ownership, management should see a working system handling a real process inside the business.
Days 60 to 100: build the pipeline. Once the first system is live, management can judge it on evidence and nominate other processes. Set the hold-period roadmap at this point, prioritised by P&L impact and assigned to owners with dates.
What to keep out of the first 100 days
Three things do not belong in the plan:
- Enterprise AI platform selection. Choosing a platform before you know your use cases is how portcos end up paying for software nobody uses.
- Customer-facing AI. Chatbots and AI product features carry brand risk and need mature data foundations. They are hold-period projects, not day-one projects.
- Headcount promises. If the plan says AI will remove thirty roles, adoption dies on the spot. Frame the early work as capacity: the same team handling more volume, faster.
The test
A simple test for whether AI is in your 100-day plan correctly: by day 100, is there at least one AI system running in production, doing work the business used to do manually, with a number attached to it?
If the only output is a strategy document, add a production build and a measurable operating result to the plan.
At Healf, the FT's #1 fastest-growing company in Europe, the fastest project we ran paid back in twelve weeks. Hold the first build to that standard. Everything we build is aimed at the P&L, and the first hundred days set that expectation.