What the sessions surfaced
Three phrases recurred: “there is no need to reinvent the wheel”, “everyone works alone, there is no shared brain”, and “there is no time”. Individual use of AI was already sophisticated. What was missing was the shared layer that turns personal habits into a firm asset — and protected hours to build it.
Four ground rules, agreed before any tool
Buy first and build only for the gap. Keep a human in the loop, always. Keep the knowledge in the firm, with support designed to taper and end. Start small and measure against the firm’s own work, never a vendor demo.
Eight domains, scored and sequenced
Administration and bids became the first pilot: repetitive work, hard-to-recruit skills and zero AI use, so every gain is visible. The project archive is the foundation for later phases. Visualisation was already about ten times faster and was left alone. Financial insight was excluded at the client’s request — recorded, not argued.
Two of the three items in the “start here” quadrant were not software at all: programme ownership and three management decisions (one archive or two, what is already paid for, who may access what).
A pilot that can be judged
One domain, two people, three to five weeks. The output already exists today, so “better” is measurable. The system learns from past inputs and outputs rather than asking busy experts to explain their judgement. Five measures are agreed in advance: cycle time (40% faster or better), accuracy (above 90%), unprompted weekly use, load displacement and new capability. If it misses, the tool changes — not the scope.
Economics the approver can trust
The 24-month model counts every cost, including the internal lead’s own hours, and prices only time saved. Break-even lands in month seven; value reaches about 3.9 times the investment by month 24. The number that converts a cautious approver is the exposure before proof: about 18% of the total, reached at week eight.