You are Dana Whitfield, managing partner of Atlas Advisory Partners, a strategy and compliance advisory firm in Philadelphia. Your new artificial-intelligence platform has just been brilliant and disastrous in the same fortnight, and this Monday you must set how the firm will work with it for the first quarter.
The situation
Atlas has 240 professionals, about 180 engagements a year, USD 68 million of revenue and an 18% EBITDA margin. 65% of its revenue is fixed-fee: every hour saved is margin, and every hour redone is lost. Keystone Health Systems brings in 22% of revenue, and its contract lets it claw back fees on a defective deliverable.
The two-week pilot left two documents on your desk. A market-sizing model the platform produced in 4 hours against a 60-hour budget, which two partners scored at 94% of the human benchmark. And a client-specific regulatory memo, flawless in form, that cited three regulations that do not exist; a second-year consultant sent it on without opening the citations, and the client's counsel did open them. This quarter the firm has booked USD 380,000 of rework, against 45,000 in a normal quarter.
Beacon & Rowe, your closest rival, says it delivers the same engagements 40% faster with 30% fewer staff hours, and last quarter it beat you on two fixed-fee bids. Keystone has put "AI usage controls" on the agenda of a review six weeks out. The firm's work reduces to six task families that take 255 hours a quarter done by hand. For each one you can decide who does it and who checks it, and you can commission at most two controlled probes a quarter. Inside the firm, one camp wants everything delegated now; the other wants the platform banned from anything a client sees.