AI forecasting: revenue and margin in real time

Not everything in a consulting business is forecastable to the same degree, and treating it as if it were is what makes forecasts lose credibility. Committed backlog and recurring engagements can be projected with reasonable confidence. New logo revenue in a quarter cannot, and no model improves that much.

The productive approach is to forecast each layer separately and present them as separate layers: signed work, extensions with a history of renewing, weighted pipeline, and everything else. A leadership team that sees the four layers argues about the right one instead of about the total.

Revenue forecasts get attention; margin forecasts get built once and then quietly diverge. The drift comes from a small number of recurring causes: seniority mix heavier than sold, non-billable time growing, rate increases not applied, and overruns absorbed rather than negotiated.

Each of those has a leading indicator available weeks before it shows in the accounts. Tracking the four as a small set — rather than waiting for the month-end margin figure — turns a forecast miss into a series of small corrections, which is the only form in which margin is actually recoverable.

A forecast that arrives as a single number invites a negotiation about the number. One that arrives as a range with the assumptions written down invites a conversation about the assumptions, which is the conversation that changes behaviour.

The practical format is short: the range, the three assumptions that move it most, and what would have to be true for the low end or the high end to happen. Keeping the same three assumptions visible month after month also makes the forecast improve, because the misses become traceable to a specific belief that turned out wrong.