A second worked example: the forecast the agent refuses to give

The most valuable forecast an agent produces is sometimes the one it declines to commit. Picture a solo founder closing the month with 220k of open pipeline across eighteen deals and a board call on Friday.

The naive number. Weighted pipeline says 132k will close. That is the figure the old spreadsheet would have handed up, clean-looking, defensible, wrong.

What the agent does instead. It splits the 220k by data quality before it weights anything. Eleven deals (140k) have a fresh next step, two-way activity inside fourteen days, a verified contact and a close date that has not slipped. Four deals (52k) have one missing or stale signal each, a close date pushed once, or no activity in three weeks. Three deals (28k) are effectively unknowable, blank next step, no logged meeting since the deal was created, one with a contact the decay rule flagged as a job-changer.

The honest output. "Of 220k open, 140k rests on verified-fresh data and forecasts at 95k weighted; 52k is fragile and I have marked it Best Case, not Commit; 28k I cannot forecast at all and recommend you either qualify it out or call the contacts this week. Your defensible commit is 95k, with a credible upside to 130k if the fragile deals firm up." The founder walks into the board call with a number she can defend line by line, plus a three-deal call list that materially changes the forecast if she works it.

Contrast the two. The naive 132k is a single brittle figure that breaks the moment one phantom deal slips. The agent's version is a distribution with a worklist attached, and it converts the forecast from a guess into a plan. This is the same discipline that makes the funnel metrics that move revenue trustworthy, never report a number you cannot click down into the records and signals behind it.