Deal Pipeline Tracker Agent

When to use

Use this when origination is run off a list of names and nobody can say what will close. Typical triggers: a partner asking whether the year's target is reachable, a review where every opportunity is somehow still active, or a team fully occupied and closing nothing. Reach for it when the question is not which deals are live but how many will land and whether the team can carry them.

What it does

It produces a managed pipeline: stage definitions with objective entry criteria, cohort conversion between stages, the top-of-funnel volume implied by the closing target, work-in-progress limits set against real capacity, staleness rules, and a weighted forecast reconciled to what closed.

Method

1. Define the stages MECE. One opportunity, one stage, no overlaps.

   • Mapped, contacted, qualified, engaged under NDA, indication submitted, exclusivity, signed, closed — with passed, lost, and watchlist as terminal states.

2. Make each entry criterion an observable event. Not a feeling.

   • "Owner agreed a second meeting and sent revenue" is a criterion; "warm" is not. If two people would stage the same deal differently, the definition is not written yet.

3. Give every opportunity an owner and a dated next step.

   • A live opportunity with no date on its next action is not live, and that one field catches most of what a pipeline review would otherwise miss.

4. Measure conversion by cohort, not snapshot.

   • Follow the deals that entered a stage in one period through to their outcome; a snapshot ratio flatters whenever the top of the funnel is growing.

5. Back-solve the top of the funnel from the closing target.

   • Multiply the stage conversions into a cumulative rate, divide the target by it, and test the answer against the contacts the team can actually make in a week.

6. Set work-in-progress limits against real capacity. Use Little's Law.

   • Work in progress equals throughput times cycle time: deals entering a stage each year times the months they sit there gives the concurrent load to set against the number of teams.
   • When it says the constraint is capacity, more origination makes the numbers worse, not better.

7. Kill or park the stale ones. Carrying them is not free.

   • Flag anything past 1.5x the median time in its stage or lacking a dated next step, then move it to the watchlist with a re-approach date rather than leave it in the count.

8. Report a weighted forecast and reconcile it. Adjust the rates, not the story.

   • Weight each opportunity by its stage's cumulative probability of closing, then compare last period's forecast against what closed and re-cut the conversions.

Inputs

• The closing target for the period and the fee or capital per closing
• The opportunity list with stage, owner, and last action date
• Historic outcomes for at least two years of cohorts
• Median time in stage, by stage
• The number of deal teams and what one live process consumes
• Screening criteria, so a mis-fit is passed rather than staged

Output format

• Stage definitions, each with its entry criterion stated as an observable event
• Stage-to-stage conversion by cohort, naming the period measured
• The implied top of funnel: qualified opportunities and contacts needed per period
• A capacity section applying Little's Law, stating concurrent load against teams
• A staleness list naming what to kill, park, or restart, with dates
• A weighted forecast in expected closings, reconciled to the prior period
• Present the pipeline and its metrics in prose, never as markdown tables

Example

For Ardsley Partners (fictional, illustrative), a mid-market desk targeting three closings a year: trailing cohorts convert qualified to engaged at 40 percent, engaged to indication at 50, indication to exclusivity at 25, exclusivity to signed at 70, and signed to closed at 90 — a cumulative 3.15 percent. Three closings therefore need 95 qualified opportunities a year and, at 30 percent qualification, 317 contacts, or six a week. Little's Law then binds: 4.8 deals must enter exclusivity a year, each holding a team for four months, a concurrent load of 1.6 against two teams. The active list of 41 — 24 qualified, nine engaged, five at indication, two in exclusivity, one signed — weights to 4.41 expected closings; stripping the 12 with no dated next step, eight qualified and four engaged, leaves 29 and 3.85. The forecast fell by 0.57 and the outcome by nothing: against a target of three, the binding constraint was never origination.