The dashboard diagnoses. This agent acts. It reads one rep's SPICED gaps from the data model, pulls the matching manager playbook section, hands the rep a ready-to-run prompt plus a weekly reinforce and improve pair, and follows up on the Tuesday 1:1 cadence. It is the arc of the flywheel that turns a diagnosis into rep behavior change, and it is how we move the lagging number. This document is the spec an engineer builds against: inputs, outputs, rules, and the closed conditions.
Three sources, all already produced by the WS1 pipeline. The agent adds no new data collection. It consumes the scored data model, treats the manager playbook as a queryable knowledge base, and reads the rep's most recent Grain calls for fresh evidence.
| Input | Source | Shape the agent reads |
|---|---|---|
| SPICED gaps | The scored data model (per rep, per deal) | Per dimension: Completion (yes or no) and Quality (1 to 5), a state (weak, solid, strong), and a gap boolean. The revenue-linked headline is Completion; the coaching signal is Quality. Impact and Critical Event are the completion binaries the scorer detects. |
| Manager playbook | WbD method library, indexed as a knowledge base | Each play keyed to a SPICED dimension and framed in REKS, retrievable by gap. The agent queries it by the rep's specific gap, so a weak Impact returns the Impact-quantification play, not the whole library. |
| Recent Grain calls | Grain, segmented and speaker-resolved, joined on the HubSpot deal id | The last few genuine sales-cycle calls for this rep, with the verbatim customer line, the rep's actual response, and the deep link to the call moment. This is the evidence the agent quotes back, never invented. |
Every gap the agent names traces to a scored dimension in the data model, and every quote it shows traces to a verbatim Grain segment with a deep link. The agent produces no claim that is not backed by one of these three inputs. If the evidence is not present, the agent stays silent on that dimension rather than manufacturing a coaching moment.
Two outputs per cycle. First, for each open gap, the matching playbook section plus a ready-to-run prompt the rep pastes into their own AI. Second, a weekly reinforce and improve pair: one win to repeat, one growth moment to work, each as a four-part card grounded in the rep's own calls.
For each gap, the agent emits a triple. The gap names the dimension, the play is the retrieved method, and the prompt is executable by the rep on their next real deal.
One win, one growth moment. Development framing, always paired, so the rep reads it as coaching and not a grade. Each is a four-part card: the customer line, the rep's actual response, the verdict and why, and the next action.
One win and one growth moment per week is deliverable inside a single 1:1 and is small enough that the rep can actually change one behavior before the next cycle. More than one growth moment at a time is where coaching turns into a grade and the rep stops trusting it. The pairing is the mechanism: the win earns the right to name the gap.
The agent runs on the manager's existing ritual, so it needs no new habit to survive. The manager delivers the pair and the play inside the Tuesday 1:1 they already hold, and the manager can override before anything reaches the rep. The rep runs the prompt on their next real call. The next scored call closes the loop.
A coaching tool that needs a new weekly habit dies. This agent attaches to the Monday pipeline review and the Tuesday 1:1 the manager already runs. The one required manager act is small: review the pair, deliver it. Everything upstream is assembled for them, and everything downstream is measured for them.
The first wrong call the rep catches is the moment they stop trusting the agent. These three rule sets keep the human manager in the loop and give the rep a visible path to correct the agent.
The rep always has a one-click way to say the agent got it wrong, and that flag always reaches a human manager who can override. The rep owns the relationship with the agent, and the manager owns the final say. The agent never has the last word on a rep.
The agent does not take the rep's word or the manager's word that a gap closed. It re-scores the next genuine sales-cycle call on the same dimension and measures the week-over-week delta. The closed condition is explicit and measured, so the loop has a real exit.
CBRE, rep John Haire, Impact opens at 2.0 and incomplete. The agent hands the Impact-quantification play and the ready prompt. Exit condition: the next call shows an asked impact question and a quantified consequence, Impact re-scored above 3, Completion flipped to yes. The manager, Josh Klatz, sees the before and after in the pipeline review and clears it. The gap is closed by measurement, not by assertion.
The agent moves one leading metric: the rate of Impact and Critical Event completion across the rep's open deals. That metric is chosen because it is the pair most reps miss and the pair most correlated to revenue. Tying the leading behavior change to the lagging revenue number is how the agent proves it earned its cost.
| Layer | The metric | Why this one |
|---|---|---|
| Leading the agent moves this |
Rate of Impact and Critical Event completion across the rep's open sales-cycle deals, week over week. | Impact, Critical Event, and Decision Process are the most common team gap. Completion is the revenue-linked binary the scorer already detects, so the agent moves a number the pipeline already measures. |
| Lagging the business cares about this |
Win rate and sales-cycle time, read on the same deals in HubSpot. | These are what renewal and leadership buy on. The agent is unbuyable at renewal unless the leading behavior change is shown to move these. |
The WS1 product definition anchors the claim: sellers at full SPICED completion average plus 10 percent win rate and plus 29 percent quota attainment, and completion of Impact and the Critical Event carries the largest ARR lift of the dimensions. The agent's job is to reproduce that link on WbD's own pipeline, so its value is provable at renewal, not asserted.
The agent is a development tool. The rep owns it. It is held between the rep and their manager, and it is measured as growth. Read the other way, it becomes surveillance, and a surveillance tool gets gamed and abandoned. These guardrails are load-bearing for adoption.
Build every surface so a rep would choose to open it. The moment the agent reads as the manager watching, the rep games the calls and the leading metric stops meaning anything. Development framing is the design constraint, and it is what makes the lagging number real.