Discovery to handoff · point of view
Deals close without the requirements needed to go live.
Not because reps are careless — because the missing thing is almost always vertical-specific. Age verification, lottery, scale labelling. Knowledge, not effort.
4 of 8
sample deals slipped or doubled their timeline
3 of 4
traced to a vertical requirement never discussed
1 of 4
was not a discovery problem — and no tool should claim it
Why this piece
The bridge between where knowledge exists and where it's needed.
The rep · in the field
Records on whatever they already carry. Confirms with taps. Never types.
Transcript-source agnostic by design — the recorder they bought themselves works. The vertical requirement set flags what was never discussed.
→
The consultant · in Salesforce
Opens a record with nothing to chase. No re-interview.
Native fields, Tasks and notes — no new consultant-side tool. The one real risk is flagged above the fold before kickoff, not found during configuration.
Rep touches Salesforce once: a confirm screen showing exactly what will be written. The tool types.
Answers arrive in Slack, where reps already are. No new inbox to forget.
Soft gate: an unresolved blocker flags the record but never blocks the deal. Hard gates teach reps to route around tools.
Deliberately not built
What I cut, and why the cut is the point.
A consultant workspace
Consultants run ~25 go-lives inside Salesforce. The tool writes native fields, Tasks and notes there — and leaves. No second login.
Transcription
Commodity, and reps already own recorders. The contract is (transcript, vertical) in → structured requirements out. Any source qualifies.
A hard gate on closing
A real deal can outrun a verification — competing bid, fixed timeline. The flag travels with the deal instead of blocking it.
Close-rate claims
Leadership's first goal, deprioritized on purpose: the mechanism here is rep time back and rework avoided. If close rate moves too, good — but I won't promise it.
The wider motion is mapped, not built: 33 annotated frames cover discovery health, onboarding queue by risk, and sequencing — the pieces I'd reach for next, in order.
Build vs buy
Buy the commodity. Build the knowledge.
Buy
Transcription + capture — recorder-agnostic; reps keep what they carry
LLM extraction — model spend concentrated on details: vendor, model, counts, dates
Slack + Salesforce rails — standard APIs, upsert on the opportunity key, never duplicates
Build — this is the asset
The vertical requirements library — ~15 cards per vertical, blockers vs standard. No vendor has Toast Retail's knowledge. An SE maintains it in-app; if it needs engineering, it rots.
The extraction contract — closed enum (the model can't invent a requirement), confirmed / inferred / null, and: not discussed is never not needed.
Offline tier — checklist and alias matching run local and instant; a dead-zone parking lot still shows what matters.
How we'd know it's working
One number decides. Everything else is a leading indicator.
Go-lives hitting the expected date
If discovery improves and this doesn't move, the problem was somewhere else — and I'd say so.
Baseline*50%
→
Target*75%
LeadingVertical blockers confirmed pre-close
LeadingConsultant re-interviews before kickoff
OutcomeRep hours pulled back post-close
Counter-metricFlags dismissed as not applicable — if it spikes, the flags are wrong
*Illustrative — first step is tagging two quarters of real go-lives by whether the slip traced to discovery.