[!info] This is the canonical discovery-call practice as of June 2026. Earlier discovery-call guides have been moved to archive/.

Discovery Call Playbook — What’s Working, What to Fix

A living guide synthesized from real call debriefs. Updated after each call.

Evidence base: 2026-05-18-brett-1949-sagetap-call, 2026-05-20-kevin-5542-sagetap-call, 2026-05-20-joseph-1446-sagetap-call, 2026-05-29-josephb-sagetap-call, 2026-06-29-nicholas-e-sagetap-call

The Chat Surface — Probe Before You Spend a Call (Bryan, 2026-07-01)

Sagetap lets you send a chat message to any applicant, for free. That means the funnel is not “read their answers → decide call or decline.” There is a cheap middle step: chat.

Rule: every initial evaluation must decide a specific chat probe, not just yay/nay. For each applicant — regardless of fit rating — ask: is there a specific question or two I can send in chat that would (a) resolve whether the call is worth it, or (b) gather more useful survey data? The written answers are the free harvest; chat is the free follow-up harvest, and it de-risks the one expensive resource (the call).

Why this matters, by tier:

🟡 / 🟠 borderline: one sharp question usually decides accept-vs-skip. Example: a strong-answers applicant whose only initiative is security — ask “is the reasoning-visibility pain part of your security budget, or a separate line?” The answer converts a coin-flip into a decision without spending a call.
🟡 with 0 initiatives (real pain, no buying motion): ask “is there an active project where you’d put budget behind better agent visibility, or is this exploratory right now?” If the answer is a project, they graduate to a call; if exploratory, you saved the credits and kept the relationship warm.
🔴 / clear skip: you’re not going to call them, so ask the question that sharpens the thesis or campaign at zero cost. Security CxOs are a free survey panel: “when you say visibility, do you mean seeing what agents did, or blocking what they shouldn’t — and which is the budgeted priority?” Their answers calibrate how the campaign titles are being read.

The chat question should reference something concrete from their own answers (so it reads as “I read yours,” not a template), be answerable in one reply (low friction), and — because it’s pasted into Sagetap — be plain text.

This is now encoded in sagetap-intake (Hard Rule 10 + the required Chat Probe section) so every evaluation produces one, and tracked per-applicant in sagetap-applicant-tracker.

The Structural Problem

Version 1 (Brett, Kevin calls): Discovery is excellent. The pitch gets crushed. Rich discovery for 20-22 minutes → compressed pitch in the final 5-7 minutes → soft close. The prospect understands the problem but not the product.

Version 2 (Joseph_1446, JosephB calls): Discovery barely happens. The pitch takes over from the start. Extended monologue in the first 7-10 minutes → prospect asks clarifying questions to understand what you do → concrete product features arrive too late → soft close or pass.

The problem is getting worse, not better. The Joseph_1446 call showed the first regression (pitched first, discovered second). The JosephB call took it further — the first 7 minutes were continuous pitch with no questions asked, despite JosephB opening with an invitation to focus the conversation. The prep had specific discovery openers designed for his qualifying answers. None were used.

This is now two distinct failure modes requiring different fixes:

Fix 1: Don’t pitch first. Discover first.

The first five minutes of every call should be questions, not statements. The prep notes have specific openers designed for each prospect’s qualifying answers. Treat them as a literal script for minutes 0-5. Once the prospect is talking, freestyle. But do not open with a vision pitch.

Hard rule: The words “data substrate,” “super structures,” and “systems of understanding” should never appear in the first 10 minutes. Lead with the prospect’s own language from their qualifying answers.

Fix 2: When asked “what do you measure?” — name five things

JosephB asked “what exactly are you going to measure?” three times and got architecture instead of metrics. The five metrics (from 2026-05-29-roi-measurement-framework): cost per merged PR, token waste, agent comparison, rework rate, review burden. Say them by name. Print them on a card. This is the answer to the #1 enterprise buyer question.

Fix 3: Weave, Don’t Stack

Don’t save the product for a separate section after discovery. Let the discovery moments be the pitch. When the prospect describes a pain point, name the Pathbase feature that addresses it in that moment, then move on. By the time you reach minute 20, they’ve already heard the pitch — it was embedded in the conversation.

Example from the Kevin call:

Kevin describes agents duplicating functions across the codebase at ~6:29
Instead of filing that away for the pitch later, say: “That’s exactly what cross-session traces catch — you’d see Agent A and Agent B both generated implementations of the same logic, before the PR.”
Then ask the next question. Don’t monologue. One sentence that connects pain → product, then move on.

Example from the Brett call:

Brett describes the null-check bug at ~10:40
Instead of pivoting to the Minority Report concept, say: “If the agent session trace existed, you could open it and see the exact moment it decided the null check was redundant. That’s what Pathbase captures.”
Then probe deeper: “How often does this happen? Is this a one-off or a pattern?”

Mental Timer

By minute 15, start transitioning. Say: “I want to make sure we have time to show you what we’re building — can I give you the 5-minute version?” This guarantees 10+ minutes for pitch, questions, and close.

What’s Working

1. Giving people space to talk

Both Brett (quiet, direct, short answers) and Kevin (verbose, tangential, generous with context) produced their best signals when given room. The null-check bug, spec-as-prompts, cross-codebase blindspot, two doors, rules fragility — all volunteered, not extracted. Keep doing this.

2. The dead-ends / cadaver / Minority Report concept

This is the single strongest product concept in the pitch. Both Brett and Kevin engaged most with this idea — that engineering knowledge lives in the process (dead ends, corrections, decisions) and is thrown away today. Kevin restated it better than you pitched it: “there’s great knowledge in that, in just learning the code base and getting people up to speed.”

Lead with this concept, not the architecture. “Toolpath is an open format” and “we create a super graph” are implementation details. “You can see the dead ends and decisions that led to a PR, not just the final diff” is the idea that lands.

3. The accordion problem framing

Landed with Kevin. Useful metaphor for the generate → review → re-review loop. Keep in the toolkit.

4. Peer-to-peer tone

Both calls felt like conversations between peers, not vendor pitches. The Schneider Electric connection with Brett, the co-founder’s Google background with Kevin — these build credibility without overselling. Keep the human, conversational register.

5. Reading the room

Bryan adapts well to different communication styles. Short questions for Brett (man of few words). More exploratory conversation with Kevin (riff-friendly CTO). This is a real skill — don’t lose it by over-structuring the calls.

What to Fix

1. Ask the universal probes — every call, no exceptions

Three questions that should happen on every call. Zero excuses for skipping them.

Probe 1 — The mental model reveal:

“If I showed you a structured trace of everything one of your engineers’ AI agents did yesterday — every tool call, every file edit, every decision — what’s the first thing you’d look for?”

This sorts Archetype 1 (“whether the work was good”) from Archetype 2 (“whether it touched anything it shouldn’t have”) in one answer. Not asked on either call.

Probe 2 — The champion chain:

“Who else in your org would care about this data?”

Reveals internal buyers, multi-stakeholder opportunity, and whether this person is the decision-maker or a referral. Not asked directly on either call (Kevin mentioned the Digital Performance Team organically, but the explicit question wasn’t posed).

Probe 3 — Purchase intent:

“What would you need to see to put budget behind something like this?”

Direct and necessary. Especially critical for Kevin (6 active initiatives, Jun 2026 target — he’s spending right now). Not asked on either call. Brett volunteered the pre-budget signal organically but without the explicit probe you can’t gauge timeline.

2. Probe deeper on pain — don’t pivot to solutions

When a prospect describes a concrete pain point, the instinct is to jump to the solution. Resist. Probe twice before pivoting.

Call	Pain described	What happened	What should have happened
Brett	Null-check bug	Asked “how did you approach that postmortem?” (good), then pivoted to Minority Report concept	Second probe: “How often does this happen? One-off or a pattern?” Third: “What would you have wanted to see in the agent’s history?”
Brett	Specs written as prompts	Acknowledged, moved on	”Can you give me an example? What does a spec-as-prompt look like vs. a traditional spec?”
Kevin	Code review tools failing	Agreed they demo well but fail in practice	”What specifically are you testing for when you insert bad code? What do they miss?”
Kevin	”Two doors” automated triage	Agreed, moved to general commentary	”What would the criteria be for door A vs. door B? What signal would you need to make that routing decision?”

Rule of thumb: When a prospect describes a pain point unprompted, they’ve given you a gift. Unwrap it fully before moving on. Two follow-up questions minimum.

3. Use their pre-call answers

Both calls left high-signal pre-call answers unprobed:

Kevin: “None have solved the problem for us yet” — never asked what they tried and why it failed
Kevin: “Our qual data is showing tremendous benefits in both HOW we work and WHAT we work on” — never probed what “how we work” means
Brett: No qualifying answers (outbound call — this was the lesson; send questions ahead next time)

Before every call, pick 2-3 phrases from their qualifying answers to probe. Write them on a sticky note. These are the sharpest entry points you have — the prospect already told you what they care about.

4. Say “Pathbase” — not “Pathways”

Said “Pathways” on both calls. If a prospect searches for “Pathways” after the call, they find nothing. Fix this. It’s a small thing that erodes professionalism.

5. Close with a specific action, not a vibe

Call	Close	Strength
Brett	”Let me talk to a couple of my senior engineers, maybe we can get something together”	Specific — named who and what action
Kevin	”I’m intrigued… we’ll see about taking it forward”	Vague — no who, no what, no when

The Brett close is better because it has a specific actor (senior engineers) and a specific action (talk to them, schedule something). The Kevin close has neither.

Template for a stronger close:

“Based on what you’ve shared, I think [specific thing] would be the most useful next step. Can we get [specific people] on a 30-minute demo call in the next [timeframe]?”

If the prospect deflects to “I’ll review and follow up,” pin it down: “Happy to send you [specific thing]. When would be a good time to reconnect — next week?“

6. Don’t overpromise on timeline

“Any day now” and “if you need this today, we can probably get it live for you today” sets an expectation that the product is fully ready. If Kevin follows up this week and the onboarding isn’t smooth, credibility is damaged. Better: “We’re onboarding early design partners now. I’d love to get you in — how about I set up a walkthrough for [specific date]?” Same urgency, less risk.

7. Stop signaling “we’re early” — NEW from JosephB call

Phrases like “our early design partners,” “we’re launching with,” “beginning to have a unified data substrate,” and “we can probably get it live for you today” all send the same signal: this product isn’t ready. JosephB’s feedback — “not mature enough” — was partly about the product, but was definitely also about how the product was described. Paolo flagged this in the post-call debrief: “We shouldn’t at all act like we’re exploring the territory.”

The fix: Talk about what the product does, not what stage it’s at. “Pathbase captures session traces and computes cost per merged PR, token waste, agent efficiency, rework rate, and review burden” vs. “We’re launching with some early features for our design partners.”

8. Don’t position as infrastructure — position as a product with opinions — NEW from JosephB call

When JosephB asked “is this basically a platform to query?” — that was him downgrading Pathbase from “product I buy” to “infrastructure my team builds on top of.” His Sagetap feedback confirmed this: “just delivers information, but without specific insights, suggested action items, or downstream process.”

Enterprise VPs buy products that answer questions. They don’t buy infrastructure they need to build on top of. Datadog doesn’t say “we give you a data substrate for observability.” They say “here’s your error rate, here are your slow endpoints, here’s what broke.”

The fix: Lead with the five questions Pathbase answers (see 2026-05-29-roi-measurement-framework). Present the data layer as how the product works, not what the product is.

9. Define roles for joint calls — NEW from JosephB call

On the JosephB call, both Bryan and Paolo pitched without a defined division of labor. Paolo’s “that’s our wheelhouse” at 10:40 interrupted JosephB’s best discovery moment (the easy-to-hard ROI spectrum). His “no developer friction” at 18:15 answered a question JosephB didn’t ask.

The fix: Before any joint call, define: who asks questions (discovery), who provides product answers when tagged in (pitch), and what the handoff signal is. One person should never interrupt a prospect’s train of thought to pitch.

Call Structure Template

Based on two calls, here’s the target structure for a 30-minute Sagetap call:

Segment	Time	What to do
Warm-up	0–2 min	Build rapport. Acknowledge their time. One personal note.
Context check	2–5 min	Confirm/expand qualifying answers. “You mentioned X — is that still the case?”
Discovery	5–15 min	Open-ended questions. Let them talk. Probe twice on pain points. Weave product in — one sentence per pain point connecting to Pathbase.
Universal probes	15–18 min	The three probes: mental model, champion chain, purchase intent.
Pitch	18–23 min	5-minute structured pitch. Lead with the dead-ends concept. Anchor to their pain, not abstract architecture.
Questions & demo tease	23–27 min	Let them ask. Show one thing if possible (even a screenshot).
Close	27–30 min	Specific next step. Named people. Concrete timeline.

For outbound calls (no qualifying answers)

Send qualifying questions via Sagetap message before the call. Even 24 hours in advance gives you material to work with. The Brett call wasted 5-7 minutes establishing baseline context that written answers would have provided.

Phrases That Land

Collected from calls — use these, they resonate:

“You can see the dead ends, not just the final diff” — the core concept, stated simply
“The accordion problem” — expanding generation + compressing review = game of telephone
“GitHub for agent traces” — quick anchor, people get it immediately
“Process is more valuable than the output” — the philosophical hook for engaged CTOs
“Not another Jellyfish” — for anyone burned by rigid engineering intelligence tools (Kevin’s exact anti-pattern)
“Optimized for manual human review” — disarms skepticism about “AI code review” promises. Positions Pathbase as helping humans, not replacing them.
“Vanity metrics” — Joseph’s term for lines of code, prompt counts, token spend, seat utilization. Sharper than “macro indicators.” Use it.
“Did the agent do its homework?” — one-line test for session quality. Landed on the Joseph call.
“Where is your AI spend working and where is it waste?” — the reframe from “what’s the ROI?” to an operationally answerable question. Developed after the JosephB call failure.
“Cost per merged PR” — the atomic unit of AI engineering cost. Concrete, measurable, decision-driving.
The date bug story (anonymized): A Fortune 500 engineering team shipped a production bug where the AI model used its 2025 training year instead of the current year on a date form. A trace-visible failure that only AI-generated code produces. JosephB’s own example — more concrete than the null-check bug because the failure mode is novel and AI-specific.

Phrases to Retire

“Super graph” — too abstract, nobody engages with it
“Structured provenance format” — accurate but academic. Say “the decision history” or “what actually happened”
“Novel data structures” — sounds like a research paper, not a product
“Nervous system of agent activity” — used to open the Joseph call. Too abstract, too architectural. Joseph’s response was literally “Okay.”
“Data substrate” — used on the JosephB call. He heard “platform to query” and downgraded Pathbase to infrastructure. This phrase is now confirmed as actively harmful with enterprise buyers. Say “the missing layer” or “the decision history”
“Systems of understanding for AI” — used to open the JosephB call. Too abstract to mean anything. JosephB’s response: “Cool, I think that can break down to a lot of different stuff.”
“We can probably get it live for you today” — signals the product is either trivially simple or not ready. Undermines the enterprise positioning.
“Unified graph structure interconnected with the Git provenance” — nobody processes this in real time. Say “you see the agent’s decisions alongside the code changes”

Prospect-Specific Lessons

Brett Dolecheck (Xylem/Sensus) — 2026-05-18

Quiet, direct, short answers. Give him space and silence.
His feedback: he wants to know “what is it actually gonna do?” — concrete, not conceptual.
The null-check bug is the single strongest incident-level pain across all discovery. Use it in every subsequent pitch (anonymized).
Don’t monologue — his best contributions came when he had room.

Kevin_5542 (AI Software, Ireland) — 2026-05-20

Talker, tangential, generous with context. Let him riff.
Has evaluated and rejected multiple tools. He knows what doesn’t work. Ask what he’s tried.
The “two doors” concept validates the annotation pipeline roadmap. Bring this up in the demo.
He’s a CTO assembling an entire AI infrastructure stack — 6 initiatives. Position Pathbase as one piece of that stack, not a standalone tool.
Co-creates with enterprise clients — code quality is a client deliverable, not just internal hygiene. The stakes are higher.

Joseph_1446 (DevOps Director, FinServ, Canada) — 2026-05-20

Systems thinker. Processes by asking targeted technical questions. Give crisp 2-3 sentence answers, not vision speeches.
His mental model is Datadog distributed tracing — he wants span-by-span waterfall views of agent sessions. Position against this: “Datadog captures the transport layer. Pathbase captures the decision layer.”
Coined “vanity metrics” for lines of code, prompt counts, token spend. Use this phrase going forward.
Independently arrived at the annotation/scoring pipeline — “marry it with a scoring model to emphasize expected behaviors.” Third Sage to intuit this (after Kevin’s “two doors”).
This was the weakest call execution. Pitched first, discovered second. Monologued when he asked crisp questions. Skipped all universal probes. The prospect quality is high; the call didn’t match it. The follow-up materials need to recover what the call missed.

JosephB (VP of Product Gen AI, Fortune 500 Insurance, Wisconsin) — 2026-05-28

Strongest applicant in the combined set — 2,400 developers, token allocation crisis, explicit observability purchase initiative. Passed on Sagetap.
The call failure was a positioning failure, not a fit failure. He saw the market need (4/5 differentiation, timing) but not the product value (3/5 value). His feedback: “just delivers information, but without specific insights, suggested action items, or downstream process.”
He asked “what exactly are you going to measure?” three times and got architecture instead of metrics each time. The five-metric answer (cost per merged PR, token waste, agent comparison, rework rate, review burden) didn’t exist yet as a framing. It does now. See 2026-05-29-roi-measurement-framework.
His date bug — model fell back to 2025 training year instead of 2026 on a production form — is the strongest AI-specific failure story across all calls. More novel than Brett’s null-check bug because this failure mode is unique to AI-generated code. Use it (anonymized).
He has a VP peer in charge of “AI for the developer experience” who is the likely implementer. JosephB is the strategy leader, she’s the buyer. She was mentioned but never named or probed.
$50/month per developer token budget. Peer at Datadog has $500/week. Two orders of magnitude variance. He doesn’t have a framework for knowing whether his allocation is right. This is the strongest “where should the next dollar go?” signal in the entire program.
The prep was excellent and entirely unused. Specific openers, probes, time-boxed structure, listening-signals table — none deployed. The next call must use the prep or the prep is wasted effort.
Joint call with Paolo had no role definition. Both pitched simultaneously. Paolo’s “that’s our wheelhouse” interrupted JosephB’s best discovery moment. Before any joint call, define who discovers, who pitches, and what the handoff signal is.
Full debrief: 2026-05-29-josephb-sagetap-call

NicholasE (Engineering Manager, Swedish Retail, 200K employees) — 2026-06-29

The first Product Pitch call (not Research/Awareness). He came ready to see the product and said so.
Volunteered the richest unprompted discovery monologue across all calls — 4 distinct pain points in 2 minutes. Let prospects talk; the best signal is unprompted.
“18 different ways of doing things” across 20+ teams — the most quantified organizational pain metric from any call. Quotable.
90% AI-generated code — highest adoption rate from any prospect. Validates the problem is systemic, not occasional.
Contractor session transparency is a new use case: requiring offshore consultants to provide the full agent session with their code. Bryan surfaced this with a brilliant question; nobody followed up on feasibility. Pursue in the follow-up.
“It’s like a diary, almost” — his mental model. The session trace is a record, not a dashboard. This framing may land better with ICs/managers than “the decision history” or “GitHub for agent traces.”
The “we’re early” problem is now a joint-call problem. Alex said “v zero,” “currently working on,” and “focused single users” three times. Bryan added “design partners.” The prospect was ready to try the product and pulled back to “I’ll send engineers.” The cooling was visible and correlated. Before any future joint call, brief Alex on what not to say.
The live-install close window opened and closed. When Nicholas asked “I guess it’s pretty easy to start using it?” at 20:16 — that was the moment. The answer should have been “Want to try it right now?” It wasn’t. Rehearse this close.
The prep was used. First call where pre-call answers were explicitly referenced and the trajectory opener was deployed. Mark the progress.
Follow-up: mid-August (Swedish midsummer). Pin a specific date in the Sagetap message.
Full debrief: 2026-06-29-nicholas-e-sagetap-debrief

After Every Call

Update this playbook if anything new was learned
Add the prospect to the comparison table above
Ask: did I ask the three universal probes? If not, why?
Ask: did the pitch get compressed into the last 5 minutes? If yes, what ate the time?
Ask: did I say “Pathbase” or “Pathways”?
Ask: when the prospect asked “what does this measure?” — did I name the five metrics? (cost per merged PR, token waste, agent comparison, rework rate, review burden)
Ask: did I use the prep notes? If not, why did I prepare them?
Scorecard across all calls:

Call	Discovery before pitch?	Universal probes asked?	Named five metrics?	Used prep?	Outcome
Brett (outbound)	Partial — no qualifying answers	0 of 3	N/A (pre-framework)	N/A (outbound)	Demo follow-up
Kevin	Yes — good discovery	0 of 3	N/A (pre-framework)	N/A	Soft close
Joseph_1446	No — pitched first	0 of 3	N/A (pre-framework)	N/A	Mixed
JosephB	No — pitched first	0 of 3	No	No	Passed
NicholasE	Yes — good opener, cut short	0.5 of 3	N/A (Product Pitch)	Yes — first time	Interested — follow-up mid-Aug

Progress on NicholasE: The prep was used for the first time (pre-call answers referenced, trajectory opener deployed). One universal probe was partially asked (champion chain). But “we’re early” language was the worst yet (4 instances across Bryan and Alex), and the live install window was missed. Discovery-first discipline improved; probe discipline and messaging discipline did not. The next structural fix is briefing Alex on “what not to say” before joint calls.

Discovery Call Playbook — What's Working, What to Fix