Demo Flow

In this module

1. The 25-minute demo arc
2. Setup (2 min) — the opening that sets up everything
3. Act 1 — The Loop (8 min) — VOC → Idea → Brief → outcomeHook
4. Act 2 — The Execution (8 min) — agent dispatch + verification + done-flip
5. Act 3 — The Attribution (5 min) — predicted vs. realized + recalibration
6. Close (2 min) — back to the differentiator + the close
7. Tuning the demo to the persona
8. The research foundation — Gong on demo dynamics, Dunford on demo content
9. Anti-patterns to avoid
10. Discussion prompts

The 25-minute demo arc

A demo is theater with a thesis. The thesis is the closed-loop differentiator. Everything you show must serve the thesis. The 25-minute arc:

The 25-minute budget is non-negotiable. Demos that run 35-40 minutes leave the buyer cognitively saturated and less able to engage in the close. If the buyer wants more, you've earned the next meeting — don't blow the budget by extending in real time.

Gong's research on 326K+ calls found winning demo calls run at a 65:35 talk-to-listen ratio (you talk more once buyer wants to see it work). That's different from discovery, where winning calls are 46:54. Internalize the shift — you're allowed to drive the demo, but check in with buyer reactions at the act boundaries.

Setup (2 min)

The opening sets up everything. If you blow the setup, you spend the rest of the demo trying to recover.

Open slide 01. Engineer-track 01-pm33-strategic-execution-platform diagram.

State the differentiator. "PM33 is a closed-loop strategic execution platform. Other AI dev tools optimize for shipping speed. We optimize for outcome attribution. The differentiator is the loop back from shipped work to strategic metric movement."

Pause. Wait for the buyer to nod or say "OK." If they look confused, take 30 seconds to re-frame — but don't move on until they've signaled "got it."

The pause is the most important part of the setup. Most reps barrel through it. Don't. The buyer needs to be oriented before the demo will land.

Act 1 — The Loop (8 min)

This is where you sell the closed loop. The act has three beats.

Beat 1 — The morning state (2 min)

Live demo: open Pam's morning summary on staging. Show the overnight summary block — what shipped, what's blocked, what needs the PM's attention. Reference the PM-track day-in-the-life view (overview-day-with-pm33).

The buyer reaction you're looking for: "wait, this is what they see every morning?" That's the wedge.

Beat 2 — VOC triage (2 min)

Show an inbound VOC item being triaged. Walk through how Pam scored it against active strategic objectives, clustered it with similar items, and drafted an Idea (or surfaced a candidate Brief).

Key teaching point: the PM is not reading every VOC item. They're confirming or correcting Pam's judgment on the top 5-10. The triage layer is where 80% of the PM's old VOC-reading time goes away.

Beat 3 — Brief authoring (4 min)

Promote the Idea to a Brief. Walk through the structured fields:

• Title in imperative form
• Machine-verifiable acceptance criteria
• outcomeHook — predicted metric movement (this is where you sell the closed loop)
• Specialist + LLM tier assignment
• TDD phases

The outcomeHook field is the moment the demo earns the closed-loop framing. Spend extra time here. Show the field. Read it aloud. Connect it to the strategic objective the buyer cares about.

If the buyer is technical, dive deeper on the Brief schema (function signature, input/output contracts, dependencies). If non-technical, stay at the conceptual level.

Act 2 — The Execution (8 min)

This is where you sell the agent governance. The act has three beats.

Beat 1 — In-flight state (2 min)

Show a Brief that's in_progress. Agent dispatched. Per-agent worktree shown (if buyer is technical). Lifecycle events streaming to the audit log.

Key teaching point: the agent is working in isolation. Per-agent worktrees prevent the absorption-class bugs covered in PM33's CLAUDE.md. This is the architectural answer to "what if AI agents step on each other?"

Beat 2 — Verification + done-flip (3 min)

Show a Brief auto-flip from in_progress to done. Walk through what fires: CI passed, schema-drift gate passed, security checks passed, independent code review approved, PR merged, lifecycle event pr_merged fired, AutoDoneVerificationService computes done = true.

Key teaching point: done is computed, not claimed. No agent or human marks the Brief done unilaterally. Done emerges from verification signals. This is the closed-loop pattern's structural defense against the Perry et al. 2023 finding that AI-assisted users write less secure code AND are more confident in its security.

Beat 3 — outcomeHook window opens (3 min)

Show the outcomeHook measurement window opening. The Brief shipped at T0. The configured measurement window (e.g., 14 days) opens. Metrics start accumulating against the predicted impact.

Key teaching point: the loop is open until the measurement closes. A shipped Brief isn't done from an attribution perspective until the outcome data lands. This is the bridge to Act 3.

Act 3 — The Attribution (5 min)

This is where you sell the compounding advantage. The act has two beats.

Beat 1 — The OAR view (3 min)

Open the strategic objective dashboard. Show the OAR (Outcome Attribution Report) — predicted vs. realized for shipped Briefs in this objective. Reference exec-02-oar-mockup.excalidraw.

Walk through the four sections:

• Section 1: objective status (current vs. target, AR(1) forecast, confidence interval)
• Section 2: top contributing Briefs (predicted vs. realized, with band assessment)
• Section 3: misses & surprises with diagnostics (READ THIS ONE FIRST — the misses are the learning)
• Section 4: recalibration history (workspace prior over time)

The OAR is the executive surface. If you have an exec on the call, this is the slide they'll remember.

Beat 2 — Drift alert + AR(1) recalibration (2 min)

Show a drift alert (or simulate one). Walk through what triggers it (forecast trajectory diverging from target by more than the CI allows) and what the system recommends (capacity-aware scheduler proposal for additional Briefs to close the gap).

Show the AR(1) prior recalibrating. Q1 prior μ = 0.10 → Q2 prior μ = 0.07 (recalibrated down based on Q2 misses). Variance decreased 18% over 2 quarters. The compounding advantage is the workspace prior beating industry default by 15-30% after ~100 Briefs.

Key teaching point: the lock-in is the workspace prior, not the vendor. This is the long-game compounding case (covered in Executive Module 5).

Close (2 min)

Back to slide 01. Repeat the differentiator verbatim from the setup. The buyer has now seen 23 minutes of demo; the closing repetition crystallizes what they should remember.

Ask the specific next-step question. "Where in your org would this start?" or "What's the strategic objective you'd want this pointed at first?" These are concrete questions that surface the pilot scope.

Confirm next step. "Great — let's schedule a discovery call with your CISO next week, then I'll draft a pilot scope based on what you said. Sound good?"

The close has to be specific. "Let me know what you think" is the loss signal. "Here's what happens next, here's when, here's who" is the win signal.

Tuning the demo to the persona

The 25-minute structure stays constant. The emphasis shifts by buyer:

A demo with 2-3 personas in the room is harder than a single-persona demo. Tactic: do the demo at the least technical persona's level, but pause at act boundaries to ask the more technical persona "want me to go deeper on this?" Lets them drive the depth without losing the rest of the room.

The research foundation

Gong demo-dynamics research (primary)

Gong's analysis of 67,000+ demos:

• Unsuccessful demos contain monologues up to 106 seconds uninterrupted. Successful demos have frequent buyer chime-ins.
• Talk-to-listen ratio for winning demos: 65:35. Different from discovery (46:54) — you're allowed to drive once they want to see it work.
• Implication: the 25-minute arc above includes natural pause points at act boundaries. Use them. Ask "any questions on that section?" before moving on.

Gong demo-to-close research: industry-average demo-to-close rates are sobering. The deals that close are the ones where the buyer left the demo able to explain the differentiator in their own words. That's why the closing repetition of slide 01 matters.

Dunford on demo content (practitioner-canonical)

April Dunford, Sales Pitch (2023): demos should "demonstrate why your differentiated value matters" rather than walking through features. The Act 1 → Act 2 → Act 3 structure above is Dunford's "Insight → Alternatives → Differentiated Value → Proof" pattern applied to live software.

The "I have to see it work" buyer reaction

The post-2024 AI-product buyer is more skeptical than the 2019 SaaS buyer. They've seen too many AI demos that didn't work in their environment. The structural defense: demo on staging with real data, not on a marketing-canned scenario. PM33's staging is the canonical demo environment for this reason.

Honesty flag: the "AI buyers must see it work" claim is practitioner consensus across 2024-2026 GTM writing (a16z, First Round, Lenny's newsletter, etc.) but I couldn't find a single peer-reviewed primary study to cite. Treat it as field wisdom, not research.

Anti-patterns to avoid

1. Skipping discovery — demoing cold. Re-confirm 2-3 specific pain points before the demo starts. "Last time we talked, you said X, Y, Z were the issues. The demo is going to show how PM33 addresses each. OK?"
2. Demoing every feature. Pick the 3-5 things that matter to THIS buyer. The other 50 features earn the next demo.
3. Letting the demo wander. The closed-loop narrative is the through-line. If the buyer pulls you into a feature tangent, acknowledge the question and bring it back: "Great point — happy to demo that next time. Right now I want to make sure you see how the loop closes."
4. Reading the slide aloud. They can read. Your job is to narrate the why behind what's on screen.
5. Skipping the close. A 25-minute demo without a 2-minute close is a demo that loses to "let me think about it." Always close with a specific next step.
6. Over-engineering the technical depth. A non-technical exec doesn't want to see the per-agent worktree config. Save that for the CTO's separate technical-deep-dive session.

Sidebar — how PM33's demo environment supports the flow

Staging at [staging URL] is intentionally pre-loaded with:

• A realistic strategic objective ("Reduce TTFCV by 30% by EOQ") with active Briefs
• A VOC triage queue with diverse signals
• An OAR with mid-quarter drift alerts visible
• AR(1) recalibration history spanning 2 quarters

You shouldn't have to create demo data live. The environment is designed to support the 25-minute arc end-to-end. If you discover gaps (e.g., "I needed to show X but the demo data doesn't have a good example"), file a Brief to PM33 — the demo-readiness of staging is owned by the GTM team.

Discussion prompts

For team practice:

1. The 25-minute test: do the full demo on staging, recording your screen + audio. Stop at exactly 25:00. Did you finish? If not, where did you over-run? Common culprits: Act 1 Beat 3 (Brief authoring) and Act 3 Beat 1 (OAR walkthrough).
2. The pause-point test: review your demo recording. Count the pause-for-buyer-reaction moments. Aim for 5-7 across the 25 minutes. If you have 0-2, you're monologuing.
3. The persona-shift test: a peer plays a different persona than the previous practice run. Do the same demo. Did the emphasis shift correctly? Or did you do the same demo regardless?
4. The close-line test: write 3 different closing CTAs tuned to different deal stages (first demo, second demo, late-stage). Practice each.

Further reading

• Module 4 — Objection Handling — what to do when the demo surfaces objections
• Module 5 — Pricing Conversation — what to do after the demo lands well
• Engineer-track Modules 1-4 — technical depth for the demo content
• Executive Module 2 — Reading the OAR — what the OAR section of Act 3 means to an exec
• Gong demo research — the 67K-demo dataset + 65:35 talk-ratio finding
• References: ../product-manager/references.md