Lesson 6 of 6 · 50 min

Capstone: discovery → solution → demo

Run the full arc end to end — discovery to a quantified spec, a solution narrative in the customer’s vocabulary, and a hero/problem/resolution demo that re-states the value — then defend it under pressure across the two-round solutions-engineer loop (architecture + discovery/demo).

The whole job, in one arc

Everything in this track composes into a single three-act arc: discovery produces a quantified spec, the solution narrative maps required capabilities to that spec in the customer’s vocabulary, and the demo delivers the story and closes by re-stating the quantified value from act one. The three acts share one artefact — a one-page Storyboard: Hero / Problem / Required Capabilities / Product Story / Confirmation of Value. This capstone runs the arc end to end and rehearses the two-round solutions-engineer loop that tests it.

The typical AI-SE loop stacks the rounds you’ve trained: a 30-minute scenario scoping (translate a vague use case into a spec), a 60–90 minute architecture + live build, and a 30–45 minute panel demo — and the unspoken bar is the handoff between rounds, not any one round in isolation. The documented FDE-AI loop is a 30-minute scenario → 90-minute live build (with an AI coding assistant) → 60-minute stakeholder presentation. Rehearse the seams: the spec you produce in discovery is the input to the architecture, and the architecture is what the demo dramatizes.

The Customer Demo — When Should We Do It, and How to Build ItWe The Sales Engineers (Ramzi Marjaba)

Act 1 — discovery to a quantified spec

Open with use-case selection, not a platform pitch: “Which business decision do you wish you could automate today, and how would you know it was working?” Run the three-beat discovery (outcome/stakeholders → today-workflow/pain → constraints/risk), and use Implication questions to put a number on the board in the customer’s own hand. Walk the Geodesic five pains explicitly — data readiness first, because it is the call-one filter that prevents a dead PoC. Leave the call with two artefacts: the AE’s MEDDPICC qualification snapshot and your process map (people / process / technology / data / integration / security). The quantified pain and the process map are the spec.

code

1THE ONE-PAGE STORYBOARD (shared across all three acts)23  HERO ................. the prospect's role (a developer, an ops lead) -- NOT the product4  PROBLEM .............. the pain, QUANTIFIED in the customer's words (e.g. 8-12 hrs/wk)5  REQUIRED CAPABILITIES  the 3 capabilities that resolve the problem (customer's nouns)6  PRODUCT STORY ........ the demo: each capability answers the quantified pain7  CONFIRMATION OF VALUE  re-state the From->To->Gap; 200-600% range, customer's numbers89  Build it BEFORE the demo; review it with the AE. Sandwich the solution10  between the hero's problem and the resolution.

Build the one-page Storyboard during discovery, not after, and review it with the AE before the demo — the practitioner rule is that process discovery is “sharpening the axe,” where every ten minutes invested saves thirty minutes of demo rework. The Storyboard’s five fields force the discipline: if you can’t fill Problem with a number the customer said, you haven’t run discovery; if you can’t name three Required Capabilities in the customer’s nouns, you’re about to demo a tour; and if Confirmation of Value isn’t a From→To→Gap the champion can defend, the deal dies in committee. The two artefacts you carry out of every call — the AE’s MEDDPICC snapshot and your process map — are the raw material the Storyboard is assembled from.

Interview angle. The 30-minute scenario round scores whether you ask scoping questions on end users, latency, auditability, and regulatory exposure before drawing a box — and whether you resist proposing an immediate AI solution. Candidates who jump straight into architecture fare worse than those who clarify; candidates who invent fake systems and APIs (because the case gave none) fare worst. The strong move under ambiguity is to walk the buyer through their own process to surface the missing constraints.

Act 2 — the solution narrative, in the customer’s vocabulary

Transition from problem to design with named Required Capabilities and Positive Business Outcomes (Command of the Message), keeping the same nouns and KPIs the customer used. This is where the architecture from Lessons 1–4 earns its keep: you propose the walking skeleton (gateway → orchestrator → retrieval → tools → evals/guardrails → observability), state the SLOs, and — critically — answer the five AI pains the discovery surfaced. Data not ready? Show ingestion + cleansing on a sample. Legacy systems? Show the MCP connectors / API coverage of the named systems. Agent security? Show identity controls, the allow-list, and the audit log. ROI unclear? Show the eval dashboard, guardrails, and rollback. AI demos fail when you show capabilities the customer never named, so the architecture narrative must be answer-only, not a tour.

Have the architecture answers ready for the live-build and defense, because the panel will probe them: the eval-the-AI question (“how do you know it’s working?” → golden set + online sampling + calibrated judge + drift dashboard, Lesson 2), the observability question (“how do you find why it’s slow?” → TTFT span split, Lesson 3), the connector-security question (“how do you contain a destructive tool call?” → consent + allow-list + IdP + audit, Lesson 4), and the freshness/multi-tenancy questions (Lesson 1). Each is a 30-second answer with a number or a named mechanism.

The architecture round and the demo round are the same story told twice — once as a system of seven layers to the technical panel, once as a hero’s resolution to the business stakeholder. The senior candidate is the one whose seven layers and whose three demo capabilities are visibly the same solution, anchored to the same number.

Act 3 — the demo: hero / problem / resolution

Structure the demo as hero / problem / resolution: the hero is the prospect (a developer, an ops lead — never the vendor or product), the problem is the pain quantified in act one, and the resolution sandwiches the solution between them. Deliver it with the seven demo disciplines: understand the audience; tell a story tailored to them; lead with pain, not features; keep it focused (one message, three capabilities, ~ten minutes); show relevant use cases; use the prospect’s language; and make it interactive (let them drive part of it). The panel demo is 30–45 minutes but the slides/context cap is 5–7 minutes — a 25-slide deck shown in 7 minutes beats a 7-slide deck shown in 25; use the rest of the time for live Q&A and to re-anchor on ROI.

code

1THE DEMO ROUND (30-45 min) -- where the time goes23  Intro / agenda .................... 1-2 min4  Customer overview (their pains) ... 2-3 min   <- their words, quantified5  Why your product + what it is ..... 2-3 min   <- 5-7 min of slides TOTAL6  The demo (hero/problem/resolution)  ~10 min   <- 3 capabilities, interactive7  Live Q&A + re-anchor on ROI ....... remainder89  Per feature, pre-write: "this saves/makes you $X / Y hours."10  Tough question? "Here's what I heard -- correct?" then answer or "I'll verify."11  A 25-slide deck shown in 7 min beats a 7-slide deck shown in 25.

Communicate need over features, and attach every capability to a number. The demo-round rubric hinges on whether you drive buying logic through business outcomes — “you have 70% retention; a 1-point lift is $2M revenue” — rather than industry buzzwords or product acronyms. Pre-write the ROI sentence for every demo moment. And close the circle: the final beat re-states the From→To→Gap and the 200–600% range from act one, in the champion’s language, so they can re-tell it. Interview angle. For every tough question, the strong move is to listen, summarize it back (“here’s what I heard — is that correct?”), then answer or honestly say “I need to verify that against our docs” — bulldozing objections or inventing facts is the weak tell.

Putting it together: the capstone scenario

Run the whole arc on one scenario: “A logistics firm wants an AI agent to handle automated shipment rerouting.” Act 1 — discovery: who reroutes today and how (hero = the ops lead), what a missed reroute costs per quarter (quantify the risk lever), which systems the process touches (TMS, ERP, carrier APIs), what data exists and at what quality (readiness filter), and who approves an automated reroute (agent-security pain). Act 2 — solution: the walking skeleton with CDC ingestion from the TMS for freshness, MCP connectors to the carrier APIs behind a gateway with an allow-list and audit log, an eval harness with a golden set of reroute decisions and a drift monitor, and observability with TTFT spans. Act 3 — demo: hero/problem/resolution on three capabilities (detect → propose reroute → execute with approval), each tied to the quarter-cost number, closing on a 200–600% ROI range in the ops lead’s words.

code

1THE STORYBOARD, FILLED IN (logistics rerouting)23  HERO ............... the ops lead who reroutes shipments by hand today4  PROBLEM ............ "~6 missed reroutes/quarter x $40k each = ~$960k/yr"  (their number)5  REQUIRED CAPS ...... 1) detect at-risk shipment   (CDC from TMS, retrieval)6                       2) propose a reroute          (model + carrier-API tools via MCP)7                       3) execute WITH approval      (human-in-the-loop + audit log)8  PRODUCT STORY ...... demo the 3 caps; each answers the $960k pain, not a tour9  CONFIRMATION ....... From $960k loss -> To ~$160k -> Gap ~$800k; 3-yr ROI 250-400%1011  Architecture (panel) and demo (stakeholder) are the SAME solution, one number.

Notice how every lesson shows up exactly once: the architecture (L1), the eval harness (L2), the observability (L3), the MCP/connector governance (L4), and the discovery + value story (L5) — composed into one defensible arc. That composition is the senior solutions-engineer signal: not depth in one round, but a coherent handoff from a quantified pain to an SLO-anchored architecture to a number-closing demo. Interview angle. Map each rubric category to one story you personally owned — an architecture you drew, a discovery you ran, a demo you delivered, an eval you owned — each with a ROI claim and a “what broke and how I detected it” coda; that satisfies the behavioral 5-point scale, the technical 1–4 scale, and the AI-specific overlay simultaneously.

Implementing MEDDIC — MEDDPICC Explained in 10 MinutesMEDDICC article7 Sales Demo Best Practices, from a Sales Engineer (Dan Pecher, Dialpad)Dialpad articleForward Deployed Engineer Interview: The Definitive 2026 GuideExponent articleStorytelling in Demos: The Hero’s Journey in 60 Seconds (pain → resolution)Great Demo!article5 Demo Tips From the Best in the Business (customer-as-hero, the “critical few”)PreSales Collective

Checkpoint

You have a 30–45 minute panel demo for the logistics-rerouting scenario. How do you allocate the time?

ASpend most of it walking every feature so they see the full productB5–7 minutes of slides/context, ~10 minutes of a focused hero/problem/resolution demo on three capabilities, then live Q&A and re-anchor on the quarter-cost ROICSkip slides entirely and improvise the demo to seem confident

Checkpoint

During the demo a stakeholder asks a sharp question about an AI limitation you’re not 100% sure about. Best response?

AConfidently give your best guess so you don’t look unpreparedBListen, summarize it back (“here’s what I heard — correct?”), answer what you can, and say “I’ll verify that against our docs,” then re-anchor on the quantified valueCPivot to another feature to change the subject

Checkpoint

In Act 2 you’re tempted to showcase the agent’s impressive multi-tool planning, which the customer never raised. What’s the senior call?

ACut it — the solution narrative must be answer-only, mapping each capability to a pain the customer quantifiedBShow it — more demonstrated capability always strengthens the caseCShow it but only briefly at the end

Checkpoint

The panel asks, mid-architecture, “how would you know this rerouting agent is actually working in production?” Strongest answer?

AWe’d collect user thumbs-up/down on each reroute and review the negativesBA golden set of reroute decisions gated in CI, online sampling with a calibrated judge, and a drift dashboard tracking quality, latency, and cost — plus rollbackCWe benchmark the model on public datasets before launch

Checkpoint

Across the loop (scenario → architecture/build → demo), what most distinguishes a strong candidate?

AMaximum depth in the architecture round, treating the others as warm-upsBA coherent handoff — discovery’s quantified spec feeds an SLO-anchored architecture that the demo dramatizes and closes on the customer’s numberCUsing the most advanced AI techniques in every round to show expertise

Interview prep

The capstone maps to the full solutions-engineer loop: a scenario-scoping round, an architecture/live-build round, and a panel demo — graded on a behavioral 5-point scale (problem solving, awareness, collaboration, perspective), a technical 1–4 scale (algorithms, coding, communication, problem-solving), and an AI-specific overlay (RAG depth, agent orchestration, LLM eval, OWASP LLM Top 10). The through-line: scope before solving, anchor on numbers, and own your contributions (“I,” not “we”).

01“Run the arc from a one-line use case.” → discovery to a quantified spec → solution narrative in the customer’s vocabulary → hero/problem/resolution demo that re-states the value.
02“What’s the shared artefact?” → a one-page Storyboard: Hero / Problem / Required Capabilities / Product Story / Confirmation of Value.
03“How long are the slides in the demo round?” → 5–7 minutes; a 25-slide deck shown in 7 min beats a 7-slide deck shown in 25; spend the rest on Q&A + ROI.
04“How do you handle a tough demo question?” → summarize it back, answer or say “I’ll verify,” then re-anchor on the customer’s number — never bulldoze or fabricate.
05“How do you know the AI works?” → golden set + online sampling + calibrated judge + drift dashboard + rollback (Lesson 2), tied to the scenario’s decisions.
06“What separates strong across the loop?” → the handoff between rounds — a coherent arc, not depth in one round.
07“Why this company / role?” → connect personal motivation to customer-facing work and a specific product decision; articulate individual contributions (“I”).
08“Defend your design under pressure.” → strip the complexity, acknowledge the AI limitation honestly, re-anchor on the quantified value.

Going deeper, the cross-round probes are where loops are won or lost: “the case gave no systems or constraints — what do you do?” (ask scoping questions and walk the buyer through their process; never invent fake APIs); “the eval round — how do you know your golden set represents production?” (harvest from the production retriever’s real runs); and “map a rubric category to a story you owned.” Prepare four stories — an architecture you drew, a discovery you ran, a demo you delivered, an eval you owned — each with a ROI claim and a “what broke and how I detected it” coda, and rehearse them out loud. That single preparation satisfies all three scoring scales at once.

Could you run the full discovery → solution → demo arc on a fresh scenario and defend it across the scenario, architecture, and demo rounds?

New to itGetting thereConfident

Takeaways

The job is one arc: discovery → solution narrative → demo, sharing a one-page Storyboard.
Discovery produces a quantified spec; the solution retells the architecture as the customer’s story.
Demo slides cap at 5–7 minutes; attach a dollar/hour figure to every capability and re-state the value.
Defend under pressure by stripping complexity and re-anchoring on the customer’s number, not by bulldozing.
The senior signal is the handoff between rounds — a coherent arc, not depth in one round.
Prep four owned stories (architecture, discovery, demo, eval), each with a number and a failure-mode coda.

You’ve run the full arc — pair with a peer for a live mock loop, then take it into a real engagement.

Sources

Free to read · better with Enzo

Learn it with Enzo

Save your progress, answer the checkpoints, and let Enzo quiz you on what you just read.

Continue with Enzo Sign up free