# Demo Template The demo is your first sale. It can replace the pitch deck and help you move faster post a discovery call. If the buyer doesn't feel it in the first meeting, you probably won't get a second one. Enterprise buyers testing AI products that underperform are not coming back to try another vendor. They're burned, backlogged, and moving on. This is a repeatable framework for building demos that create conviction. ![[Screenshot 2026-03-13 at 14.01.18.png]] # Step 1: Pick the Painful Workflow **Find the task everyone hates.** Not the one that sounds impressive on a slide. The one that makes the buyer's team groan on a Monday morning. Data entry in SAP. Email chains to approve a purchase order. Manually triaging support tickets at 2am. The more universally dreaded the task, the more visceral the demo lands. The test: describe the current workflow to someone outside the company. If they wince, you've found it. # Step 2: Show the Work, Not the Product Don't walk through features. Don't show a settings page. Show the AI doing the actual work, in real time, on screen. The best demos create a **"lightbulb"** moment. The buyer watches software handle a complex, human task and thinks: "that just replaced three hours of my team's day." Voice agents are particularly effective because the replacement of labor becomes audible. You hear the AI negotiating, answering questions, switching languages. It's not a dashboard. It's doing the job. Procurement agents that collapse a weeks-long purchase request into five minutes. Support agents that resolve tickets with infinite patience in any language. The pattern is the same: compress time dramatically and make the compression visible. # Step 3: Use Their Data, Not Yours Generic demo data kills conviction. The buyer sees sample data and thinks "that's a controlled environment, it won't work with our mess." Seed your demo sandbox with real customer data (anonymized if needed) or synthetically generated data that mirrors their actual environment. Same field names, same edge cases, same formatting quirks. When the buyer sees their own world reflected back, the mental leap from "demo" to "production" shrinks dramatically. If you can't get their data before the first meeting, generate synthetic data that matches their industry. A logistics company should see freight lanes and carrier names, not placeholder text. A procurement team should see their vendor categories and approval hierarchies, not "Acme Corp." # Step 4: Kill the POC Timeline The traditional sequence is: pitch, discovery, POC proposal, security review, POC execution, results analysis, business case, procurement. That's 6-12 months before anyone sees value. Compress it. **The demo IS the proof of concept.** Run it in a sandbox that mirrors production. Show results that would have taken a formal POC to generate. If your first sales meeting contains the proof, the "pilot" stage becomes a formality: paperwork, not validation. This connects directly to [[Deployment Velocity]]. If your deployment hours are still high, your demo-to-production gap will be large regardless of how good the demo is. [[Wright’s Law]] applies: every deployment should compress the next one. If it doesn't, you have a services problem masquerading as a product demo. # Step 5: Quantify the Before and After During the demo, make the math explicit. Don't let the buyer do the ROI calculation later in a spreadsheet. Do it live. > "That purchase request took 4 minutes. Your team currently averages 3 weeks. At 200 requests per quarter, that's X hours returned to your team." > "That support ticket was resolved in 90 seconds. Your current average handle time is 12 minutes. At your ticket volume, that's Y fewer agents needed." Outcome-based framing does two things. It makes the value undeniable in the room. And it sets up [[Selling AI MOC|outcome-based pricing]] naturally: if the buyer can see exactly what the AI delivers, charging per outcome feels obvious rather than risky. # Step 6: Prepare for the Trust Questions The demo creates excitement. The next 10 minutes are where trust gets tested. Come prepared with clear answers: - "Do you train on my data?" No. - "How are prompts and outputs logged?" Here's our audit trail. - "What happens when the AI gets it wrong?" Here's our escalation protocol and human-in-the-loop triggers. - "What compliance certs do you have?" SOC 2, and here's our [[AI Verification]] approach beyond just certificates. Don't wait for these questions. Pre-empt them. The strongest teams show the governance layer as part of the demo, not as a follow-up slide. [[Domain Experts as Eval Builders]] matters here: if you can show eval suites and rubric-scored scenarios during the demo, you're proving both capability and trustworthiness simultaneously. ## The Checklist Before every demo, confirm: - [ ] Workflow chosen is something the buyer's team genuinely hates doing - [ ] Demo shows AI doing the work, not a feature tour - [ ] Data mirrors the buyer's actual environment (real or synthetic) - [ ] Time compression is visible and dramatic (weeks to minutes, hours to seconds) - [ ] ROI math is done live, not left as homework - [ ] Trust answers are pre-loaded, not reactive - [ ] Sandbox mirrors production closely enough that the POC becomes a formality - [ ] The demo can run in under 15 minutes. Longer than that and you've lost the room # Common Mistakes 1. Showing a dashboard instead of doing the work. Dashboards are reporting tools. The buyer wants to see the AI replace labor, not visualize it. 2. Using generic sample data. Kills the "this could work for us" feeling instantly. 3. Saving the best for last. Lead with the most dramatic moment. If the buyer is checking email by minute 8, nothing at minute 15 will save you. 4. Over-explaining the tech. The buyer doesn't care about your model architecture. They care that the purchase order got processed in 4 minutes instead of 3 weeks. 5. Treating the demo and the POC as separate events. In AI sales, they should be the same thing. [[Land-and-Expand in Enterprise AI]] shows that most pilots never expand. If your demo doesn't already contain the proof, you're adding months of timeline and risk. Links: - [[Selling AI MOC]] - [[Land-and-Expand in Enterprise AI]] - [[Deployment Velocity]] - [[AI Verification]] - [[Domain Experts as Eval Builders]] - [[AI-first GTM]] - [[Consultancy-to-Platform Transition]] - [[Team Growth x Product Market Fit]] --- Tags: #deeptech #systems #kp #firstprinciple