DEMOPREP_SELLING_STATUS.md

DemoPrep Status Brief

Purpose

DemoPrep is a demo-generation application built to help sales and solution teams create relevant, live analytics demos quickly. Given a company and a use case, the application researches the business context, builds a realistic analytics dataset, deploys it to Snowflake, creates a ThoughtSpot model, and generates a liveboard that can be used in customer conversations.

The goal is simple: reduce the time and friction required to create credible, company-specific demos, while improving demo quality and consistency across the field.

What We Built

DemoPrep now supports an end-to-end workflow from company input to live ThoughtSpot assets:

1. Company and use-case intake Users provide a target company, select a vertical / line / function, or enter a custom business context.

2. Research and context generation The system gathers business context and uses it to shape the demo narrative, metrics, dimensions, and liveboard story.

3. Dataset-first generation We moved away from purely schema-first synthetic data generation. The current approach builds coherent business datasets first, then derives the schema and DDL from that dataset. This has materially improved realism and reduced impossible metric combinations.

4. Snowflake deployment DemoPrep creates the schema and loads populated tables into Snowflake.

5. ThoughtSpot deployment The application creates the connection, tables, joins, model, and liveboard. The liveboard path uses MCP creation followed by TML post-processing for layout and polish.

6. Quality test harness We built an end-to-end Playwright quality suite that submits demo requests, waits for the pipeline, resolves the exact ThoughtSpot model/liveboard, samples Snowflake data, and grades both data quality and liveboard quality with an AI rubric.

Current Status

The application is now consistently completing end-to-end runs in local testing, with meaningful improvements in data quality.

Recent local six-pack result:

Test	Data	Liveboard	Total
EY Professional Services	76	64	79.0 / B
Datadog SaaS Sales	70	73	78.2 / B
Nike Retail Sales	80	73	83.2 / B
Delta Airline Sales	65	74	76.0 / B
Wells Fargo Banking Marketing	78	69	81.2 / B
Starbucks Store Operations	72	58	75.5 / B

Average: 78.9 / B

This is a major improvement from earlier runs where data quality was frequently in the 40s and 50s, with common issues such as returns exceeding sales, load factors over 100%, marketing funnel metrics out of order, and generic data that did not match the target company or use case.

What Improved

The biggest improvement is the move toward a more coherent data-generation approach:

• Dataset-first architecture: We now generate business-realistic datasets before deriving DDL.
• Scenario families: The system maps requests into known scenario families such as SaaS sales, retail sales, banking marketing, airline operations, store operations, sports venue engagement, and professional services.
• Custom use case handling: Custom prompts now influence the dataset structure more strongly. For example, EY now generates professional services data instead of incorrectly producing sports and venue data.
• Better test identity: The test runner now uses page URLs and exact run context instead of schema-prefix guessing.
• More reliable diagnostics: Logging has been improved so run identity, model selection, stage completion, and errors are easier to trace.
• MCP contingency path: If the ThoughtSpot MCP service is unavailable, DemoPrep can create a clearly marked backup liveboard path rather than silently hiding the failure.

Why This Matters Commercially

Demo quality directly affects seller confidence and customer credibility.

Today, high-quality custom demos require manual work from people who understand the account, the industry, the data model, and ThoughtSpot. That creates a bottleneck. DemoPrep reduces that bottleneck by turning a company and business problem into a working analytics experience.

The commercial value is:

• Faster demo creation: Move from days or weeks of prep toward same-day demo generation.
• Better account relevance: Generate demos that reflect the prospect’s company, industry, metrics, and business questions.
• Repeatable quality: Create a structured path for demos instead of relying on ad hoc manual builds.
• Field leverage: Give more sellers and solution engineers access to credible demos without requiring every demo to be hand-built.
• Higher confidence in AI demos: The quality harness creates a feedback loop so we can measure and improve the demos instead of relying on subjective inspection.

Remaining Gaps

The application is not finished, but the core value is now visible.

Current gaps:

• Liveboard quality trails data quality: Data scores are now often B-level, while liveboards still need better narrative layout, stronger KPI selection, and more consistent visual variety.
• Some scenarios are still generic: Certain industry/function combinations need deeper specialization to feel truly customer-specific.
• Custom extraction needs hardening: We improved this significantly, but custom prompts still need more testing across unusual industries and ambiguous company contexts.
• Operational resilience: ThoughtSpot deployment and MCP availability remain external dependencies. We now handle some failures better, but this should continue to be hardened.
• Packaging and ownership: To sell or scale this internally, we need clearer ownership, deployment expectations, support model, and roadmap.

Call To Action

We are asking for alignment on whether DemoPrep should move from a sprint prototype into a supported internal product motion.

Specific asks:

1. Executive sponsorship Confirm that rapid AI-generated demo creation is a priority worth operationalizing.

2. Pilot group Identify a small group of sellers and solution engineers to use DemoPrep on real opportunities and provide structured feedback.

3. Quality bar Agree on what “demo-ready” means. Proposed starting bar: pipeline completion, data score 75+, liveboard score 70+, and no critical business-logic violations.

4. Scenario prioritization Pick the highest-value vertical/function combinations to harden first based on pipeline impact.

5. Production path Decide whether this should remain a specialist tool, become an internal field application, or eventually become a customer-facing/productized capability.

Recommended Follow-Up

After the meeting, the recommended next steps are:

• Run a structured pilot with 5-10 field users.
• Select 10 priority demo scenarios and harden them to a consistent quality bar.
• Improve liveboard generation and post-processing so visual storytelling catches up to dataset quality.
• Add a dashboard for quality trends, failure reasons, scenario coverage, and user feedback.
• Define ownership for deployment, monitoring, support, and roadmap.
• Prepare a short internal demo showing before/after quality improvements and a live end-to-end generation.

Bottom Line

DemoPrep is now showing the shape of a valuable internal sales enablement platform. It can generate complete ThoughtSpot demos from company and use-case input, and recent work has materially improved data quality and pipeline reliability.

The next decision is whether we want to treat this as an experiment, or invest enough to make it a reliable tool for the field.