AI-Powered Messaging Gap Analysis in Two Hours at Docebo

The drop-off was happening after the demo stage. Our hypothesis was straightforward: we were not telling the market the right message and were not meeting buyer expectations once they reached a demo.

I needed to investigate what this segment expected when they came in, why we won or lost opportunities, whether outcomes changed by persona, company size, or industry, and how our past product reality and future roadmap should shape what we leaned into or avoided. The request landed around noon on a Friday. Pre-AI, gathering win-loss data, NPS trends, call insights, and industry signals alone would have taken days or weeks before any real analysis began.

I started by defining the exact inputs required: win-loss data, NPS scores over time, customer conversation insights, and external industry/voice-of-market signals.

Next I curated and cleaned those inputs so they were usable. I used existing automations and agents to surface six-month NPS sentiment trends for the segment, sliced win-loss interviews by segment/persona/use case, ran natural-language queries across Gong calls for expectations, demo reactions, questions asked, and stage progression, and gave Perplexity Enterprise Pro tight parameters (trusted analysts, influencers, Reddit, Substacks, podcasts) to map industry expectations and gaps.

I spot-checked continuously for hallucinations by following each tool’s audit trail back to source calls, pages, and stats. I then consolidated everything into a single context document and fed it to a custom GPT. Using tight prompting practices, I iteratively queried the data to produce a full messaging audit—how we positioned today versus prospect expectations, discovery pains, and questions—without knowing every answer in advance. The goal was clear recommendations and a format consumable by demand gen and our SDR/BDR teams.

I shipped the output the same afternoon for validation, reviewed it with sales, product, and other stakeholders (only minor tweaks needed), and handed the updated messaging, use-case needs, personas, KPIs, value drivers, and industry variations to demand gen so they could begin refreshing ads and top-of-funnel assets. We kept everything else constant so we could treat it as a natural before/after test and planned to iterate as results arrived.

• Gong for call recordings and conversation data
• Tool removed agents to pull and trend NPS scores automatically surfaced in Slack
• Klue (maintained by our competitive intelligence manager) for structured win-loss interviews sliceable by segment, persona, and use case
• Internal Audience Insights agent (built by demand operations on Unity Apps) for natural-language queries across Gong + Salesforce data, benchmarks, value trees, and testimonials
• Perplexity with a curated parameter doc of trusted analysts, influencers, Reddit communities, Substacks, and podcasts
• Custom GPT (ChatGPT) loaded with the consolidated insights doc and tight system prompts
• Supporting stack pieces: Salesforce segmentation, automated NPS → Slack flows, and later consideration of trackers inside Gong/Clari-style tools as workarounds

I signed off around 2:00 p.m. the same Friday with a complete messaging audit and recommendations ready for my boss’s first review—roughly two hours of elapsed time. He later told me the depth and turnaround far exceeded expectations; the team had assumed the work would take a full week even with our tools.

We surfaced real disconnects: places we were over-positioning on capabilities that were not current strengths (and sometimes competitor differentiators) and under-emphasizing areas where we led and buyers already cared. The deliverable included updated messaging, persona KPIs, value drivers, and vertical-specific framing.

After quick stakeholder validation we began updating ads and top-of-funnel assets. Quantitative pipeline lift is still early, but the process compressed what used to be weeks of manual gathering and analysis into an afternoon, gave us higher confidence in the inputs through systematic spot-checks, and created a repeatable playbook we can run again as results come in—potentially even faster next time.