B2B SaaS
Gave product managers a portal to ask anything of their customer data
A mid-market technology company’s customer feedback was everywhere and nowhere at once. We unified 30+ sources into one searchable store and made it answerable in plain English, with the evidence attached.
- Sector
- Mid-market technology · B2B SaaS
- Engagement
- Signal discovery → AI feedback platform → product enablement
- Built with
- n8n · Pgvector · AWS Bedrock (Anthropic’s Claude models) · evidence-cited answers
2,000+
Real customer signals behind each roadmap decision
30+
Feedback sources unified into one searchable store
The challenge
A mid-market technology company’s customer feedback wasn’t missing, it was everywhere and nowhere at once.
It lived in sales and customer calls, in public reviews and forums, in support tickets, surveys and internal systems. No single person could hold it all in their head, and no report pulled it together. So the product team did what most teams do: prioritised on instinct, recent memory, and whichever customer had complained most loudly that week. The signal to build the right thing existed, it was just impossible to act on.
What we did
We scoped the signal that actually mattered
We started by mapping where useful feedback actually lived, and settled on around 30 distinct sources spanning three worlds: conversations (sales and customer calls), public channels (reviews, forums and the open web), and internal systems (support tickets, surveys and product data). Deciding what to listen to, and what to ignore, came before any build.
We unified everything into one searchable store
We built a pipeline in n8n to pull from all thirty sources and load them into a single vector store on Pgvector, so every comment, ticket, call and review lived in one place, searchable by meaning rather than keyword. For the first time, the company’s entire body of customer feedback had one home.
We made it answerable in plain English
Using Anthropic’s Claude models on AWS Bedrock, we built a portal where a product manager can ask a question in plain English, “what’s frustrating enterprise customers about onboarding?”, and get a synthesised answer drawn from across all the feedback, with the underlying evidence attached. No tagging, no spreadsheets, no trawling through call recordings.
We turned raw feedback into Jobs To Be Done
Beyond answering one-off questions, the system produced ongoing Jobs To Be Done (JTBD) analysis, surfacing the jobs customers are really trying to get done, ranked by how often and how strongly they showed up across the data. The roadmap stopped being a record of who shouted loudest and started reflecting what customers actually needed.
Every answer came with its evidence
Because each answer cites the source feedback behind it, product managers could trust what came back, and defend it in a roadmap review. The system didn’t ask anyone to take its word for it; it showed its working.
The outcome
- Product managers query the company’s entire body of customer feedback in plain English, and get a synthesised, evidence-backed answer in seconds.
- Data from over 30 sources (calls, public channels and internal systems) unified into one searchable store.
- Ongoing JTBD analysis ranking the jobs customers are really trying to get done.
- Hundreds of hours of manual feedback trawling saved per quarter.
- Roadmap decisions backed by over 2,000 real signals instead of a handful of loud accounts.
Why it worked
The value was never a smarter dashboard. It was closing the gap between the feedback a company already had and the decisions it was trying to make. The feedback was never the problem, acting on it was. Now the answer is one plain-English question away, with the evidence to back it.
Sitting on feedback you can’t act on?
Start with a fixed-scope AI Opportunity & Risk Audit.