Natural language queries on top of a complex analytics platform.
Embedded as Fractional AI Product Partner for 6 months to design and ship an AI copilot that lets non-technical users query their operations data in plain English.
The Challenge
Powerful analytics. Used by 12% of the user base.
Operata had built one of the most sophisticated workforce analytics platforms in the contact center space. Their power users — analysts and operations managers — loved it. But those users represented only 12% of the seats. The other 88% — supervisors, team leads, business stakeholders — never logged in because the query builder was too complex.
The product team knew the answer was some kind of natural language interface. What they didn't know was how to scope it, what models to use, how to handle the reliability problem (what happens when the AI gives a wrong answer), or how to build it without a six-month detour away from the core roadmap.
That's where we came in.
Our Approach
Embedded, not advisory.
We joined Operata's weekly product reviews, architecture discussions, and design critiques. Not as consultants delivering recommendations — as a working partner helping the team ship.
Scoping & Risk Assessment
First 3 weeks: mapped the query patterns of the 12% who were using the platform, identified the 20 most common query types, and defined the reliability floor. The copilot would only handle queries it was confident about — everything else would fall back to the query builder.
Architecture Design
Designed the text-to-SQL pipeline with a verification layer: the LLM generates the query, a validator checks it against the schema, and a confidence scorer decides whether to execute or escalate. Chose a fine-tuned model over a general LLM for the SQL generation step.
Product & UX Design
Designed the copilot interface as an overlay on the existing platform — not a separate page. Users ask a question, see the answer, and optionally see the query it ran. Iterated through 4 rounds of user testing with actual customers.
Ship & Iterate
Launched to 5% of customers in week 14. Monitored query confidence scores, error rates, and session recordings. Shipped 3 improvement cycles based on what we saw. Full rollout in week 22.
Handoff & Team Upskilling
Documented the architecture, prompt engineering decisions, and model fine-tuning process. Ran 4 sessions with Operata's engineering team so they could own the system independently after the engagement ended.
Results
What we delivered.
4×
increase in weekly active users within 60 days of launch
67%
of new queries now handled by the copilot
NPS +31
improvement in product NPS score post-launch
"Having Venkat embedded in our product reviews changed how the whole team thought about AI. We shipped something we'd been talking about for 18 months — in 6 months — and it actually works."
Sarah Okonkwo
VP Product, Operata
Ready to Build?
Let's build your intelligent product.
Book a 30-minute strategy call. We'll listen to your idea, challenge your assumptions, and tell you exactly how we'd approach building it — no sales pitch, no commitment.