One Researcher, Twenty Pods

The hard part of research was never collecting the signal. It was making sense of it fast enough to matter. You can run the interviews, sit in on the sessions, pull the survey data, read through the support transcripts, and still lose the thread, because by the time you’ve synthesized all of it the decision got made without you in the room.

I support more than twenty product pods at Sprinklr, more or less on my own. The traditional playbook doesn’t survive that ratio. You either go shallow on everything or deep on a lucky few and absent everywhere else. For enterprise CCaaS products, where a bad launch is something a customer actually pays for, neither one holds up. So I rebuilt how I work around a single idea: let AI carry the mechanical weight of synthesis, and put the hours it gives back into the parts of the job that produce truth.

Data is the whisper of a behavior

The whispers always outran my bandwidth. One discovery effort might leave me with hours of recordings, a stack of transcripts, a survey, planning docs, and a notebook of half-finished observations. “I read everything” usually meant “I read what I had time for and trusted my gut on the rest.”

That’s the part AI fixed. I run synthesis and first-pass product thinking through it: pulling themes across transcripts, flagging the places two sources disagree, drafting structure I can then argue with, shaping rough notes into something close to requirements. It reads all of it, every time. It doesn’t get tired on the ninth transcript, and it doesn’t quietly skip the quote that complicates a tidy story.

What I don’t hand off

This is the part that gets glossed over, and it’s the part that counts. AI is in the loop for the labor. It is nowhere near the loop for the judgment.

It didn’t tell me, on Project Saral, that the problem teams were chasing wasn’t really missing data. The data was there. What people couldn’t get was a clear account of what the routing system had decided, and when. That came from sitting with the people doing the work and watching what they reached for and couldn’t find. AI doesn’t read a contact center floor, where the most important thing in the room is often what nobody says out loud. And it doesn’t hold the relationship with a PM who needs to trust my read before they stake a release on it.

So the reframe, the time in the field, and the quality of the call stay with me. The triangulation slog, the rough clickable prototype I need for a fast usability pass, six pods’ worth of synthesis in a week — that’s the machine’s.

It isn’t really about speed

People hear “AI-heavy research” and picture someone buried in a screen, going more remote and more asynchronous. For me it went the other way. Handing off synthesis is the only reason I could afford three onsite contact center deployments, across retail, consumer electronics, and CPG operations, watching forecasting and scheduling and incident response happen live instead of hearing about them on a call.

AI didn’t help me do less research. It let me do the kind I can’t fake, the slow work of turning scattered signal into something a team can actually build against.