Case study

AI tools were everywhere. Velocity wasn't moving. The gap between the two is where this case study lives.

I joined a SaaS company that had hired me to answer that question. The board's ask was layered, but the one metric that mattered was time to value: how long it took a new customer to produce their first piece of work in our platform after signing. The number was ~3.5 months. They wanted ~3 weeks. We had the team. We had the tools. The operating model was the gap.

Over the next twelve months, we closed it. Time to value landed at ~3 weeks. Release velocity moved ~40%. Quality moved ~25%. Below is what changed, how we measured it, and what I'd run on any product team this week.

Situation

The mandate was layered, and pointed at one outcome. The board wanted three things at once: ship more features, faster, with higher quality; put AI in the product, and in how we built the product; and move our key customer metric, time to value, from ~3.5 months to ~3 weeks. The first two were the path. The third was the destination.

When I arrived, the product team was four PMs, three product designers, and eight engineers. We were pre-AI. Some of the team were using LLMs as personal chat or search. None of it was in the product. None of it was in how we worked.

The harder problem was upstream of the tooling. PMs told me they didn't have time to talk to customers or run discovery. PRDs took weeks to write. Features shipped and landed flat. We didn't have real product trios (PM, designer, tech lead working together as one). Just PMs handing specs over the wall.

The one thing that would have changed the curve: we were spending all our energy on the back of the product development lifecycle (PDLC), building and shipping, while the front (customer interaction, problem discovery, solution validation) was the actual constraint. Bolting AI on top of broken discovery would have produced more features customers didn't want.

The stakes were not abstract. Board confidence was thinning. Customer churn was climbing. We had to move.

What changed

Workflow restructuring

I picked three stages of the PDLC to redesign first: discovery, metrics synthesis, and requirements writing. All three were front-of-PDLC work, the actual constraint.

The clearest before/after was discovery synthesis. In month 1, pulling together what we knew about a problem (customer interviews, product feedback, support tickets, win/loss data, churn signals, market and competitive intel) took weeks to months. By month 6, a PM ran the same synthesis in a couple of hours with AI assistance, then walked into the trio with a sharper problem definition than we'd ever had.

The trigger was not a top-down mandate. It was a single experiment with one product trio. One PM, one designer, one tech lead, willing to try the workflow change for a sprint. When it worked, the rest of the team noticed.

Measurement

The first metric I put in place was customer interactions per week per PM. We hadn't measured it before. Customer interactions weren't a regular part of the job. They happened ad-hoc, mostly when an Account Executive asked if a PM could join a call.

Counting interactions worked for about six weeks, then I caught myself measuring the wrong thing. The number was an output. Time to value was the outcome the board cared about, and that's what we rebuilt the measurement around. What that looked like in practice: a customer-cohort view that tracked the gap from signature to first piece of work, refreshed weekly.

Defensibility

Two AI features the team shipped were not easy to copy. The first was AI-assisted content generation tuned to our customers' subject matter, where the real time savings landed. The second was adaptive paths through the product that personalized in real time to each user's behavior.

What made them defensible was platform extensibility. Customers could bring their own LLM model. Configuration templates lived inside our system, not theirs. The features stopped being commodity the moment customers wired their own model and config into the platform.

Results

Time to value was a rolling average across the customer base over those twelve months. The number moved from ~3.5 months pre-AI to ~3 weeks by the end.

Release velocity, measured as releases per quarter, moved ~40% comparing Q1 baseline to Q4 result. Quality, measured by customer-reported issues per release, moved ~25% over the same period. I'm keeping the absolute numbers private out of respect for the company. The deltas are what mattered to the board, and what would matter to yours.

A fair caveat: AI was not the only thing that changed. We made team changes (the trio model itself was a structural shift) and process changes that had nothing to do with AI. Some belongs to those moves. The AI work was the load-bearing piece, but it wasn't the only piece.

The surprise was on the GTM side. Sales, customer success, and marketing began trusting product more, and the product story they carried internally and externally got sharper. We hadn't optimized for that effect. It became one of the loudest signals that the change was real.

What's still imperfect: we haven't cracked AI-assisted technical writing. PRD writing got faster. The technical specs engineering uses to build still take real human time, and the AI tools we've tried have not closed that gap.

Self-assessment

Three questions worth running on your own team this week:

  1. 1. Workflow vs. tools. In the last six months, did you change how product gets built, or just which tools you use to build it?
  2. 2. Measurement. Can you point to a number (release velocity, defect rate, time-to-PRD, anything) that has moved because of AI in the last six months?
  3. 3. Defensibility. If a customer asked which 2 to 3 AI features in your product they couldn't get from a competitor, could your team answer in under a minute?

If you want the full diagnostic with scoring guidance, take it at getproductlabs.com/resources.

If you recognized your team in this story, this is the work I do. I help product orgs move from adding AI to being AI-native, starting at the front of the PDLC. If this is your team's spot, I'm taking 2–3 new engagements this quarter. Let's talk.