How Welch’s three‑person insights team runs seven projects at once

Christian Lavoie runs consumer insights for all of Welch’s with a team of three. At Groceryshop 2026, he joined Merciv CEO Shaia Erlbaum and Modern Retail’s Gabriela Barkho on stage to walk through how that team now carries more requests, turns them around in days instead of weeks, and keeps several analyses moving at the same time.

The most useful part of the conversation had little to do with the advanced features most vendors lead with. What changed Christian’s week was a platform that handled the slow, unglamorous work accurately enough that he could stop checking it and spend his time on the story the data was telling.

Three people serving an entire company

Christian has spent about 30 years in marketing research, half of it on the research partner side and half in corporate roles, the last three at Welch’s. His insights team is three people with one open role, and it supports everyone: brand teams managing and growing the core business, a transformational innovation team working outside the category, licensing, category management, and sales whenever a customer meeting or sell-in comes up.

That load was workable only through tight prioritization, and new leadership then raised the bar on speed. Requests that once could wait a week now had to land quickly so the business could keep moving.

“We got to this tipping point that we recognized that hiring another person wouldn’t do it. We needed to completely reinvent the way we work.”

The first tools fell short within a week

The team started with the enterprise AI tools already available to a company of Welch’s size. Christian expected the evaluation to take a month or two. Within about a week, he could see none of them would meet the team’s needs.

They struggled with the volume of data the team loaded in, and detailed analyses across sources came back with gaps and errors. The slides built for leadership needed so much correction that the shortcut stopped being one.

“I almost spent more time fixing what was not right than if I would have just done it myself.”

That was the point at which Welch’s decided to invest in a more capable platform.

The basics changed the daily work

Shaia noted that most vendor presentations lead with their flashiest feature, while the work that actually slows insights teams down tends to be menial: cleaning up a deck, tightening a survey, reworking a focus group workflow. Christian agreed. The market talk centers on digital twins and in-platform concept testing, but the change he wanted was simpler.

“Just getting the basics right will completely change my daily life.”

For him, the basics start with three things.

• A library holding everything the team has ever done, searchable in minutes. Before, finding a past study meant digging through folders and personal hard drives, or emailing someone who left a year ago.
• Analysis he can trust without rechecking every line, since the earlier tools made assumptions he then had to catch.
• A slide he can present as it stands, rather than one that takes two hours to fix.

In practice that means asking the platform to read five past studies and return a single page he can take to the CEO with what he found and what he recommends. Welch’s has not yet turned to the more advanced capabilities. The priority is getting the core right, rolling it out to more people across the company, and adding features once that foundation is solid.

General AI arrives as a brilliant stranger

Shaia described most general-purpose AI tools as brilliant strangers. Given the right data and a narrow enough question, they do excellent work. What they lack is the history behind the question: the methodology a team refined over years, the survey it ran last year and ten years ago, and what it learned in between about how its research tracks with point-of-sale data or assortment planning.

A team like Christian’s sits on decades of that knowledge, and a frontier model starts with none of it. Shown a survey, a general model can comment sensibly on its programming logic, but without your past methodology and surveys in view, Shaia said:

“It’s blind to the nuance that makes your team so expert on your own knowledge, your own topics.”

Merciv is built to work from that full body of research and connect it to everything else the organization knows.

Testing it against work they already trusted

The team proved the platform out against its own record. It loaded about ten past reports whose conclusions had been reached the traditional way and checked whether the platform arrived at the same place. It did.

The bigger test changed how Christian works with research partners. The old cycle ran from a partner’s report through rounds of revisions and calls, then five more revisions, and finally an executive summary he wrote himself. This time he asked for the raw data file, ran the full analysis on his own, and sent it to the partner to check. “They looked through everything,” he said. “They said it was 100 percent right.” When he asked what insights they would add, they had nothing to add.

Work that used to take weeks of finding time for each step now closes in a few days. The energy that went into checking, rechecking, and assembling data now goes into testing hypotheses and trying different ways of showing the results, so he arrives at the executive summary with something left to give.

“I spent all of my time just creating that story.”

The result is a one- or two-slide readout leadership can make a decision from.

Catching what careful people miss

Christian sorts the platform’s value into two kinds of catch: errors that people overlooked, and insights nobody saw.

The first came from a long, complicated jobs-to-be-done survey spanning several categories and demand spaces. He doubted anyone could catch every programming error in it, and the partner assured him it had been triple-checked. Fifteen minutes after he ran it through Merciv, he had a question-by-question error log showing which logic was right, which was wrong, and how to fix it. The partner’s response: “We didn’t catch that.” He now uses the same check on his own work whenever a task is too tedious to trust to a human read.

The second came from three related studies. The partner had analyzed each one separately, so Christian asked the team to wait until all three were complete and analyzed them together. One finding about the product had skewed the results of the first study, and it only surfaced when the three were read side by side.

“We would have gotten the first study back and made a whole bunch of changes to the product that were the wrong thing to do.”

Instead, the team read the first study through the right lens and used the other two to guide how to improve the product.

A 5 p.m. deadline and a 9 a.m. debrief

The most recent example came on the day of the panel. After lunch, Christian got a text: the team owed the CEO a presentation at 5 p.m. and needed results from a study that had just finished. That morning he had already loaded the data file and typed in what he wanted to see. By the time he walked the 20 minutes back to his room, the analysis was done. He sent it to the brand director as a first run with the story not yet smoothed out, and the reply was that it was exactly what he needed.

The week before Groceryshop, the team ran focus groups all day and into the night on Tuesday and again on Thursday. After each group, Christian loaded the transcript and the platform generated a report. While the next group was underway, he kept asking new questions and pulling in data from a related quantitative study. He also used the gaps he was spotting to tell the moderator what to probe. By the 9 a.m. debrief, he had a full report of findings in the team’s hands, where normally the debrief would be followed by a two-week wait for the partner’s report.

“It actually made the groups more interesting to listen to.”

Anyone who has sat through two and a half hours of focus group at a stretch will understand why that matters.

Where the reclaimed time goes

With several analyses running in parallel, a new request no longer means telling a stakeholder to wait a week or giving up a weekend. Christian described it as a command center at his desk, where he starts one analysis, kicks off another, and moves between as many as seven projects at once. More requests now add to the list without adding to the stress.

Speed is the easiest gain to measure, and Shaia argued it is not the most important one. When the menial work is handled reliably, an insights team gets room to be creative and confidence in answers it can verify and return to, and that is what turns research into better decisions for the whole company. For any team evaluating the tools on the floor at Groceryshop, he suggested that is the standard worth holding them to.