Diagnosis of the operational workforce in retail stores at iFood
- Year
- 2025
- Duration
- 2 weeks
- Tools
- interviews, in-app survey, AI
- Context
- Multicategory B2B · picking
- Industry
- Marketplace · Wholesale
- My role
- End-to-end research leadership
Summary
Inside one of the country’s largest grocery/wholesale marketplaces sits a workforce that makes the consumer order possible: pickers and picking managers. They receive the order, walk the store, select the product, and pack it for delivery. You’ve seen them — multi-level carts carrying several orders per loop through the aisles.
That workforce feels undervalued, unmotivated, and sidelined. The result is high turnover and a direct hit on order-quality KPIs. I led the end-to-end diagnosis that put this blind spot on the table.
The problem
The tooling hypothesis
When I joined this investigation, a hypothesis was already circulating in the business: the order-quality bottleneck lived in the tooling. If the picking app improved, the metric would improve with it. No one had evidence to support that view — it was consensus because the link between tooling experience and KPIs kept showing up.
The blind spot
At the same time, this workforce was a blind spot for product. We had roughly 33 thousand active users in the picking app and didn’t really know who was on the other side. Notice the size of the hole: whoever picks the order directly influences stockouts, delays, and wrong items — exactly the indicators the team wanted to move with tooling.
How I ran it
The full cycle ran in two weeks. Before fieldwork, I mapped stakeholders and facilitated the analysis pillars with the involved areas, from Operations to Social Impact. That’s the moment to align outcome expectations with leadership while there’s still time to change the question.
In the field I ran two fronts: qualitative inside the store, with live operations and in-depth interviews with picking managers; and quantitative in-app across the three thousand highest-volume stores, which brought 714 respondents among pickers and managers. Alongside that, five benchmarks. I used AI in planning and in crossing qualitative with quantitative, and reallocated the hours saved into participant recruitment — where field research usually stalls.
What we found
60% under one year
The base turns over faster than any training can keep up: 60% of pickers have less than a year in the role. Tooling improves execution for people who already know what they’re doing, but in a base that arrives unprepared the tooling gain gets wiped out. This is where the tooling hypothesis lost strength.
57% flexible or outsourced
Most of this workforce isn’t store staff either. 57% is flexible or outsourced, versus 35% permanent. In other words, the product had been designed assuming an employment relationship most of the base doesn’t have.
74% not fully valued
Recognition closes the loop: 74% don’t feel fully valued in the store where they work.
What I decided
I recommended discarding the tooling hypothesis — the one that already had sponsorship — and redefining who the subject of the problem is.
Discard the tooling hypothesis, despite the sponsorship it already had in the business.
Segment by operating model, not demographics — the cut that exposed that the profile most dedicated to picking is also the most outsourced.
Treat the outsourced worker as a first-class user — something no one had asked for.
Four verticals left out of scope, including Pharmacy (the largest volume in the base): a diagnosis that tries to cover everything concludes nothing.
What changed
The team’s investment moved: out of tooling and into onboarding and category-valuation factors. Prioritization started to include peak-hour labor reinforcement, with advance scheduling of pickers on high-demand days.
Operational profiles became an artifact reused by tactical teams to interpret tooling-usage behavior — the outcome I value most, because it’s my material used in a decision I wasn’t even in the room to defend. And it unlocked a new front: revisiting onboarding to train the picking function itself, not just how to use the app.
Retrospective
The subject of the problem
Diagnosis is also a decision
My principle is that diagnosis isn’t only collection — it’s also deciding who the subject of the problem is. All the material was already available before me: the numbers were in the database, the partners were there, the stakeholders too. What didn’t exist was the willingness to say out loud that the most important user of that operation is someone the company doesn’t employ, doesn’t identify, and doesn’t see.