Nestlé Professional
A concept case study.
Self-directed exploration, not a shipped feature.
It's 7am. A restaurant's espresso machine throws an error code, and the doors open in an hour. Right now, that means digging through a PDF manual or calling a support line and waiting on hold. This concept gets that manager an answer in under a minute. Or an honest "here's when a technician can get there," instead of a guess dressed up as one.
The Real Part
I worked on the Nestlé Professionals account at Almost Impossible Agency, building out the web and digital presence for the B2B/HORECA side of the business: hotels, restaurants, cafes, and offices running Nestlé equipment. That's the foundation this concept builds on. The AI layer below never went live. I'm not implying it did.
The problem worth solving
Nobody reads an equipment manual for fun. They open it when something's broken and the clock is running. The existing path, dig through a PDF or call and wait, works. It's just slow, and slow costs a business owner real money in downtime. That gap between "something's wrong" and "someone tells me what to do" is the whole concept.
Who it's for
A HORECA account holder, not a tech enthusiast looking for a new chatbot to poke at. Someone managing a busy floor who wants the machine running again. Every extra tap or screen is a cost to a person who's already under pressure, so the design has to earn its way past that.
What gets designed
Four screens made the cut, and each one exists because it maps to a real decision point in the moment, not because AI products are expected to have four states.
Entry point
A photo of the error code, or a typed description, whichever is faster in the moment. No forced format. Mobile-first, because this happens on the floor of a restaurant, not at a desk.
Confident-answer state.
The assistant walks through the fix step by step and shows the actual manual excerpt the answer came from. That's the trust mechanic. Nobody follows instructions from a black box when the machine costs thousands and the fix might void a warranty.
Low-confidence state
When the assistant isn't sure, it says so plainly and routes straight to booking a technician. The design call here is resisting the urge to make the AI sound more certain than it is. A wrong guess costs more trust than an honest "I don't know" ever will.
Same assistant, different job. "I'm low on syrup" turns into a short back-and-forth that lands on an actual order, not a product catalog built for someone browsing shelves instead of restocking under pressure.
Reorder flow
The decisions that actually matter here
Anyone can polish a screen. The design thinking shows up in a few specific calls:
Showing the manual excerpt instead of a clean, confident-sounding answer trades polish for trust, on purpose. A restaurant manager trusts a cited source more than a system that sounds sure of itself.
Escalation is framed as the assistant doing its job well, not failing at it. "I don't know, here's a technician" is a better outcome than a bad guess dressed up as a fix.
The input flow doesn't force a format, because the person using it is often typing one-handed with the other hand on a broken machine. A photo or a fragment of text both need to work.
First-time users get one short screen explaining what the assistant can and can't do. Setting that expectation early heads off the "why didn't it just fix it" frustration that shows up later if it's skipped.
A concept I designed exploring how AI-assisted support could extend the Nestlé Professionals
Fitting into what already exists
This sits inside an existing professional portal, so the components extend the platform's visual system instead of replacing it. Build order in Figma: typography and spacing first, reusing what's already there, then the four core screens, then the states within each (loading, confident, escalate, error) as variants off shared components, so the file stays manageable instead of sprawling into one-off screens.