Feedback from all 260 stores, answered the same day

Trolley sells the two minutes between a car and a coffee. It opened in Kuwait in 2010 with four shops and a curated shelf of imported snacks, and it now runs around 260 convenience stores inside petrol stations, malls and universities across Kuwait and Saudi Arabia. Everything about the format is built for speed.

The feedback form was not. It asked for three minutes from a shopper who had just spent two inside the store, and the numbers said exactly what you would expect: 92% of the people who opened it never reached the end. The handful of responses that did arrive were slow to read, hard to attribute to a branch, and older than the shift they described. A complaint about a fridge in Riyadh reached the person who could fix it days after the shopper had already decided not to come back.

HoopAI rebuilt the loop end to end — the form, the code on the counter, the AI that reads every answer, and the dashboards the answers land in. It runs in every store Trolley operates.

• Six questions, 45 seconds: a HoopAI form written for a phone held in one hand, not a desk.
• A QR code for every store: one unique code per branch, so each response arrives already tagged to a store, a city and an area manager.
• Layers of AI on every response: language, sentiment, urgency and ownership are all decided in the seconds after the shopper hits send.
• A dashboard for every area manager: each manager sees their own stores, and only their own stores.

The challenge

Trolley already knew its customers had opinions. It could not get to them fast enough to matter.

The old form was long, generic and the same everywhere. It carried no idea of which branch the shopper had just walked out of, so someone had to be asked to type it in — one more field on a form that was already losing people. Nine out of ten walked away. What survived was a thin, self-selecting sample: mostly people annoyed enough to spend three minutes saying so, which is a poor way to learn what an ordinary Tuesday looks like across a few hundred stores.

Then there was the second half of the problem, the half a shorter form does not fix. Even a good response is worth nothing if it sits unread. Trolley needed the bad ones to reach a human within minutes, in the language the shopper wrote in, and it needed the good ones to turn into something the business could see — a rating, a review, a reason for the next shopper to walk in.

One QR code per store

Every Trolley store has its own code. Not a shared campaign link with a branch dropdown behind it, but a distinct QR code per branch, generated in bulk and printed for the counter of the store it belongs to.

That single decision removes work from both ends of the loop. The shopper never has to say where they are. The response arrives already carrying its store, its city, its country and the area manager it belongs to, so nothing has to be sorted, matched or reconciled afterwards. Attribution stops being a data-cleaning job and becomes a property of the code itself.

A second code sits beside it. Some people do not want to answer questions — they want to say one good thing and get on with their day. That code goes straight to Trolley’s Google listing for that branch, so a shopper who would rather skip the survey can still leave a review where it counts. Two codes, two intentions, no dead ends.

Six questions, 45 seconds

The HoopAI form asks six questions and takes an average of 45 seconds to complete. The old one took three minutes.

Most of that came out of the design rather than the question count. The form knows which store it belongs to before it renders, so it never asks. It runs in the shopper’s language. It is built for one thumb on a phone screen in a car park, with answers you tap instead of type, and it never asks for a name, an account or a login to begin.

Drop-off fell from 92% to 21%. Four out of five people who start the form now finish it — and because the people who finish are no longer only the ones with a complaint to file, the picture Trolley gets back is a fairer one.

The layers of AI behind one answer

A shopper taps send and the response is not filed. It is read.

Several AI steps run in sequence, each one answering a narrow question, each one handing its answer to the next. The point of splitting the work this way is that no single step has to be clever about everything, and every step’s output can be inspected on its own.

• Language: the response is read in the language it was written in, and that language is carried through everything that follows — the internal summary, the reply, and the email that reaches the shopper.
• Sentiment: every response is scored, not only the ones that use obvious words. This is the switch the rest of the system runs on.
• Urgency: a mild grumble about queue length and a food-safety complaint are both negative. They are not the same event, and they do not get the same response time.
• Ownership: the response is routed to the right expert using skills, availability, capacity limits and business rules, so it lands with a person who can actually close it rather than in a shared tray.
• Reply: on-brand replies are drafted for the team to send, in the shopper’s language, with the store and the specific complaint already in them.
• Rollup: the response joins its store’s numbers, its area manager’s dashboard and the national picture, so one person’s Tuesday becomes part of a trend somebody can act on.

All of it happens between the shopper putting their phone away and reaching their car.

When the answer is bad news

A negative response does not wait for a report. The customer service team is notified as it lands, with the store, the shift, the language and the substance of the complaint attached, and it is assigned to a named person rather than announced to a room.

The follow-up email goes out in the shopper’s own language. That is not a courtesy detail in a market split across Arabic and English — a recovery message in the wrong language reads as a form letter, and a form letter is how a fixable complaint becomes a lost customer.

The store manager and the area manager see the same response at the same moment. Nobody finds out at the end of the month that a branch had a bad week.

When the answer is good news

Most responses are not complaints, and a satisfied shopper is the most under-used asset in retail. Trolley now asks them, at the one moment they have already decided to say something nice.

A positive response triggers an email with the ask in it and the buttons right there — one tap to the Google listing for the exact branch the shopper visited, no searching, no hunting for the right pin on a map. People click them. Store ratings move, and they move per branch rather than as a national average that hides the difference between the best store and the worst one.

It closes a loop that used to run one way. The same 45 seconds that tells Trolley a store is doing well now also tells everyone searching for that store.

A dashboard for every area manager

Feedback that only reaches head office changes nothing on a shop floor. Every area manager at Trolley has their own dashboard, scoped to the stores they are responsible for.

The scoping is enforced, not cosmetic. Access runs on roles, field-level permissions and team scopes that mirror how Trolley is actually organized, so a manager in Kuwait sees Kuwait, a manager in Riyadh sees Riyadh, and neither has to be trusted to look away from the rest. That is what makes it safe to hand this data to every area manager in the company instead of keeping it in one report that six people read.

What each of them gets is their own stores ranked against each other, the responses behind every score, and the ones still open.

The security layers

Trolley is a listed company collecting thousands of pieces of customer sentiment a month across two countries. The programme was built so that the interesting question — what do our shoppers think — never turns into a second, worse question about who can see it.

That starts below the programme, with the platform it runs on.

• SOC 2: the controls behind the platform are audited rather than asserted.
• Encrypted in transit and at rest: a response is protected on the way from the shopper’s phone and once it has landed.
• GDPR: the data-protection obligations that follow personal data are built into how it is stored, accessed and removed.
• HIPAA, with a signed BAA: a convenience store does not need healthcare rules. It is worth knowing the platform underneath is run to a bar where a hospital will sign its name.

On top of that sits the way this particular programme was designed.

• Collect less: the form asks six questions and nothing else. No account, no login and no name required to leave feedback.
• Attribute by code, not by interrogation: the store is known from the QR code, so the shopper is never asked to identify themselves or their location.
• Role-based access: roles, field-level permissions and team scopes control who sees and changes what, mirroring Trolley’s real structure.
• Governed storage: customer detail is stored with the privacy, governance and access controls that regulated data requires.
• A full audit trail: every action taken by a person, an automation, an integration or an AI step is tracked, so an answer to “who saw this, and what did they do with it” always exists.
• Single sign-on for staff: managers reach their dashboards through SSO rather than another password to reuse.

None of these layers is remarkable on its own. Together they are the reason a feedback programme can be handed to hundreds of managers in two countries without becoming a liability.

Saudi Arabia first, then Kuwait

The rollout started in Saudi Arabia, across all 85 stores in that market. That was deliberate: a network large enough to prove the model and contained enough to fix it in, with one set of area managers learning the dashboards before anyone else depended on them.

Three months later the programme extended across Kuwait — the older and larger half of the business — bringing it to every Trolley store in both countries. The second launch needed new codes and new managers, not a new system.

It has not stopped there, because the code is generated per store rather than per campaign. Trolley opened 25 net new stores in the first half of 2026 alone. Every one of them joined the programme on its opening day, with its own code, its own numbers and its own line on an area manager’s dashboard.

The results

The first full month across the network brought in 5,000 responses. It now averages around 250 a day.

• 45 seconds to complete six questions, down from three minutes.
• 658% more feedback responses a month than the old form produced.
• 21% drop-off, down from 92%.
• Every store across Saudi Arabia and Kuwait — around 260 of them — each with its own code and its own numbers.

The number Trolley cares most about is not on that list. It is the gap between a shopper having a bad experience and somebody at Trolley knowing about it, which used to be measured in days and is now measured in minutes.