0 complaints in August 2026Behind the scenesEmails, web forms, calls, retailer returns and social posts all land in one queue, as written, in whatever language, with any photos attached.What is automatedCollecting every complaint in one place, whatever the channel.
Today A team codes about one in five by hand, weeks later, into a spreadsheet.
Customer's photo of the neck
One complaint, as the system reads itBehind the scenesA language model reads the words in any language, and the photo, and fills the same fields every time, including the customer's tone, so angry repeat buyers are seen first. The batch code printed on the neck is read from the photo, which ties the complaint to a line, a day and a site.In this demoScripted reading. The batch code format, line, date, time and site, is invented.Could go furtherAsk the customer for the batch code in the reply when there is no photo, so the link is made either way.
Every complaint, grouped by what went wrongBehind the scenesComplaints are grouped by the problem described, not by the words used, so flat, no fizz and sin gas land together. Each group is compared with last month to show what is new.What it achievesTwo groups stand out as rising sharply. Both point at the bottle cap, which a keyword search would not have joined up.
From complaint to causeBehind the scenesBatch codes link complaints to production records: which line, which shift, which materials went in. Joining those to quality test results shows what changed.AssumptionInvented factory, lot and figures.What it achievesThe cause is found in days, with evidence a supplier cannot argue with.
Cap torque tests on those shifts
1.3 against a specification of 1.6 to 2.2
The caps from that lot were not tightening to specification, so gas escaped and some leaked.
The taste complaints trace nowhere in production. They follow a heat wave in South West stores of one retailer, so they go to distribution, not the factory.
Quality alert, drafted for the quality managerBehind the scenesThe alert is drafted from the trace: which batches, where the stock is, what to tell the supplier and what to tell customers. The quality manager decides; the system does the paperwork.Could go furtherSend the supplier the evidence pack directly, and watch the next month's complaints to confirm the fix worked.
0%
of complaints read, up from 20% coded by hand
0 days
to find the cause, down from 19 (modelled)
0
units held before reaching shelves
£0
refunds and returns avoided (modelled)
Every complaint, batch, test and alert sits in one model, so the same view serves quality, the factory, the supplier and the board.AssumptionInvented maker, sites, supplier and volumes. Modelled: days to find a cause, units held and refunds avoided.What it achievesThe real figures come from running last quarter's complaints and batch records through the same steps.
Complaints, August 2026
3,412
traced to one line and one cap lot
Interactive demo
Watch it run in 94 seconds, or click through at your own pace. The i buttons show what happens behind the scenes.
That was one day
3 days from complaints to cause, not weeks
Modelled on an invented maker. Your own complaints and batch records would give the real figure.