Industry 04
Retail
A prompt change is a product change. Your shoppers see it at once.
In your search, catalogue and merchandising, a prompt change reaches your shoppers the day it ships. So a release goes out when its score on a fixed dataset clears the threshold, and you see drift before conversion does.
What changes here
A scored evaluation suite that gates the release
Catalogue scale, where one change touches millions of records
Seasonality, where the baseline and the alert move together
Personalisation under the consent rules of each market
Your shoppers touch the model
So a prompt change ships on a score. Your shoppers meet it in search and the catalogue, with no layer between.
- A
The score decides the release
- B
Without it, your quality drifts until conversion notices
- C
One change reaches millions of records
- D
Seasonality moves the baseline, and the alert with it
Work in this industry
AI work first, then the engineering underneath.
Retail · Brand reputation software provider
Millions of reviews, asked a question in plain English
Topic modelling over millions of customer reviews, with a conversational agent on top so the answer takes seconds rather than an analyst.
- 15 seconds or less to answer a question
- 4 months to deliver the conversational agent
- Millions of reviews analysed
Retail · Promotional data provider
Printed promotions, cropped and read by machine
Computer vision crops each advertisement off the page, reads printed and handwritten text, and classifies the offer, at a volume no manual team could match.
- 99.3% cropping accuracy
- Almost 93% accuracy recognising the offer type
Retail · Brand management SaaS
Unwanted reviews flagged before a person reads them
A fine-tuned BERT model reads incoming reviews for spam, off-topic and negative content, including irony and sarcasm, and tells the team which ones to remove.
- 80% accuracy predicting unwanted reviews
- 334 person-hours over 2 months
Engagements
What we take on here
The same 10 services, held to this industry's constraints.
Next step
Name the drift.
Tell us where your search or catalogue model last drifted. You get a written answer on what breaks first and what it costs.