
Very Lets an AI Decide Who Pays Less
Drapers reports that Very's website can now spotlight discount offers unique to individual shoppers, and the only result Very has quantified comes from a single strategy. A discount reserved for shoppers a model picks is a personal price, and Britain's guidance on targeted offers was written for categories a shopper can recognise.
Sir John Crabstone
Drapers reports that The Very Group’s website can now “spotlight discount offers unique to customers based on individual shopping trends”. A discount reserved for shoppers a model picks deserves a plainer name: a personal price. Very has put a number on the results of only one strategy.
On Monday Walmart promised never to use “shopping history, urgency or willingness to pay to set an individualized price”. Very’s feature runs in the polite direction, but it is still a price chosen for one shopper.
In Drapers’ account, Made With Intent’s “agentic decision engine” has run on Very’s site since early September, reading more than 800 behavioural signals to help decide “what they see and when”. Sarah Hanna, Very’s digital director, reported “revenue per customer increasing by a fifth in one of our agentic strategies, and an uplift in overall conversion”. The retailer’s wider claim that revenue has risen since launch, which the headline rests on, comes without a figure.
The vendor’s manual never mentions Very, so it can only suggest what the fifth measures. Every agentic campaign keeps a control group, whose share starts at the midpoint of a 5% to 50% range and moves with the campaign’s results. If a Very strategy is one such campaign, the fifth is that campaign against its own control. Drapers gives no certainty score and no blended result for every strategy against shoppers left alone.
Made With Intent’s discounting page promises incentives “only when a shopper is hesitating, instead of giving them away to everyone”. Showing one discount to all, it warns, tells “your most engaged buyers that the price they were happy to pay wasn’t the real price”. Targeting makes that price no more real; it keeps the engaged buyer from finding out.
The vendor frames its trigger as hesitation, yet every discount the engine withholds is a bet on what that shopper will pay anyway.
Connecticut’s new ban on surveillance pricing takes effect on 1 July 2027 and binds retail sellers. It exempts a discount offered to attract or keep a customer, provided the price is verifiably below the generally available market price. A price raised with personal data must carry the warning “THIS PRICE WAS INCREASED USING YOUR PERSONAL DATA”, or a substantially similar one; a price that falls needs none.
Britain’s competition regulator put discounts in the same category in 2021. Its discussion paper on algorithms treated “providing discounts to selected customers” as a form of personalised pricing, while allowing that it can benefit some consumers. A footnote on loyalty penalties adds that shoppers paying the higher advertised price also get a price “personalised” to their behaviour.
The CMA’s price transparency guidance, issued last November, says in paragraph 5.50 that an offer open only to certain consumers “must also make explicit which consumers will qualify for that offer”. Its examples are new customers and NHS workers, categories a shopper can recognise. A model that picks its qualifiers shopper by shopper is a harder case, and Drapers’ report does not say how Very tells anyone.