Albertsons' AI grew baskets 10% to 26%. It didn't pay back at once

Albertsons says shoppers using its AI assistants buy 10% to 26% more per order, yet the rollout gave no immediate return. Where it expects the money instead.

Albertsons Companies, the US grocer behind Safeway, Vons and Jewel-Osco, says shoppers who use its AI shopping assistants spend more per order: 10% more for its Ask AI search, according to its chief executive on the January 2026 earnings call, and up to 26% more for its fuller recipe and dietary assistants, according to the executive who runs them. The same executive said in August that the rollout did not pay back immediately. The money Albertsons is now counting on, about 200 million dollars a year, sits in a reorganisation of buying, stores and back office, not in the chat box.

That combination is unusual and useful. Most companies that cannot prove their AI spending pays off simply do not publish anything. Albertsons published a result, admitted the payback was slow, and then told investors where it expects the real savings to come from. For anyone running a store chain, that sequence is the lesson.

What Albertsons actually deployed

Over roughly 18 months Albertsons launched three customer-facing tools on its websites and apps: Ask AI, a conversational search that lets a shopper type a request in plain language; a planning tool that turns a meal or an event into a list; and a buying tool that turns the list into a basket. Ask AI was built on Google Cloud's conversational commerce agent, which the company says it was one of the first retailers to take to market. It is now merging the three into one assistant, and in August it added a Safeway plugin inside ChatGPT.

Behind the counter it runs a separate set of systems. In May it switched on a computer-vision tool, built in-house, that rates strawberries and grapes at distribution centres against its own quality standard and recommends a grade for an inspector to approve. On the July call it listed four enterprise AI priorities: digital customer experience, merchandising intelligence, labour optimisation and supply chain. It said better forecasts were already reducing manual intervention in ordering, and that an AI workforce management platform was on track to reach every part of the business in early 2027.

The basket number was real. The payback was not immediate. The savings target sits in the back office.

What do the 10% and 26% figures actually measure?

They measure how much more a shopper who uses the tool spends per order than one who does not. They do not prove the tool caused it. A shopper who opens an assistant to plan a dinner party was already going to buy more than one topping up milk. Albertsons has not published how many shoppers use the tools, over what period the figures were taken, or a test against a control group. Read them as a promising signal, not as extra revenue.

The 10% figure is the stronger of the two because the chief executive said it on an earnings call, on the record to investors. The 26% figure comes from an interview with Jill Pavlovich, senior vice president of digital shopping experiences, reported by the Wall Street Journal and picked up by trade titles. She also said the biggest baskets came when people used AI to build recipes around dietary needs, because they stop forgetting items.

What did it cost, and did it pay back?

Albertsons has not said what the tools cost. It has said the payback was slow: Pavlovich told the Journal the company did not realise an immediate return on the investment, even though baskets grew straight away. Consolidating three assistants into one is partly an attempt to change that.

There is also a cost the basket figure hides. In its second-quarter results on 13 October, Albertsons said its gross margin fell, mainly because of delivery and handling costs from growing digital sales, plus higher shrink. Digital sales rose 11% in the quarter, but identical-store sales fell 0.8% and the company withdrew its outlook for the year pending a full business review. A bigger online basket is only worth what is left after someone picks, packs and drives it.

Where Albertsons expects the money instead

On the July call, chief executive Susan Morris described ACI Edge: 11 operating divisions folded into four regions, and centre-store buying centralised in one team. She called technology and AI foundational to it, and put a figure on it in writing: about 200 million dollars of incremental annual run-rate benefit, building through fiscal 2026 with most of it in fiscal 2027, for about 50 million dollars of transition costs. That sits on top of an existing three-year productivity target of 2 billion dollars.

Be careful with that figure. The 200 million dollars is for the whole reorganisation. Fewer divisions and a single buying team would save money without any AI at all, and Albertsons has not split out what share it credits to software. What the company is signalling is that the dependable savings in grocery come from forecasting, ordering, labour scheduling and buying, where a small accuracy gain repeats across thousands of products every day.

Does this apply to a 200-person retailer?

Partly. The lesson on order of investment applies at any size; the customer chat assistant mostly does not. A large grocer can spread the cost of an assistant across millions of online orders. A small one usually cannot.

Our own arithmetic, not Albertsons': a regional grocer with 2 million dollars a year of online sales, where one in ten online shoppers uses an assistant and spends 10% more, gains about 20,000 dollars of sales. At a gross margin in the high 20s, before picking and delivery costs, that is under 6,000 dollars a year of gross profit, which is unlikely to cover a licence. The same grocer's fresh waste, ordering hours and rota planning are usually worth far more.

What to do on Monday

First, price your waste and your ordering time before you price any AI. Take one month of fresh shrink by department and the staff hours spent on ordering and scheduling. That is the pot Albertsons is now chasing.

Second, if a vendor shows you a basket uplift, ask the question Albertsons has not answered in public: against which control group? Ask for the result of a test where half of comparable shoppers saw the tool and half did not.

Third, if you do add a shopping assistant, decide in advance what payback you need and by when, and measure it net of picking and delivery cost. Albertsons is a reminder that a real increase in basket size and a slow return on investment can both be true at once.