Lowe's says Mylow users convert at triple the rate. What that proves

Lowe's online shoppers who use its Mylow assistant convert three times as often as those who don't. It is a real result, but not a controlled test.

Lowe's says online shoppers who use Mylow, the AI project advisor it built with OpenAI and launched on Lowes.com in March 2025, convert at three times the rate of shoppers who do not. Chief executive Marvin Ellison gave the figure on the company's earnings calls in May and August 2026, up from "more than doubles" in November 2025. The same system runs on handheld devices for associates in all 1,700-plus stores. Lowe's has not said what it cost, how many shoppers use it, or how much revenue it adds.

That makes it one of the few retail AI assistants with an outcome number the company has said out loud, repeatedly, to investors. It is worth taking seriously, and worth reading precisely, because a comparison between people who chose to use a tool and people who did not is not the same thing as proof that the tool made the sale.

What Lowe's actually built

There are two products on one foundation. Mylow sits on Lowes.com, at first for MyLowe's Rewards members on desktop and mobile web. Customers describe a job rather than type a product name, and it answers with steps, product suggestions and how-to content. Mylow Companion went to every associate on the sales floor two months later, on the handheld devices they already carry.

The design choice that matters is that both are about projects, not products. In OpenAI's customer story, Lowe's digital commerce chief Joe Cano says most visitors arrive with a problem, not a product in mind. The same story describes the associate tool suggesting what Lowe's calls attachments, the items people forget, and store operations chief Vinny Scalese gives the example of tape and sandpaper for someone patching drywall. A project question is a basket, not a single item.

Voice turned out to be a requirement rather than a feature. Chief digital and information officer Seemantini Godbole says in the same story that more than half of associate interactions are spoken, because people on a sales floor have their hands full.

How the numbers moved

Read Lowe's calls in sequence and the adoption curve is visible. In November 2025 the two tools were answering nearly one million questions a month, and Ellison said conversion "more than doubles" for online customers who engage with Mylow. In February 2026 he put it at "roughly double". By May the customer side alone was handling more than a million inquiries a month, associates had asked Mylow Companion more than 5 million questions, and the conversion gap had become triple. In August he repeated triple, with more than 25 million questions answered across customers and associates since launch.

In stores, Lowe's reports customer satisfaction up about 200 basis points where associates use the Companion.

Lowe's has published the ratio. It has not published the denominator.

Does a triple conversion rate mean the assistant made the sale?

Not on its own. The figure compares shoppers who chose to ask Mylow a question with shoppers who did not. Someone asking how to fix a leaking faucet is already closer to buying than someone browsing. With no published holdout test, usage share or attributed revenue, the true lift is likely below three times.

That is not a reason to dismiss it. The figure rose from about two to about three while usage grew. If the keenest early users were the whole story, you would expect the ratio to fall as more ordinary shoppers joined, not rise. And Ellison listed Mylow as one driver of online growth alongside free and same-day delivery, a redesigned site and app, visualisation tools and a new marketplace. Online sales rose 15.7% in the second quarter while total comparable sales grew 0.2%. The assistant is part of that, not all of it.

What did it cost?

Lowe's has not said. The build was with OpenAI, and Lowe's said in November 2025 that OpenAI had recognised it for passing 100 billion tokens of use, which tells you the running cost is metered and not trivial. Group capital expenditure guidance for fiscal 2026 is up to $2.5 billion, which covers stores, supply chain and technology together and cannot be split to Mylow.

What can be said is how Lowe's ran the project. Chandhu Nair, its technology lead for data and AI, says every AI system at Lowe's needs a business sponsor at senior vice president level before it is built, and that engineers walked store aisles with operations staff from day one. That is a governance cost as much as a software one.

Why a retailer would build its own agent now

The fashionable argument this week is that agents will replace apps, and that the companies owning the agent people talk to will own the customer. Lowe's is doing both at once. It told investors in February that it is working with leading digital platforms so it can take part in agentic commerce, and it is keeping its own agent on its own site, where it controls the answer, the product data and the add-on suggestion.

The reason is in the numbers above. A third-party agent can send Lowe's an order. It cannot put tape and sandpaper next to the drywall patch, and it cannot help a seasonal hire in the plumbing aisle answer a question about a part they have never sold. The associate half of Mylow is the part no outside platform can replace, and it is the half with a published customer-satisfaction effect.

Does this apply to a 200-person retailer?

Only the store half, and only if your products need explaining. A mid-sized retailer cannot build what Lowe's built, but it can buy a staff-facing assistant that answers from its own product data. Try that before a customer chatbot, measure it against a store that does not have it, and count questions asked per associate per shift.

What it does not mean

It does not mean an AI assistant triples conversion. It means shoppers who use Lowe's assistant convert at triple the rate of those who do not, in a category where questions are unusually common and baskets are unusually large.

It does not mean the tool paid for itself. Lowe's has not published a cost, and neither has anyone else.

And it does not mean a customer chatbot is the place to start. Lowe's own most transferable result is on the sales floor, where the problem it names is familiar at any size: new, seasonal and cross-trained staff facing questions they cannot answer. If you run stores, the next action is to count how often that happens in a week before you look at any vendor. If you run a web shop, run any assistant against a holdout from its first day. That is the one number Lowe's has not published, and the one you will need.