Enterprise Field Notes · Issue #18

Target Hired Lowe's AI Chief. His Page Was Gone in Six Days.

Five million logged questions. Every one records what happened. Not one records why.

By Ben Pickett · August 20, 2026

New here? Subscribe to Enterprise Field Notes, one new issue every week.

The system nobody wanted to learn

Before I went to work in retail I was at a healthcare company, and I was the only person there who knew how to code the system that adjudicated claims and referrals.

That is the confession.

Last week Target hired the man who built Lowe's AI, and it put me back in that room.

I took the contract details and wrote the rules that made them comply, the plan terms and the regulations around them, and the logic ran deep. It was documented. That is the part people get wrong about this. The documentation existed, it was fine, and nobody wanted to learn it, because the fastest way to get an answer was to walk over and ask me.

Then I took a new job and gave my notice, and everyone found out at once what that meant.

Here is the part I want to be honest about. There was nobody to review my code. Not because anyone forbade it, but because there was no one else who could. I was careful. I cannot tell you I never got a rule wrong.

Careful is not a control.

But I want to be accurate about what that cost, because it was not money going out the door unchecked. There was a second look. An appeal would come in, or a claims analyst would pull the claim, or a doctor would review it, and those controls work. What they cost is time. The claim sits longer. The patient waits. Somebody in a doctor's office spends an afternoon on the phone finding out why. So the price of my being the only person who understood the logic was not a bad payment nobody caught. It was delay, and it landed on patients and on the people taking care of them. It was quality of care.

So on the way out we did the thing we should have done years earlier. I sat with my team and we built a front end, so that instead of one person writing rules in code, the people assigned to a contract could walk through it together and check each other. It was not finished when I left. It was reviewable, which it had never been.

That system is where I started caring about testing and reliability the way I do now.

I had not thought about that in a long time. I wrote about Home Depot two issues ago and I was not looking for another big box retailer. But on August 11 I went to look Chandhu Nair up.

What is already gone

Chandhu Nair does not start at Target until August 24. His page is already gone from the Lowe's leadership site.

I checked it against a colleague's page in the same browser session, seconds apart. That one loads. His redirects to an access denied notice, and his name is off the leadership index. It was gone by August 17, six days after Target announced the hire, and it was still gone this morning when I ran the same check again.

On August 19 Lowe's held its second quarter earnings call. Marvin Ellison and Joe McFarland both talked about the AI companion Nair's team built, and Ellison put a new number on it: "Since its inception, MyLowe, which also powers our associate AI companion application, has supported over 25 million questions from customers and associates." Neither of them said Nair's name, and I did not find a question about him anywhere in the transcript.

Companies take pages down and this one is probably routine. What interests me is narrower. Twenty five million counts customers and associates together, as of this week. The five million I am going to divide in a minute is associates alone, as of May. You cannot subtract them, and I am not going to try. Both are counts of what happened. The thing I have just spent five hundred words on is in neither of them.

The hire

Nair starts on August 24 and reports to Target's chief information and product officer. He spent six years at Lowe's, most recently as senior vice president of stores, data, artificial intelligence and innovation, and he also ran Lowe's Innovation Labs.

The trade press covered it as a hire. Retailer appoints AI chief, arms race continues, here is the turnaround context. Every outlet ran roughly the same piece.

I want to look at it from the other side. Not what Target just gained. What Lowe's just lost, and whether it is the thing anyone thinks it is.

What he actually built

Lowe's launched Mylow in March 2025, a customer-facing home improvement advisor built with OpenAI. Two months later it launched Mylow Companion to every associate across more than 1,700 stores, and called it the first time a retailer had put this kind of tool in associates' hands at that scale.

The numbers came from named executives on the Q1 earnings call, which is a higher bar than a press release.

CEO Marvin Ellison said Mylow was supporting over a million customer inquiries a month, and that online customers who use it convert at triple the rate of customers who do not. Joe McFarland said associates had asked more than five million questions through Mylow Companion since launch. Ellison put a number on the outcome, in the Q and A rather than the prepared remarks: "200 basis points of customer satisfaction improvement based on the use of the companion tool." He did not say two hundred basis points of what, measured how, against what base.

Divide the five million, which nobody on the call did. Against 250,000 associates over roughly twelve months, that is about twenty questions per associate per year, one every eighteen days. Hold it loosely in both directions. McFarland said more than five million, which pushes the number up. Retail turnover means well over 250,000 people held a badge across that year, which pushes it back down.

Then take the reading that flatters Lowe's most. Nobody disclosed monthly actives, so the real users may be a narrow band asking far more. Say thirty thousand associates do nearly all of it. That is about a hundred and seventy questions each, two in every three shifts on a full time schedule, from twelve percent of the workforce, and near silence from the rest. The generous reading does not rescue the headline, it relocates it.

Putting AI in 250,000 hands describes a deployment and not a habit.

The distance between those two things is the change management everybody says is 70% of the work.

The conversion figure and the satisfaction figure both deserve a caveat the earnings call did not give them. A shopper who opens an advisory tool is already further into a project than a shopper who does not, so the conversion gap is not all lift. And the associates who take up a new tool are probably the ones already doing the job well. Neither is a controlled result. They are still the most specific numbers any retailer has put on the record.

Chief digital and information officer Seemantini Godbole put the point plainly: whether associates have been on the job for five weeks or five years, they can be confident they are giving expert-level advice.

How it got that way is stranger than what it does.

The decision almost nobody copies

Early on, it was wrong a lot.

Nair has said so plainly. "We were about 60% accurate." That number attaches to the early rollout, not to 1,700 stores. What matters is the posture it describes.

Think about what most companies do with a tool that is right six times out of ten. They hold it. They put it back in the lab, they add evaluations, they schedule another pilot, and they wait for a number that looks safe enough to defend in a steering committee.

Nair shipped it, and his reasoning is the most useful sentence in this whole story: "The only way to make it accurate is for more people to get into it and give me feedback, thumbs up, thumbs down."

Every answer carried a thumbs up and a thumbs down. The team worked the ratings daily. Accuracy was not a gate to clear before deployment. Accuracy was downstream of deployment, and holding the tool back would have starved it of the only thing that could improve it.

I want to be careful here, because this is the kind of idea that gets repeated badly. It works for an advisory tool with a trained human standing between it and the customer. An associate who gets a wrong answer about a faucet asks a colleague. That is not a license to ship a system that is right 60% of the time and let it cut checks or approve claims. The human in the loop is what made the low bar survivable, and the human in the loop is exactly what most agentic deployments are removing.

The thing that actually moved it

The first version was not even for associates.

It started as an analyst tool sitting on catalog data, useful to managers who wanted to ask questions about sales. Someone from the business side pushed to point it at all 250,000 associates instead. That redirect is most of the program.

The voice feature had problems. Associates used it anyway. Roughly half of them, with a customer standing in front of them, preferred it to typing.

The data did its job. It said what was happening. What it could not say, in any quantity, was why.

Nair supplied that, and not from a dashboard. "If you have a customer in front of you, you don't want to be looking down on the phone and typing, so they were using the voice feature, which had problems, so we quickly pivoted."

The telemetry produced the signal. A man standing in an aisle produced the explanation.

They are two different things, they live in two different places, and only one of them is in the file.

That is the whole of what I want you to take out of this. Five million logged questions are five million records of what happened. Not one of them contains the sentence Nair just said, and the sentence is the part you would need to build the thing again.

Adoption did not come from the technology either. It came from the operations organization, which Nair credits directly. They embraced it, communicated it, gamified it, and created incentives to use it. The team has said uptake ran ahead of what they expected, though nobody has said what they expected, and the arithmetic above is the reason I would want that number. Either way the lift came from an operating decision rather than an engineering one.

Nair has a line about how his team works that I have not been able to stop thinking about, and I want to give you all of it, because the second half is the part that argues against me: "One thing that we do consistently as an AI team is we are on the floor. You cannot be an AI engineer or a data scientist on our teams unless you're walking the floor and seeing the tools in action."

So he did write it down. It is a condition of employment, said to a reporter, on the record. It is the most direct answer to this problem I have seen any retailer give, and it is the strongest version of the case against what I am about to argue. I want it on the table before I argue it.

Because a rule can be real, and stated, and still have no ticket, no line item, and nobody whose actual job it is to defend it.

And then the sentence the rest of this issue rests on, given to CDO Magazine: "It's 30% tech and 70% change management."

Hold onto the voice story, because it is going to matter later. I want to be accurate about the claim. A good analyst could have gotten there from the data eventually, with the right segmentation, a quarter to work in, and somebody thinking to ask the question in the first place. A man standing in an aisle already had it.

And be accurate about the 70% as well, because it is not what I would like it to be. Change management at Lowe's was a funded program with an owner. Comms, gamification, incentives, a ratings queue worked every day. All of that is documented and all of it stays. The part with no file is smaller than the quote makes it sound, and it is not the change management. It is the noticing that tells you which thing to build next.

The number that says he is the exception

Randy Bean and Tom Davenport have run a benchmark survey of senior data and AI leaders for fifteen years. Ninety percent of this year's respondents are the most senior data or AI person at their company, and there are more than a hundred of them, and Lowe's is one of the listed participants. It skews heavily to financial services and North America, so read it as a leadership panel and not a census.

This year, 93% of them picked cultural challenges over technology as the greatest impediment to AI adoption at their company. Read that as the forced choice it is, two options that sum to a hundred, so technology did not place last in a wide field, it placed second of two.

Then Davenport asks the question in the foreword, and it is the sharpest thing I read all month:

"But how many CDOs spend 93% of their own efforts addressing human issues?"

Almost none, and everybody knows it.

Nair's team had a standing answer to it before Davenport asked, and it was not a philosophy. It was a rule about where engineers spend their Tuesdays.

What actually does not transfer

Here is what Target is buying, and it is a lot: a proven operator who has shipped a large program twice, who ran the innovation unit Lowe's had been funding before most companies had one, and who knows what the second year of one of these looks like.

And here is what Lowe's still has, all of it, none of it leaving: the models, the OpenAI relationship, the platform, the governance committee, and more than five million logged associate questions, with a thumbs up and a thumbs down offered on every answer.

And a thumbs down is not nothing. It says an answer was wrong. What it cannot say is why, and it cannot see the thing that never became an answer at all. Nobody rates the moment they decide not to look down at their phone. That is the case the voice pivot turned on, and it is the one the ratings queue could not have caught.

The logs showed that associates kept choosing a voice feature that had problems. They could not say it was because an associate will not look down while a customer is standing in front of her, and they certainly could not say the answer was to fix voice first. Somebody had to be standing in the aisle for that, and then had to decide it mattered enough to move the roadmap.

This is the thing we keep getting backwards. We document the AI system exhaustively. Model cards, evaluation suites, governance registers, vendor contracts. All of that survives a resignation automatically, because none of it lives in a person.

So be precise about what walks out on August 24, because it is not everything and the honest version is more useful than the dramatic one. The artifacts survive and explain nothing. The rule survives. The team survives, and while those people are still walking the floor the explanation is still being made. Lowe's is not left holding an empty bag.

The part that worries me has nothing to do with one man's head.

A habit like walking the floor lasts because somebody with standing keeps paying for it. It has no ticket and no line item. A data scientist spending a day in a store does not show up as output anywhere. It runs on sponsorship, and sponsorship is the thing a leadership change puts back on the table.

That is not a prediction about Lowe's. It is what happens after any leader leaves. Teams get moved under different executives. Roadmaps get re-pointed at whatever the new person was hired to fix. Every practice that cannot defend itself on a resourcing spreadsheet gets asked what it is for, and a rule whose entire value is that somebody noticed something in an aisle is a hard one to win that argument with.

So the risk was never that the knowledge left in Chandhu Nair's head. Most of it did not. The risk is that the thing which kept producing the knowledge just lost the person who protected it, and a reorganization is where practices like that quietly stop. Nobody announces it. There is no meeting where somebody decides the AI team should stop walking the floor. The team just gets a new manager with a different set of priorities, and a year later nobody can remember why voice got fixed first.

Nobody has announced who inherits that at Lowe's, and nobody asked on the call. These things take time. But it is the honest question, and the whole story got written as an acquisition without anyone writing the subtraction.

That is the room I started in. My documentation was complete and it transferred nothing, because reading it did not make you able to write a rule. What made the knowledge transferable was not a better document. It was a system two people had to walk through together, which meant the reason had to be said out loud to somebody who could question it.

The question to take to work

Forbes ran a piece the same week Target made the announcement, asking who inherits your employee's AI when they resign. It is the right question and it has the wrong noun in it. The AI is the part that stays. By Monday somebody has to own it, and in a lot of companies nobody does.

And most companies are not in Lowe's position. Lowe's built the rare thing, a team that actually stands where the work happens. If you have not built that, there is no team to shuffle, because the practice was never a practice. It was one person who happened to care.

So here is what I would actually do this week.

Pick the AI capability your business already depends on. Not the pilot. The one that would be noticed if it stopped. Then get the people who built it in a room, two of you or the whole team, and ask five things about the habit that made it work.

What did somebody see, where the work actually happens, that changed how this got built?

What is the habit that produced that, and when did anyone last do it?

Of everything your logs told you last quarter, which one can nobody explain?

Who can explain it, by name, and whose budget is that person on?

Who walks this floor next, and on what date?

Write the answers down, and put a person against the last one. Until you do, the thing that keeps your AI honest is running on somebody's goodwill.

Goodwill does not appear on a resourcing spreadsheet.

The risk was never that your AI leader leaves. It is that your logs will tell you what your people did and never once tell you why. The why only exists where somebody is standing, and the practice that puts them there has no ticket, no line item, and nobody obliged to defend it in the meeting where it gets cut.

One last number. What Lowe's said this tool was for was a five week associate sounding like a five year one. What it has reported, fifteen months on, is how many times somebody asked it something. Volume is what a log can count by itself. Whether anybody got better at the job is the thing somebody has to go and see.

So here is my prediction, and you can hold me to it. The thing I expect to quietly stop is not the tool. It is the walking. If a year from now anyone at Lowe's can still say who was on the floor and when they were last there, the rule outlived the man who made it, and I was wrong about how fragile this is. I would rather be wrong. Ask me in August 2027 and I will tell you what I found.

Nobody at Lowe's has said who is walking the floor. Go ask who is walking yours.

I made you the one-pager. Five boxes. And before you tell me a document will not fix this, you are right, mine did not. The page is not the point. The point is that somebody has to say the reason out loud to a person who can question it, and then somebody has to put a name and a number against the thing that produced it. Get the Handover Sheet

References

1. Target Strengthens AI and UX Capabilities to Power Its Next Chapter of Growth. Target Corporate, August 11, 2026.
2. Target appoints its first chief AI officer as big retailers bet on AI. CNBC, August 11, 2026.
3. Target names its first chief AI officer in a bid to remake the shopping experience. Fast Company, August 11, 2026.
4. Chandhu Nair, Senior Vice President, Stores, Data, Artificial Intelligence and Innovation. Lowe's Corporate, archived February 16, 2026.
5. Meet Mylow: Lowe's Unveils First Ever AI-Powered Virtual Home Improvement Assistant. Lowe's Corporate, March 5, 2025.
6. Lowe's Deploys First At-Scale AI Assistant for Retail Associates. Lowe's Corporate, May 5, 2025.
7. Lowe's (LOW) Q1 2026 Earnings Call Transcript. The Motley Fool, May 20, 2026.
8. Lowe's leverages AI to power home improvement retail. OpenAI, May 5, 2025.
9. Lowe's SVP of AI and the Inside Story of its AI Transformation. The AI Innovator, 2026.
10. From Store Floor to Strategy Room: Here's How Lowe's Scales AI. CDO Magazine, 2026.
11. Lowe's links end-user feedback loop to AI tool improvements. CIO Dive, 2026.
12. 2026 AI & Data Leadership Executive Benchmark Survey. Data & AI Leadership Exchange, 2026.
13. Your Employee Just Resigned. Who Inherits Their AI? Forbes Technology Council, August 11, 2026.
14. Lowe's Companies (LOW) Q2 2026 Earnings Call Transcript. August 19, 2026. Marvin Ellison on more than 25 million questions from customers and associates since inception.

About the author

I'm Ben. I write Enterprise Field Notes, and by day I'm COO at Swa. Before that I was Global Director of Site Reliability Engineering at a Fortune 500 retailer, running reliability, data protection, and database operations at global scale.

Header artwork generated with Swa.

Read more of Ben's Enterprise Field Notes at benpickett.com.

© 2026 Ben Pickett · Enterprise Field Notes