The Barking Cat problem

Hello. I am Steve Webster. I have spent twenty-seven years inside retail businesses, accountable for exactly the kind of investments this talk is about. For the next half hour or so, I want to talk about why so much retail investment never quite pays back.

If we were in the same room, this is the point where I would ask you to stand up. You are watching a recording, so I will settle for something harder.

Think of the last big investment your business made. The replatform. The new agency. The AI tool. The one in the board pack.

Now: did it pay back the way the business case said it would?

Be honest. Nobody is watching you watch this. In a room, I ask people to stay standing if theirs did. Hold on to your own answer. We will come back to it.

Here is the point I want to plant early. Very few of us bought a bad technology. We bought a good one, and asked it to work inside a business that was built to do something else. That is what this talk is about.

First, a word about the name. My practice is called Barking Cat, and it has a chairman. This is Casper, giving the board his full attention.

He sits above the operation. He declares loudly and often. He is paid in freeze-dried chicken hearts, which by the standards of the advisory industry is a reasonable fee. He does not attend client meetings.

In all the time he has chaired the business, he has never once barked. I have stopped asking. Which, it turns out, is the entire point of this talk.

For the avoidance of doubt, here is the Chairman addressing the board from the top shelf. It looks promising. I have had it independently verified. It is not a bark.

Which brings me to a Nobel laureate.

In February 1973, Milton Friedman wrote a Newsweek column called Barking Cats.

A reader had written to him. The reader agreed that a regulator was performing badly, but thought the answer was to staff it with better people. Friedman replied, in effect: what would you think of someone who said they would like a cat, provided it barked?

His argument was that the behaviour was not an accident, and not a staffing mistake. It followed from how the thing was constituted, in the same way a meow follows from the constitution of a cat. You cannot assign characteristics at will.

To be fair to Friedman, he was writing about a government regulator, not a retailer. I have borrowed the image, with respect, because I have spent twenty-seven years watching retail businesses ask for a cat that barks.

Expecting an institution to behave against its nature is expecting a cat that barks. Half a century after Friedman, we still do the same to our businesses.

We demand digital outcomes our operating model cannot produce by nature. Then we wonder why the investment never quite pays back.

And when it does not pay back, we do what Friedman’s reader did. We blame the people. We change the agency. We change the platform. We hire a new ecommerce director, which, speaking as a former ecommerce director, I can confirm is a very popular option.

What we rarely change is the cat.

Think back to your answer a few minutes ago.

In my experience, the technology in those business cases mostly worked. The platform launched. The tool did what it said it would. What did not exist was the business the case was written for. One where everybody wants the investment to work. Where the target in the case measures the benefit. Where every team pulls the same way. Where somebody is free to act on it. Where the people exist who know how to run it. And where the saving lands on a line that someone looks after.

And the case assumed a customer who would arrive through the channel we built, and buy what we planned.

That is the barking cat, written into a business case.

So there are two cats in every business case.

The first is the business: the one the case assumed you had. Its incentives, its measures, how its teams’ targets fit together, who is allowed to decide what, whether it has the skills, and the plumbing underneath.

The second is the customer: the one the case assumed would behave. We ask the customer to change how they buy to suit how we sell. They decline. Politely, mostly.

And then there is AI, which is the first investment I have seen that makes both assumptions at once.

Where I name a business, I have cleared it. Where I do not, you will understand why.

Cat one: the business. I want to start with the evidence, because the evidence says this is not a problem of a few unlucky retailers. Then I will take you through six assumptions most business cases make about the business, with one example of each.

McKinsey published its annual State of AI survey at the end of August. Over seventeen hundred respondents.

Eighty per cent say AI has made them personally more productive. Thirty-seven per cent can point to any impact on EBIT at all, which is about where it was last year.

And six per cent. Six per cent, the ones McKinsey calls high performers, attribute five per cent or more of their EBIT to AI.

So the hours are being saved. They are just not arriving in the profit.

My reading, and it is my reading rather than McKinsey’s: an hour saved is not a pound saved. It only becomes a pound when someone redesigns the work around it, changes who does what, or stops paying for something. If nothing else in the business changes, the hour is absorbed. The cat goes on being a cat.

So where does the return go? In my experience, it usually comes back to six assumptions the business case made about the business.

Friedman’s word was constitution. The behaviour follows from how the thing is built.

Inside most business cases I have read, written or signed, there are six assumptions about how the business is built. They are rarely written down, which is exactly why they are rarely tested.

Incentives: the case assumes everyone affected wants it to work. Measures: it assumes the target measures the benefit. Alignment: it assumes every team is pulling the same way. Decision rights: it assumes the business can act on it. Capability: it assumes the business knows how to run it. Plumbing: it assumes the benefit will land where you expect.

When they hold, quite ordinary technology pays back. When they do not, the best technology in the world becomes a very expensive meow.

This is not an exhaustive list. These six are where I most often see a business case and the business part company. One at a time, with one example of each. As we go, keep a note of which one sounds most like your business. I am going to ask.

The first assumption: everyone affected wants the investment to work.

Usually nobody objects to the technology itself. What people respond to is what it does to them. People are rewarded for what they control, and an investment that shifts sales or credit from one team to another leaves some of them worse off. Support fades. Not out of bad faith. They are responding sensibly to how they are paid.

At one retailer, store colleagues were paid commission on what went through their own till, and could do nothing to influence the website. Every year at conference, store managers stood up and said ecommerce was stealing their sales. Agreeing how stores would share in online took a year and three costed options.

Why it matters to a board: adoption slows, and the return arrives later than the business case said. That delay is rarely priced in.

For that whole year, the business had an online channel its own store teams saw as a rival. And by the end of it, online was growing while store trade was flat, in a declining market. The channel the stores were fighting was the growth.

The second assumption: the target in the business case measures the benefit.

Then a team is managed on that target, and it does the sensible thing. It improves the number, often by the quickest route available. The quickest route does not always help profit.

Economists have a name for this: Goodhart’s law, after Charles Goodhart of the Bank of England. The version everyone quotes, which is the anthropologist Marilyn Strathern’s phrasing rather than his, is: when a measure becomes a target, it ceases to be a good measure.

I inherited a paid search agency that reported, proudly, a return on ad spend well above target. A target I had not set, I should add. In the same period, it was missing the sales target and not spending the full budget. Both were true at once. The easiest way to raise a return-on-spend ratio is to spend less, on the people who were going to buy anyway.

Why it matters to a board: success can be reported that the profit line does not show, and the business keeps funding it. The board has its own version of this.

It is usually on the front page of the pack.

At one retailer, we pulled paid media back by thirty per cent. Topline revenue fell by roughly twenty per cent. A board that reads only the top line calls that a disaster.

But the bottom line was stronger without that spend. We had been buying revenue that cost more than it made.

If the measure on the front page of the board pack is revenue growth, the business will keep buying unprofitable revenue, and it will feel like success every single month. Nobody in that picture is failing. The constitution is doing its job.

The third assumption sits between incentives and measures: every team is pulling the same way.

Each team is measured and rewarded on its own KPIs, and each set makes sense on its own. Put them side by side and they can send people in different directions. Nobody is being difficult. Everybody is doing what they were asked to do.

At one business, Marketing was targeted on traffic. It bought cheap display traffic, hit the target, and redeployed the budget it saved. That traffic bounced eighty per cent of the time. Conversion fell, and the ecommerce team, which was judged on conversion, got the blame.

I explained it to the CEO this way. The Marketing Director has brought five coaches of customers to the store, and opened the doors on one of them. The other four are sitting in the car park.

Why it matters to a board: every team can hit its own target while the result the business actually needs is missed. And because no one has done anything wrong, no one is looking for the problem.

Did the explanation change everything? No. The CEO still looked at conversion rate. But from then on, he asked for my qualification first. Which I chose to regard as a victory.

The fourth assumption: the business can act on it.

Most of the value in retail technology comes from someone being free to change something, quickly: a price, a promotion, a page, a range. So the question is who is allowed to say yes, and how many others have to agree first.

At a large retailer, we reviewed machine-learning dynamic pricing. The case was made, and it was approved. It never went live. Daily prices needed digital price labels in the stores, which were still being rolled out, and an upgrade to the middleware and the ERP. Marketing pushed back, because the offline media was built around fixed price points. And Commercial was not convinced a system could improve on its process and experience. Every objection was reasonable from where that person sat. Together, they stopped an approved project.

Why it matters to a board: the investment is paid for but under-used, and the payback period in the business case keeps stretching.

It can also go the other way, where a decision is made too quickly by someone without the full picture. That one can cost you a launch day. But that is a story for another day.

The fifth assumption: the business knows how to run it.

A business case usually prices the technology and the implementation partner. It rarely asks whether the business has the people who know how to choose it, build it, run it and keep improving it. Often those skills are hired after the decision, or left to the supplier.

When I arrived at one retailer, a replatform was already under way. The platform being replaced was custom built, with no content management, no enterprise search and no merchandising tools, so the case for moving was real. But my view is that it should have waited until someone who had done it before was in the business. My first ninety days went on getting it back on track.

Why it matters to a board: the early months go on recovery rather than return. And without that experience inside the business, the supplier sets the pace and the priorities.

The sixth assumption: the benefit will land where you expect.

Data, stock and money flow where the systems and the processes send them, which is not always where the org chart says. A saving lands on whichever line it happens to land on. If no one owns that line, it is hard to find later, and nobody notices when it does not arrive.

A CFO convinced me to implement a new payment service, routing each transaction to the customer’s own issuing bank to cut fees. Finance negotiated the rates with the banks. The business case had a saving of three million pounds a year in it. We delivered the work. The saving never materialised, and it certainly never appeared in the ecommerce P&L.

I agreed to it, so this is not a story about a CFO. It is a story about a question nobody asked: when this saving happens, which line does it land on, and who looks after that line?

Why it matters to a board: delivering a project and banking its benefit are two different events. Most businesses track the first closely and assume the second.

So, which of the six do you see most often in your own business?

Incentives. Measures. Alignment. Decision rights. Capability. Plumbing.

Pause the video for a moment, if you like, and pick one. Just one. Hold on to it, because it is the one I will ask you to take away at the end.

Six assumptions, and not one of them is about the technology.

Each of those businesses got exactly what it was built to produce. None of them bought bad technology. Some of them bought very good technology.

That is the first cat. Now for the one we cannot redesign at all.

The second cat is the customer: the one the business case assumed would behave. Pictured: the Chairman, considering whether to come out. He will decide. You will not.

Twenty years ago, in June 2006, Bryan and Jeffrey Eisenberg put the idea on the cover of a book: Waiting for Your Cat to Bark? Their premise was that customers behave far more like cats than like Pavlov’s dogs. Ring the bell all you like. They will come when they want, for their own reasons, by their own route.

Twenty years on, retail still plans as though the customer will change how they buy to suit how we sell. So, as with the business, let me show you the assumptions.

The business case makes assumptions about the customer too. Five of them, in my experience, and they are even less likely to be written down than the six about the business.

The job: it assumes the customer wants what we sell. Who: it assumes the customer is who we think they are. The route: it assumes they will take the route we designed. Habit: it assumes they will change their habits to suit us. What we teach: it assumes they learn nothing from what we do.

Every one of these assumptions asks the cat to behave like a dog.

The first assumption: the customer wants what we sell.

Customers do not really buy products. They hire them to do a job, and the job is often emotional as well as functional. My favourite example comes from a furniture retailer.

The bright colours never sold. Customers bought the adjacent, safe colours. So, quite logically, the bold colours were removed from the range.

Sales fell sharply. When the merchandising team reviewed the results afterwards, they worked out why. The bright sofas had been doing the work. They drew customers in, got them engaged, and then the customer talked themselves down to a safe colour. Take away the thing nobody bought, and fewer people bought anything.

The sales report records what the customer bought. It does not record what the customer needed to see before deciding to buy. The product that sells is not always the product that creates the sale.

For a board, the lesson is bigger than sofas. Every line review, every range rationalisation, every “cut the tail” exercise judges products one at a time. The customer does not experience them one at a time. Optimise the range line by line, and you can quietly take the proposition apart.

The second assumption: the customer is who we think they are.

A fashion retailer told everyone its customer was eighteen to twenty-eight, and it competed for that customer. After six months without success, I took the customer database apart myself. The customer was thirty-six.

By then, the business had spent half a year talking to someone who was not its customer, and it had damaged its relationship with the customers it actually had. It could not get them back.

Why it matters to a board: every number in the business case, the conversion, the average order, the lifetime value, was modelled on a customer who did not exist. Before you decide what the customer wants, check who they are, on your own data rather than the brief.

The third assumption: the customer takes the route we designed.

The plan says online, or stores, or now AI. The customer uses all of them, each for a different job. Online in Great Britain went from nineteen per cent of retail sales before the pandemic to nearly thirty-eight at the lockdown peak, and settled at twenty-eight point three in September. The peak did not become the new normal. The customer took what they wanted from it.

And they still browse. dentsu surveyed two thousand UK adults this summer. Eighty-three per cent discover products while browsing in store, and around eighty per cent enjoy it.

Most of the decision happens before the till and after it: finding, comparing, checking the size and the returns policy, then living with it, returning it, coming back. The sales report records one step of that journey. The customer takes the others wherever they like.

The fourth assumption: the customer will change their habits to suit us.

Every new app, every new checkout, every loyalty scheme, and now every AI agent, has a line in its business case that quietly assumes the customer will start doing something new. Anyone who has bought a cat a new bed, and watched it sleep in the box the bed came in, knows how that goes.

Customers already have habits that work for them. The new thing has to be clearly easier, or clearly more rewarding, before the habit gives way. And there is usually some anxiety to overcome first: will this work, will I get it wrong, can I trust it?

Here is where trust sits for AI buying, today. ACI Worldwide and YouGov asked over two thousand UK adults in June. Nineteen per cent would trust an AI assistant with their everyday purchases. Fifty-five per cent would trust a human expert. That is stated intention, from a company that sells payments technology, so treat it with care. But if your AI business case needs customers to hand over the buying, that is the gap it has to close.

The fifth assumption, and the one I would most like boards to think about: the customer learns nothing from what we do.

Customers learn very quickly what we reward. If we reward waiting, they wait. If we put the price down every few weeks, they learn the full price is a suggestion.

You cannot teach a cat to bark. But you can absolutely teach it to come running at the sound of the tin opener.

At one retailer, the business had bought a job lot of stock at a distressed price, and traded it for three years. Customers loved it. It also trained them to expect aggressive prices, on margins the business could never repeat. When the stock had sold through, the customers’ expectations were still there, and the margin that had paid for them was not.

Why it matters to a board: every promotion is a lesson. A business case for a promotion, a price cut or a loyalty scheme counts what it sells this year. It rarely asks what it teaches the customer to expect next year.

Five assumptions about the customer. Here is what happens when a business finally looks at the evidence.

In 2017, at Harveys and Bensons, I was given the post-lunch slot at three conferences, which anyone who has spoken at a conference will tell you is the graveyard shift. Glasgow, Bolton and London. Four hundred store managers over three days.

I opened the way this talk opens in a room. Everyone stand up. Stay standing if you think ecommerce is stealing your sales. Over the three days, three hundred stayed standing.

Then I showed them what online was doing for their stores. Customers discovering us online and then visiting a store. The higher conversion of customers who researched online and bought in store. The investment in making store locations visible. The VR and AR work. The tools that let a customer see a product in store.

And I closed with the same instruction. Stay standing if you still think ecommerce is stealing your sales. Over the three days, fewer than fifteen stayed standing.

The customer had never been confused about which channel they were using. The business had been. Nothing about the customer changed during that session. What changed was the story the business told itself about the customer.

Which brings us to the newest cat.

With AI, we are doing something rather ambitious. We are asking the business to change its decision rights and its plumbing so the AI can work. And we are asking the customer to hand over some of the buying to an agent. Both cats, at once.

So the question for a board is not whether AI matters. It does. The question is where it pays back, and what has to change for it to do so.

IMRG asked UK retailers where they currently use AI.

Nearly eighty per cent use it to write content: product descriptions, social posts. More than half use it to make images. Six point eight per cent use it for dynamic pricing. Six point eight per cent for one-to-one personalisation.

My reading of this, and it is a reading: the high bars are the jobs that slot into the business exactly as it is already built. Somebody used to write the product description. Now a model drafts it. Nobody’s incentives change, nobody’s decision rights change, and the plumbing is untouched.

The low bars are the jobs that would change the constitution. Dynamic pricing changes who decides the price, which you have already heard about. Personalisation needs clean customer and product data flowing in real time, which is plumbing.

That may be part of why McKinsey finds the hours more easily than the EBIT. So far, much of the industry is using AI to make the cat meow more efficiently.

So where does AI pay back? I find it helps to sort AI investments into three kinds, because each pays back differently.

Low change: content, imagery, drafting replies for the service team. This is where most retailers are, the eighty per cent bar. It is quick, because almost nothing else in the business has to change. It saves hours. On its own, it rarely moves the P&L, for the reason on the McKinsey slide: an hour saved is not a pound saved. Do it. Just do not put it in the business case as EBIT.

High change: pricing, personalisation, availability. This is where the bigger return is, and it is the six point eight per cent bar. It pays back only if the business changes around it, and all six assumptions apply. Who is rewarded for using it? Is it judged on profit? Do the teams share its target? Can someone act on what it recommends, quickly? Does the business have the skills to run it, or will that sit with the vendor? Is the data complete and owned? The pricing model I described earlier did not fail on technology. It failed on shelf labels, systems and a Commercial team that was not convinced.

Customer readiness: whether an AI agent, acting for a customer, can read your product data, understand it and cite it. Today, that is a small share of traffic. It decides whether you are in the answer at all. And this is the one the customer decides, which brings us back to the second cat.

None of this says AI is overrated. The return is real. It goes to the businesses that change what needs to change around it.

IMRG measured it across 168 UK retailers. AI assistants sent 0.39 per cent of site traffic. Not thirty-nine per cent. Nought point three nine. It converts at about 2.3 per cent, close to organic search at 2.2. Customers are using AI on their own terms, mostly to research and compare.

It is small. What matters more is how it shops. An AI assistant comparing products works from what it can read: product data, price, availability, and what others publish about you. That is plumbing again, seen from the customer’s side.

At one retailer, the names, dimensions and fabrics a customer needs before they buy lived in over fifty spreadsheets. The team named it their largest challenge. The data was all there. But data spread across fifty spreadsheets is hard to keep consistent, hard to publish in a form a machine can read, and hard for anyone, human or agent, to rely on.

That opens a wider conversation about brand. Whether an assistant names you, and how it describes you, depends partly on your own data and partly on what reviews, press and comparison sites say about you. Your brand is increasingly what can be read about you.

Three things I believe about AI in retail, which follow from everything you have just seen.

One. AI services existing demand. It does not create it. Nobody asks an assistant to find them a product they have no interest in. That 0.39 per cent converts at about the same rate as organic search: real demand, arriving by a new route. Treat it as a channel to serve well, not a growth engine that will fill the gap in your plan.

Two. Agents expose commodity, not brand. An agent comparing products reads your product data, your price and your availability. It does not feel your brand campaign. If what makes you different lives only in the advertising, an agent will not find it. If it lives in the product, the data and the promise, it will.

Three. Adoption is an operating-model question, not a technology purchase. Which is this whole talk in one line. You can buy the AI. You cannot buy the incentives, the measures, the alignment between teams, the decision rights, the skills and the plumbing that let it pay back. Those you have to change.

Let me show you how I would use all this. This is an illustrative business, not a client: a fifty million pound apparel retailer with stores and a website. The figures are invented.

Here is the board pack. Online traffic up twenty per cent. Online sales up five. The average discount is deeper than last year, the returns rate is higher, and gross margin is lower. And on the table is a proposal for a new ecommerce platform, because the team says the current one is holding them back.

Would you approve it? Decide before I go on.

Fair enough, whichever way you went. It might even be the right answer. But I would not want to approve it yet, and here is why.

Six questions about the business, one for each assumption, and beside each one, what readiness looks like. They work on any investment. Let me run them over this board pack.

Incentives: are the people affected rewarded for this working? If the business rewards sales growth while margin falls, the platform will be asked to deliver more of the same. Readiness: agree how the teams affected share in the benefit before launch, not a year after.

Measures: does the measure of success move with profit? Traffic is up twenty per cent and sales five. What is the extra traffic actually doing? Readiness: judge the investment on contribution, not on activity.

Alignment: do marketing, trading and ecommerce share one goal here, or three? Readiness: one shared target for this investment.

Decision rights: could the range and the discounting be improved this season, on the current platform? Readiness: name who can act on it, and clear the blockers before you buy.

Capability: has anyone in the team led a replatform before, and who will run the new one once the supplier has gone? Readiness: put that experience in the business before the decision.

Plumbing: if a new platform delivers, which line moves? Readiness: name the line, and the person who owns it, as a condition of approval.

None of this is exotic. It is the work that usually happens a year after the money is spent, done before it instead.

And four questions about the customer.

The job: what job does the customer hire us for, and does this investment serve it? A new platform serves the job only if the job was being held back by the platform.

Who: is the customer in this board pack the one in the brief, or the one in the data?

Route and habit: does the case need customers to start or stop doing something? If the return depends on customers changing how they buy, that is the riskiest line in the case.

What we teach: what has this business been teaching its customers? Look at the pack again. Discounts deeper than last year, returns higher, margin lower. That is a business teaching its customers to wait for the price, and a new platform will teach the same lesson faster.

So on these numbers, I would look at trading before technology. Deeper discounts, more returns and lower margin are rarely fixed by a platform. The platform may still be needed. It is just not the first question.

Friedman’s point was that behaviour follows from how an institution is built. You do not get a bark by asking nicely, by hiring a better cat, or by buying it a very expensive collar.

So there are two honest options. Write the business case for the business you actually have, and the customers you actually have, and expect what they will deliver. Or change how the business is built before you spend the money: its incentives, its measures, how its teams’ targets fit together, who can decide, the skills it has, and where the benefit lands. Both are respectable. Waiting for the bark is the only option that does not work.

One request. Remember the one you picked a few minutes ago. Take that question to your next investment decision. Just one. If the answer is uncomfortable, that is usually where the return is hiding.

Thank you for watching. The Chairman has been consulted, and as you can see, he has nothing to add. Which is the most cat thing he could possibly have said.