The Third Decision: Fixing Your Operating Model Before You Buy AI with Anca Pintilie

John Emmerson: Hello everybody and welcome to another aibl Leadership Series interview. Today I am absolutely delighted to be joined by Anca Pintilie, who is a marketing expert. Welcome to the podcast, how are you?

Anca Pintilie: Great, thank you.

John Emmerson: Excellent. So Anca builds marketing functions, operating models, and various other bits and pieces, and has worked around the tech world for her whole career. You spent the last six years at Amazon, leading marketing and brand across EMEA. You scaled a team sixfold. You managed a portfolio of over 1,000 campaigns, and you oversaw an enormous budget in the hundreds of millions. You also ran a global transformation mandate, if that is correct.

John Emmerson: Before that you were at Revolut, Oracle, and various agency roles. So we are delighted to have you here. The thing that interested me most about the bio you sent was that you trained as an economist. That is interesting in itself, but you also competed as a mathematics Olympian, which I had never heard of before. I would love to hear more about that.

Anca Pintilie: Thank you. It was a long time ago, which of course ages me by saying it, but this was in school. Mathematics was my favourite subject, because there was an exploratory element to it, but there was also some certainty at the end. It was a bit binary. You either cracked it or you did not. Quite simple, as opposed to other subjects where you have to write three pages to describe a character.

Anca Pintilie: No wonder we now write very long articles describing certain things all over LinkedIn. So there you go, it comes from school. You compete against your schoolmates, then at county level, then at country level. I thought I was absolutely awesome until I hit a certain level and found myself at the bottom of the list. Clearly that was not going to be a long-term career. It is also quite an isolated thing that you do.

Anca Pintilie: I wanted to go to university to study mathematics. My whole family said, and you are going to be a teacher. Many years ago we really did not have a big vision of where mathematics could take you, as opposed to today where things are a little different. So I went to the next best thing, and I am very happy with it, an economist, where you use a lot of logic. You just look at markets rather than pure equations.

John Emmerson: I do not know what your education was like, but I was also a bit of a maths geek at school. I do not think it was properly explained to me back then that maths is actually the building block of the world. One of my daughters is moving from primary school up to secondary school. We visited a potential new school last week, and one of the teachers showing us around was an English teacher. Maya said she did not really like English and preferred maths. She argued the point, saying numbers are the building blocks of the world, and he reluctantly agreed.

Anca Pintilie: Probably not the first time he had that conversation though.

John Emmerson: Well, indeed. And I am not dismissing the importance of English language or literature, both very important. So look, as I said, delighted to have you here. Obviously here at aibl, we are really interested in AI adoption. That is the whole brand. So I would love to know, from your time running marketing operations, what is the first thing you check to know if a team is ready for AI, rather than just going out to buy it?

Anca Pintilie: It is actually quite simple. Are they able to describe how work actually happens within their function? By that I mean, how do you make decisions? Which of those decisions triggers spend? What data are you using? Where is it coming from? Does it have the same refresh rate? That sounds fancy, but let me tell you this. When I joined my team, and this is nobody’s fault, we were just not set up that way at the beginning, and then we all fixed it together.

Anca Pintilie: We pulled data from three different sources. Those were self-reported by different vendors and partners working with us, and everybody reported at a different time of the week. So the real question is, do you take the right decisions at the right time? And what is the process that supports the whole decision-making chain within your function?

Anca Pintilie: Is everything connected to a business outcome? Or does the marketing brief come in with a connection to the business that is a little loose, or that was defined seven years ago, maybe pre-pandemic, because somebody said, when this happens then we will do marketing? Maybe nobody challenged it in the past few years. This is important. And the way work happens never sits in an org chart.

Anca Pintilie: Nowadays we reorganise every other day, so any org chart is basically obsolete by the time everybody occupies their new roles. It is one of those things you have to have, but it is never a reflection of the truth. And your operating model, unless you have mapped it in the past month, is probably also obsolete. Or somebody wrote it on a slide just to present it to management. So, can you describe how work happens with accuracy? That is the readiness test.

Anca Pintilie: Because buying AI is the easiest thing. And nowadays building AI is quite easy too, if you want to build some AI workflows. But if you are not careful, you are going to automate whatever dysfunction you have within your function. This reminds me of one of those tests that were all over Instagram and TikTok a few years ago, where somebody would go up to a person and say, can I see what is in your bag? And they would start pulling out all these random objects, sometimes something you had not cleaned out for three weeks. It is exactly like that. Somebody opens your bag today, and now it is on a world stage for everyone to see.

John Emmerson: I remember this trend.

Anca Pintilie: It is the same thing but with AI and businesses.

John Emmerson: We find this. Everyone says the same thing, and it is something we harp on about a lot through our channels. Do not just go and buy an AI tool. You really need to figure out what your problem, challenge, or opportunity is and how you will use it. You need to get the infrastructure and the people lined up first. Lots of businesses really struggle with that. Is that your experience? Did you find there was a struggle there?

Anca Pintilie: Absolutely. And the bigger the business, the more complex, and probably the bigger the struggle. If you had generous budgets in the past with very ambitious targets, then most likely your operational model is all over the place, because you had to move fast and execute and go to market quickly. The way you fix it is by running an in-depth discovery, which many years ago I would have said, just get on with it, do the discovery, it will take a month, then let us get to the execution.

Anca Pintilie: But I was surprised. In my previous role, working with about six marketing teams, I thought, let us plan a maximum of two months of discovery across all of them. It took my team five to six months to actually get to the bottom of the truth. The questions are the easy part. It is not that complicated to ask. What is complicated is how you actually get to the truth.

Anca Pintilie: When I say the questions, I mean, what decisions do you make before you get a sign-off on a specific budget or campaign? What knowledge exists within this function that maybe only John and Alex remember? Why do they remember it? Where is it codified? Where is it documented? Can it be replicated? When one of them goes on leave, do we actually post certain things?

Anca Pintilie: Another interesting thing is you need to figure out exactly what metric changes behaviour and which one just gets reported. Big companies, big functions, big budgets. I used to call this the reporting festival, because you had your reporting on Monday, then the creative review on Tuesday. It is a three or four-day festival every day, where you end up discussing metrics. But which one actually drives a change of mind or a decision?

Anca Pintilie: For example, your click-through rate gets reported, but what changes your budget discussion is actually incrementality. Do you measure that? If you do not, just own up to the fact that you do not measure it. I had reports in my function where we went in with black boxes and said, if we could measure this, here is what the conclusion would be. Here are the decisions we could make. And here are the dependencies on why we cannot measure it today.

Anca Pintilie: And I would say, dear management sitting here on the call, can you please help us unblock this? Because if you want to get to this decision that will impact budget and even revenue moving forward, then we need you to do this. Sometimes it is good to put that into an equation, going back to equations, even though you have variables you do not know yet. It gives people a clear view of where you are going to get to.

Anca Pintilie: Another question I love is, which activities is the team doing that survived because they are measurable but are no longer valuable? You will find there are a lot of those. For example, we had a crisis at some point and added an extra approval step. Now we have been dragging that approval step with us for a year, which is probably half an hour extra a week for everything. Half an hour you could be using for something else.

Anca Pintilie: One of my favourites is brand trackers. Great. Most of the time they are flawed, and they never actually influence a decision, but they satisfy a lot of curiosity. This is not me advocating against brand trackers. It is, are you doing it across the territories that are actually the most painful for you? Or are we just satisfying everyone’s curiosity across most territories, with a thin layer of measurement that will never be fully complete?

Anca Pintilie: Going back to the reporting festivals, and you said you measured a lot of campaigns, the question is how many KPIs are we measuring across all these campaigns, and how many of them actually drive decisions? These are a few of the questions. And to get to the bottom of it, you need to spend a lot of time with those teams. You need to fly over, go in a room, put it on a whiteboard.

Anca Pintilie: My team and I, across six teams, ended up with about 72 calls and really did not get to the bottom of it until we flew over and said, everybody, can you just sit in one room? We are going to draw it out on a whiteboard. When people see it in that analogue world, on a whiteboard, they start realising what they are doing every day.

John Emmerson: There are so many things there I wanted to come back on. Firstly, the reporting festival. I think we have all been there, when you work for businesses and spend your entire life reporting rather than doing any work. One of the benefits of even the most simple use cases of AI is to automate all of that. People should not be spending days producing PowerPoint decks to present to others in their team. Those days should, in theory, be gone.

John Emmerson: The amount of time though, you said five to six months. There will be people listening to this going, I have not got five or six months to figure this stuff out. But what that makes me think is that generally in life you need to invest a big piece of time at the start to plan it out properly for it to be effective, whether it is a project or anything else. If you do not do that first step properly, then the outputs you can expect from adopting an AI tool are not going to be as good. Would you agree with that?

Anca Pintilie: Absolutely. And I felt it myself. I know what it means to say, this tool works in North America, we are just going to deploy it in Mexico. Great. The Mexico team said, but we have a completely different mix of channels. We take decisions differently. Our data is all over the place. That does not work. But guess what, you have already invested a lot of hours. Either you bought it or you built it. So now it is not fit for purpose.

Anca Pintilie: What happens when you deploy the wrong tool is that not only can you automate that dysfunction, but the team starts building ghost processes to manage the inappropriate tool you just brought in. So it becomes even messier. Seven months ago, in December, I spoke at a conference that was more of a tech summit. Everybody talked about AI. We had people joining with PhDs in AI. It was a fantastic conversation about the technical aspects of AI in marketing.

Anca Pintilie: About a month ago, I met a similar crowd and everybody was talking about operating models. Because whatever you bought six or seven months ago, without inspecting how you actually solve for the business, is now a bigger mess. Nobody understands it. Now you have the legacy operating model, some AI plugged into it, and ghost processes on top, and everybody is miserable. I am exaggerating and dramatising, but it can be like this.

Anca Pintilie: People now go back to basics. Those seven months, which now seem like a long time, make five months of discovery look reasonable. Because at the end of those five months, you might realise that for a chunk of your business, all you need is a simple deterministic model. Once you get into AI, AI costs money, but managing AI also means managing sophistication, which is a completely different game. Maybe all you need is to clean up your brief. Maybe your compliance and governance requirements could be embedded within the brief as a fixed part of it. That might solve a lot of things downstream, and you do not need AI for that.

John Emmerson: I do not think you are exaggerating, for two reasons. Firstly, the conversations we have with people echo exactly what you said, across different businesses. Amazon is a huge business, but even businesses at the smaller end of our spectrum, which you would consider SMEs, are facing similar challenges.

John Emmerson: The other thing is that there is so much research about how much ROI people are getting from their AI spend. Adoption levels have skyrocketed this year, as you would expect, but the ROI is still quite small for the vast majority of businesses, for the reasons you just pointed out. People have not done it properly, rolled it out properly, or thought about it properly. As a result they say, we have got this brand new shiny tool and we have spent hundreds, thousands, or hundreds of thousands on it, yet we are not seeing any ROI. Well, hold on. Have you actually embedded it properly? And have you adopted it correctly in your organisation?

John Emmerson: This brings me to another point, the human angle. You were talking about the questions you ask teams to identify their working processes and the problems they want to solve. There is an element of digging into the honesty of the responses you get. Trust is so important in every part of business, and as you go into larger organisations that becomes more diluted, just by the nature of having thousands of people. So how do you really get to the bottom of it? And how do you ensure people are given the training and development support they need to use the tools properly? It is that human element I always keep coming back to.

Anca Pintilie: It does make sense. I will start answering and you tell me if we are on the right track. When you do that discovery, if you do not surface any contradictions and everybody is reciting from the same slide, then that is your problem. That is the signal. It means you need to go deeper, because complexity pays salaries, and that is normal. Somebody who ran a process for five years is going to be attached to it. They might have a love-hate relationship with it, because they improved it, or got a promotion out of it.

Anca Pintilie: So you need to get to the bottom of why you are doing this. Is it actually relevant? Nobody has any ill intent here. It is because we work in a constant hamster wheel sometimes. Once you ask people to stop and think, you need to give them the space. There needs to be a safe space, and there need to be many stages to this. Getting on a call and saying, tell me how you go to market and how you plan your budget, warms them up to the process and helps you understand the person in front of you. But then you are probably going to have two or three more conversations.

Anca Pintilie: I was talking to a CRO last week and she told me something interesting. She said, I go into these companies and look at their goals, and they say, we need growth. Then I look at the other functions, their budgets and headcount, and if that is growth, why are we spending 60% on something else rather than growth? What is your budget in the other direction? So it is those things. Do you see contradictions in what people say, what they do, the tools they have, and their understanding of the data?

Anca Pintilie: Adoption is an interesting thing, because we need to think about adoption versus usage in AI. Sometimes you have the usage but not the adoption. That is why I love when you say embedding. Embedding means you do not work around your technology. It is not another part of your deck or your stack. It is, how do you change the way you make decisions now that you have the technology? Most teams bring in a tool and then rebuild their processes around that technology, behaving the way they used to behave before they had it.

Anca Pintilie: So you actually start with change management well before you bring the tool. And that is not training people how to use the tool. It is understanding, and you do not even need to name the tool. I just tell the team, three months from now, here is what you will get as an input into your system. Here is your operating model. The new input is that you can automate this, or do that. Now, seeing these things, how do you change the way you behave? Where is your approval process plugged in? Do you have humans everywhere, at the beginning, at the end?

Anca Pintilie: We need to think about the logic before you train folks on the tool. I know all this because we made all these mistakes. We had the tool, came in and said, let us train everyone. Three months in, I asked, why are they not using it? Because inertia plays a big role. People go back to what they know, because that is a faster path to success, even if that success is flawed or the data is not right. It does not matter.

John Emmerson: Changing human behaviour is one of the hardest things to do. It really is. In my experience people can be very receptive to change, and yet when you are asking them to change the way they act or work, it is just so hard. We are creatures of habit.

Anca Pintilie: We are. And again, it depends. Have you been rewarded for that behaviour in the past? Then that is going to be your default behaviour. So it is tough. What I usually say is this technology is your decision number three. First you map your operating model, then you start your change management, which is, how am I going to behave differently? What are the things I need to say goodbye to? There is a mourning period sometimes.

John Emmerson: Mourning period.

Anca Pintilie: Breaking this apart now, sometimes you just completely deprecate certain processes or decisions the teams were taking, and you do not replace them with anything, because they never needed them. But do not underestimate the sadness that comes with some of those things.

Anca Pintilie: And why do we need to map processes? In my previous role I worked across six marketing teams, about 300 to 400 people, a lot of processes. They were going to market in similar, sometimes overlapping, markets. When we mapped out the operating models of these six teams, we found they had about 13 of them. Nobody knew that, and again, these are talented, high-performing people. Nobody is at fault. They could have just paid more attention.

Anca Pintilie: Sometimes these operating models were driven by the conversations they were having with the business. Quality versus quantity. Performance versus growth. New acquisition versus loyalty and retention. So the marketing machine in the background was operating differently. But your machine does not have to be different. Just point it at the different goal. You do not have to build the whole decision chain downstream in a completely different way.

Anca Pintilie: When I say change management starts, part of that is codifying your decisions. If John goes away tomorrow, do we know how he takes decisions and produces a lot of value for our business? Or do we need to wait for him to come back to figure out how we are successful again? These are some of the things about discovery. I know it takes a long time, but it is an exciting piece of work as well.

John Emmerson: Yes, I think so. I do not think we can underplay the importance of it, which is what we have been talking about for the past 20 minutes. I am going to change tack slightly here. I am going to reference something you posted on LinkedIn a couple of months ago. You said you did not believe in permanent roles, that you believe in being paid to change something and knowing when that work is done. At the time you were talking about it in the context of marketing leadership. But do you think that holds true for AI projects as well, where someone often needs to stay close to the system long after launch?

Anca Pintilie: Absolutely. I think it holds for AI as well, because a mandate has milestones and a definition of what done means at every stage. You usually define those at the start. Things can change and you revise throughout the project. But at some point a senior role starts hitting diminishing returns against a specific mandate. To pay senior transformation rates to keep running a system that is already stable and just needs maintenance is the wrong spend for the business.

Anca Pintilie: And when I say to pay, that does not just mean consultancy or advisory roles. It can mean permanent roles within the company. But a mandate is, if I am done here as a senior person, then use me for something else. Let us build our next challenge together. Or maybe I have already identified a new challenge and I am building it with management and pitching it. So whenever you build a plan, you need an exit path from day one. The goalposts will change and things will be more flexible, but it is about having milestones in mind to ask, am I hitting it, am I extending it by three months?

Anca Pintilie: For an AI project, the launch is not the done part, which sometimes it becomes, especially across technical teams who are measured against launch and deployment. Launch could be the middle. It depends where done is for you. Done is when the operating model has absorbed the system and it runs as a properly defined role.

John Emmerson: Yes, I tend to agree with that. There are a variety of ways you can approach these things, and generally people need to figure it out for themselves. Often the best way to figure this stuff out is to try, mess it up, learn from your mistakes, and try again. You have also said recently that marketing problems are really about marketing. So when a marketing team is struggling to get value from AI, what do you think is the real problem that sits underneath that?

Anca Pintilie: When a team cannot get value from AI, the AI is almost never the thing that is broken. AI needs a clear business question, asked against data that everyone agrees on, to a standard that somebody owns. Then it comes up with a solution. So, can you as a marketing team give any of these three, or all three? Do you have a clear question you are pointing the AI at? Does everyone agree on the data and the story it tells? And do you agree on the standards of delivery?

Anca Pintilie: If you go back to these workflows, this is where the question is. Going back to my example with the six teams and the 13 parallel workflows, we had a multitude of versions of what good looked like at CRO level. We did not really have agreement on what the question was. If you point AI at any of this, the decision is going to be that it does not work. On the data side, we had about 68 external vendors feeding data globally on one part of our business, and there was no single measurement framework across all of it.

Anca Pintilie: You had to spend time reading different reads against all of it. It is a problem exactly because AI is not broken. AI is just amplifying whatever you have internally as a definition of success.

John Emmerson: That comes back to what we were talking about at the start, does it not? You need to define what you are trying to do first, then roll out the tool, and then you have measurable metrics and ROI to go up against. The same way you would with anything in life.

John Emmerson: I imagine that in your role at Amazon you got pitched by a multitude of AI vendors, probably thousands of times a day. I read recently that there are about 300 new AI vendors with solutions on the market globally every single day, to give you the scale of the change going through. What advice would you give a marketing leader, or any leader, about how to sort the wheat from the chaff? How do you understand what a potential solution can do for the business, and what kind of claims make you cautious when you hear them from new AI providers?

Anca Pintilie: There are loads. One claim that makes me super cautious is any claim that sounds like it is just an intelligence layer on top of your current system. That minimises the impact of AI, which can automate dysfunction and amplify a lot of things you already have in your process. When you minimise that impact, that is when I do not want to engage at all.

Anca Pintilie: Now, if you are pitching headcount savings or time savings, that is okay. It is an opening for discussion, and then you probably wait for me to come with the real challenges I have, and we put a number behind it. But if the vendor does not go back to understanding our operating model, or does not inspect whether we understand our own operating model, that is where I become cautious. You are just trying to sell me something that sits on top. That is where I become really cautious.

John Emmerson: Yes, I think that is right. We all get pitched hundreds of times a day and most of them I ignore. So if you are a vendor trying to cut through, you need to do your homework, the same way you would with any kind of sales, in my opinion.

John Emmerson: Unfortunately, we are nearly out of time. This has been an incredibly enjoyable conversation. I always try to finish with the same question, and I have not prepped you for this one, so it is out of the blue. Our audience is people in scale-up and mid-market organisations, business leaders across a range of functions. What would your single piece of advice be to those individuals about how to adopt AI in the next couple of months?

Anca Pintilie: It is the same. I am going to go back to the same thing. Understand your operating model. Do not spend six months on it, you are fine. Go in a room, be super honest about what you are doing, map it out, and then figure out what you can do faster. Do you want to do it faster? Do you want to replicate certain decisions? What are the goals of that AI? What exactly do you want to do? It is easy to start with time savings, and automate whatever can be automated.

Anca Pintilie: Sometimes it is okay to go in stages. You do not have to do it all at once. Even if you bring some tool into your business, that is fine. You can say, I am going to activate a bunch of features, I do not want to put that pressure on. But go back to the basics of your operating model. If you have metrics that you report on, whatever you report on needs to be managed. So maybe try to give that away. Simplify your reporting, simplify your decision-making, simplify all the stages. When you go back to the bare bones of what you need to produce value for the business, after that you can take your decisions on AI.

John Emmerson: Brilliant. What I love about this question, because I never prep anyone for it, is that everyone always gives me broadly the same answer. So, do not buy something before you figure it out, in simple terms. It has been an absolute pleasure to chat to you today. Thank you so much for your time, and we look forward to seeing what you choose to do next. It is going to be very exciting, I am sure.

Anca Pintilie: Thank you so much, and thank you for having me.

How do you know if a team is ready to adopt AI?

Pintilie’s readiness test is a single question: can the team describe how work actually happens inside their function? That means knowing how decisions get made, which decisions trigger spend, what data is used, and where it comes from. If a team cannot describe its own operating model with accuracy, it is not ready. Buying and building AI are both easy. The hard part is honesty about how the function really runs. Without that clarity, any tool sits on top of processes nobody fully understands, and the results will disappoint. Readiness is about self-knowledge, not budget or technology.

Why do so many AI projects fail to deliver ROI?

Adoption has risen sharply, but returns have not followed for most mid-market teams. Pintilie argues the tool is rarely the problem. AI amplifies whatever definition of success a business already has. Point it at a messy operating model and it will automate the dysfunction faster. Teams then build ghost processes to manage the tool that never fit, making the situation worse. The fix is not more technology. It is doing the discovery work first: mapping how work happens, agreeing on the data, and defining what a good decision looks like before any tool is bought.

What is automating dysfunction and how do you avoid it?

Automating dysfunction means deploying AI on top of broken or undocumented processes, so the tool scales the existing mess instead of fixing it. Pintilie compares it to opening a bag on a world stage and finding random clutter inside. To avoid it, run an honest discovery before buying anything. Map the operating model, identify duplicate workflows and legacy approval steps, and strip out activities that are measurable but no longer valuable. Sometimes the answer is not AI at all. A cleaner brief or a simple deterministic model can solve downstream problems that a full AI workflow never would.

How long does discovery take before an AI rollout?

Longer than most leaders expect. Pintilie planned two months of discovery across six marketing teams and it took five to six months to reach the truth. The questions were simple; getting honest answers was not. Her team ran around 72 calls and only made real progress once everyone sat in one room and mapped the operating model on a whiteboard. That analogue moment is when people finally see what they do every day. The work is slow, but skipping it means the outputs from any AI tool will be weaker. Treat discovery as the work, not a delay before it.

What is the difference between metrics that get reported and metrics that drive decisions?

Many teams run what Pintilie calls a reporting festival: days spent presenting numbers that change nothing. The test is which metric actually changes a mind or a budget. Click-through rate gets reported. Incrementality changes the budget discussion. If you cannot measure the thing that drives the decision, own up to it, show the decisions you could make if you could measure it, and ask management to unblock the dependency. Before automating reporting, cut the metrics that only satisfy curiosity. Every metric you report on has to be managed, so simplifying reporting is often the fastest win available to a function.

Why should change management come before AI training?

Pintilie’s order is clear: map the operating model first, start change management second, and choose the technology third. Training people on a tool is not change management. The real work is helping people change how they make decisions once the tool exists, including where approvals and human checks sit. She learned this the hard way, training everyone on a tool then finding nobody used it three months later. Inertia wins because the familiar path feels faster, even when it is flawed. Usage is not adoption. Embedding means people stop working around the technology and change how the business actually operates.

What should leaders look for, and avoid, when evaluating AI vendors?

Pintilie becomes cautious with any vendor who pitches their product as just an intelligence layer on top of your current system. That framing minimises the risk of automating dysfunction and amplifying existing problems. Headcount and time savings are a fair opening, but a serious vendor should want to understand your operating model, or check whether you understand it yourself. If they only want to sell something that sits on top, be wary. With around 300 new AI vendors reaching the market each day, the advice is simple: evaluate a solution against your operating model, not its feature list.