The Roadmap for Market Researchers in the Age of AI

We're going take you through our thinking about what market researchers and market research companies need to focus on.

I often get the complaint that the audio is not quite right. So if someone can send me a thumbs up or something, that would be really helpful.

Awesome work. Angela, you are first.

We're going to start by looking at what's changed and changing, then how we think the industry will change, what your company should be doing right now, and what you should do to reskill or change the focus of your skills in the age of AI.

We can think of all work as being either messy, manual, human performed tasks or automated tasks. It's a really clean line.

Technology helps us migrate tasks from manual to automated.

And even when you have a task where you're kind of working with the AI, you can split that up. There's a bit that is human and a bit that's automated. There's always a clean line.

AI is a massively transformative technology in two ways.

AI allows us to transform things, things from data to text, from one language to another, and it changes the way we work.

If we look at the properties of AI, the key ones that really come to mind are it's fast.

It's so broad in its knowledge base, and it's a commodity. It's very cheap, and we can turn it on and off whenever we need it, and it's online.

It's very easy to think about adopting AI as being about finding the right tools, but I don't think it's the right framework. I think, really, successful AI adoption is all about workflow design.

Tasks can only be understood.

Sorry. Tasks can only be automated when they're understood, standardized, and made reliable. So I'm gonna dig into that in a little more detail.

When is a task ready to move from messy manual to automated?

The first thing is there needs to be a way of automatically doing it. This has has to be some technology that can automate the task. So back in the olden days in market research, a really big event was when optical character recognition came in and could convert what people had written on a sheet of paper into computer readable outputs. Today, we're really talking about the AI technology.

The second condition is it must produce work that has a reliable a known level of reliability, which is acceptable to you.

And most of the time when people complain about AI, they're really complaining about this. They haven't measured it or the reliability is too low.

You only migrate if it's gonna improve the overall workflow. That is, do you become more productive?

And you should only migrate when the workflow design is correct. And this is the bit that I think is a lot harder to get right and to reason about.

For example, one of the biggest pain points that people in market research have is automated creation of PowerPoint. Or, one of the point, the creation of PowerPoint, and they want to automate it. But this really assumes that PowerPoint is a smart thing to do.

But is it? I'm not sure.

And I'm saying this, we have a lot of tools that automate PowerPoint. But have a think about it. Let's say we have a whole lot of crosstabs, and I've got a whole lot of crosstabs here. And these aren't just normal crosstabs.

They're interactive crosstabs. The user can change the filter, so there's even more crosstabs here. Now in the traditional world of market research, some lucky person reads through these, finds what's interesting, then they chart, and it becomes a PowerPoint presentation and so on. But is any of that really necessary?

What if the user can just ask a question of the tables?

And so the feature that I'm demonstrating to you now is going live over the next few days.

We've got a few little tweaks we've got to get out of it before we release it. One of them is a little slower than we'd like, but it's actually doing something pretty cool. It's reading through hundreds of crosstabs that have not been interpreted at all by a human being, that nobody has charted, and it's trying to answer this question.

That's cool. It's even giving me a reference. But now what I've shown you there, a lot of you can get that same effect by using other tools. But let's say that we really wanted to filter the data and ask the question now just amongst people in the diet.

Now it's not reading through the tables of pre created. It's rerunning in the background all of the tables to apply that filter, and then now it's going to answer this question. And so we're now in a territory which even traditional PowerPoint couldn't help you. Right? This is it's running the analysis on things that you didn't chart. So we're a long way advanced potentially from using PowerPoint.

And it will give us the result very soon, hopefully.

Alright. Cool. And so it's given us the result, and then we can click on the detail.

And we got that table. If we really, really wanted a PowerPoint, we could export this off and turn it to a PowerPoint chart.

So to just really emphasize the point, it's really important we've got the best workflow design when we automate, and it's often not the workflow that we're used to using.

And AI massively increases the number of tasks that can be migrated. Here's a different example.

For more than twenty years, I've had a standard approach to how I do data cleaning and data preparation, and the approach goes like this.

I run summary tables. One table gets created for each question in my survey. I read through them one by one, and using my enormous expertise, I go, oh, this is good. Then I change nothing, or it's bad, and then I change the data or the labels or something about it, cleaning and tidying it up as I go.

I might change some labels or I might delete some cases, and then I go to the next question. And by the time I've gone through my whole survey, I'm finished. So I've always been pretty happy with that as a workflow. But last year, we had to build an agent called the data preparation agent, and it automates the cleaning. So it's designed for people who don't have the skills or the time to manually do all of this.

And so I had to encode what I was usually just doing based on judgment into a formal set of rules, and it ends up creating a completely different workflow. So this workflow that I've used forever isn't the workflow that ended up getting built.

The workflow end up getting built isn't done question by question at all. We do operations for all of the questions, all the questions in a big loop. And we start with data tidying because it turns out you need the data to be tidy before it's easy to clean for AI because AI needs it in a stand format, whereas a human being can have untidy data and still clean it. And human beings tend to tidy after they clean. Then it goes through, flags all the dirty data, then it deletes cases, and then it goes through a last few set of activities.

Now AI is not the first technology that's come and changed how market research is done. There's been a long history of major technological revolutions massively changing how research works. And for me, in my career, which started market research frighteningly a little over thirty years ago, when I started, there was, like, ten people in every operations department doing all of these things to do with field work, and there's ten people in every operations department for every researcher, and they're largely gone. Right? So it's changed a lot. If we jump down to the AI error at the bottom, the biggest thing that's really changed a lot of things historically or historically, I'm only going back to twenty twenty two, is the automation of text coding. It's pretty unusual that people are doing manual text coding in these days.

The qualitative research is, again, it's massively changed how you process all of that open ended or the generic text data or even voice data. If we look at twenty twenty five, the big changes from our point of view were introducing the automation of data cleaning and preparation via agents, an agent that can automatically perform all of the analysis report for a basic report to the level really of a a good graduate or better than a good graduate. And then the thing that rocked the whole industry was vibe coding, where you can now, just using text instructions, get software to build new other bits of new software without knowing how to write code. And it's early days yet, but it's already making a massive difference in how we work, and I think it should be making a massive difference in how you work pretty shortly.

In twenty twenty six, I've just shown you the ability to do q and a if you cross tabs and other reports. It's almost released, and you can see in purple the things that we're going to be released coming soon. Give you a chance to have a quick read. Have some coffee.

So all this technology, how will it change the industry?

As I talked about in the last webinar, the executional roles are under a lot of pressure. Think about this way that many multinationals still work. The marketing director has a question. It gets passed down.

It keeps getting passed down and down until the worker appears.

And the workers do some work. It gets passed up. Value add. Value add. Value add. And four weeks have passed.

What does the future look like? I think the future is gonna look like this.

Notice we've only got one face in this picture. All the other people aren't needed in this future. Now is this future literally gonna be here tomorrow? Certainly not. We're gonna hope not because I'm out of business if it is because that way, we're not close to this future, sadly.

But a lot of those roles will collapse down. Right? Fewer people need to perform the execution roles than when I started, and very few are gonna be needed in some of them in the future. When does that future arrive? It it could arrive in a year.

We won't build it if it arrives in a year, sadly. I mean, somebody else has built it, and we're out of business.

I think it's more likely we get to this kind of level, maybe in ten years, but I don't know. What I genuinely hope happens is we discover some technological thing that can't be automated, some intrinsic brilliance of market researchers which no machine can replace.

I hope it happens. I have seen nothing to make me think that will happen. In fact, every single time that I have seen somebody say their job can't be automated, I've always looked at them, that's easy to automate.

The the real issue is scraping from what's in their head and putting them into the computer. I'll talk more about that shortly.

So I think it's important to really look at how the industry works at a very simple level.

We're a project based industry. People conduct a project. They go through a standard workflow. It's just done rinse and repeat in a very standard process. It's ripe for automation.

But I think automation isn't just gonna make this faster and cheaper. I think it will fundamentally restructure the industry.

The first big change, I think, we're gonna see is a lot of the execution work, running the tables, interpreting the tables, and things like that, is gonna very quickly and or or it is migrating to be done by software agents.

Now this work, which historically has been done by market research companies and internal insights teams at some companies that in house their market research, I think quickly gets done by agents. And you'd be crazy not to in house the agent. So if I'm a Unilever, I might outsource all of my concept test today, and I don't know.

But I think in the future, I'll have an agent. I'll just do my concept test internally. It doesn't make sense to outsource to another company that does an agent. It's too inefficient.

And a related change that I think will happen is projects will disappear as the main way of organizing research, and research will just move into the always on phase.

And this change is not such a speculative change. It's it's already been widespread for the last four or five years in terms of NPS tracking, customer satisfaction tracking. That's moved to an always on world. And I think we're gonna see it much more. I think brand trackers will go back to continuous trackers, and they'll be always on. But the always on isn't just the surveys. It's a socialist and all of that being done and integrated by agents.

Some research companies will continue to work a little bit similar to how they do today, but they're gonna move up the reasoning hierarchy, competing with the management consultancies.

A very small number of very strong methodologists will build what I'm calling research architecture firms. These will be firms that help clients set up their in house automated, always on research system. So these firms will help build these. But there's not gonna be many of these firms.

And the reason is that at the moment, the industry is flooded with people who know the whole research process and are strong methodologists. Because today, everyone who does a project needs to be a strong methodologist. But the minute the agents have that wisdom or intelligence in them, knowledge in them, then generic research skills of those kind become a lot less relevant. And so it's only gonna be the super strongest people that would be needed, and it's only gonna be much rare because, you know, you're gonna set up a research system, and it's just gonna stay on.

Project base disappears.

And I think the very large firms are gonna specialize in proprietary syndicated data services, and these then just get hooked up to clients' agents.

And we can see this process already starting to take place. Now some of the the biggest market research companies have always done this, but they also did ad hoc research, and they also did trackers. And I just think that's just gonna disappear in terms of being done for specific clients. As I talked about before, I think we're going to see things like brand trackers are gonna become industry wide brand trackers, and they will still be tailored to individual clients, but agents will do that.

So there'll be no point in individual brands doing their own tracking. It'll be much more efficient if they buy just hook into an agent which has got it. And those agents presumably have lots of different trackers, which they all synthesize on their own. So that's what I think is gonna happen.

What should your company be doing? There's a very common path to successfully automating a market research.

You gain expertise, and I think most market researchers are here. And then you have to make this switch from being a craftsperson to running a factory. And this is really where most market researchers historically struggle. And the thing they fail is standardization. Gonna jump into that in bit more detail.

Automation only happens after standardization.

Now standardization doesn't mean that you agree that on a color scheme. It doesn't mean that you agree on a broad sequence. It means that for each of the individual tasks that need to be performed, you've measured their reliability. You know how to reproduce them to an acceptable level of quality or reliability, which means that you've got both clear processes to follow and verification rules that are explicit, that don't depend on someone's expertise to assess. If you're in the world of I know what good looks like, you've standardized nothing.

Once you've standardized, you start to use templating. Now templating is something that's very underused in market research to the point where most market research companies say they have templates, but they don't. And the classic one of this is they've got this PowerPoint file. It contains some charts.

They call it a template. They copy the chart, and then they manually edit it. So in a very technical sense, it's a kind of template, but they haven't actually replaced much of the task. They haven't even replaced charting.

They've just replaced creating the initial chart. Right? So it's a very micro task.

There are lots of opportunities to template once you've standardized. The heavier you standardize, the easier it is to template. You can template your visualizations, how you create variables, your whole analysis workflows, and entire reports or documents. And the last thing where market research companies have really got a a very strong tradition of failing is is when they outsource the automation.

It goes like this. You hire an expert, you tell them what you wanna achieve, you show them some sample reports, and you lead them to it, you trust they know what they're doing. But the problem is only market researchers have the sufficient context and knowledge to know how to standardize. No external consultant can even know that.

And so if the market researchers aren't at the center of this whole process, it fails, and it always fails. It can be very expensive. It's not an accident that all of the companies that sell market research software have technology that was built by former market researchers. Right?

There's there's no one out there. We don't have any case of somebody who knew nothing about market research coming into the industry and building important software. There's too much context.

Successful automation really is about understanding. In the beginning, people describe their problems, call it the fog.

Then they work out what the tasks are and give them names.

Then they need to get to the stage where they can verify that a task has been done correctly without using objectives, without requiring expertise.

Then there's a flowchart describing the most common or the happiest path to do the work, and finally get to the stage where we have a detailed understanding of all the edge cases and variations. All of this has to happen for automation of the whole workflow and of any individual task. So you only have to get to this stage for a task to automate it.

So that's what the company should be doing, we think. What should you be doing?

If we go back to how the technology works and how transformative it is, there's two massive implications. The first one is that we're all going to be managers.

Now when we say managers, we're gonna be managing lots of agents. I have, this week, literally used more than a hundred different agents.

And so the skill required when you're managing agents is managing at scale. It's very different to the skill required to manage a team of four market researchers. And I'll talk more about how that difference works in a second.

The other thing that we need to start to get really good is connecting things.

Now the bits the very seldom to this. It's about using software that has APIs or command line interfaces so that an agent can talk to it.

If you've got a task that's being done away, then agent can't talk to it or get the result It doesn't work. It's about putting data in places that can be easily connected to, ideally databases, good folder structures. It's about making all of the implicit context explicit. So if you meet with a client, it shouldn't be in your head. You should record it, and that recording should be transcribed, and that transcription should be given to an agent. If the agent doesn't have the context, it can't succeed. It's about creating data in a way that's easy for an LLM or a machine to know to understand.

How do you get all this done? Well, you think about it, but a lot of it actually is about using agents to do to achieve the above. So for example, in our company, we're moving the whole company using what's probably the most popular tool for this month, which is a tool called court code. And I think lots of people who've never written code are now having to write code. It's a messy, scary experience, but you need to. You need to write code, but you don't write it until the agent to do it, to do a lot of these things so that everything can be hooked up.

So one technical thing to focus on, pharma market research, I'm gonna start playing with Claude Code. It's gonna be scary as heck at first. You just gotta remember that when you're stuck, you can ask it. Don't need to ask somebody else. Ask Claude Code.

How do management skills change? Well, for most of my career, and this could just be my personality, but I don't think so, management coaching is related to all of the soft skills. But these are less important. I'm not saying they're not important. We still have to manage people, but they're not very relevant to how you work with AI at scale. We need to change from thinking of ourselves as people managers when we work with AI to becoming system managers, much more like a factory manager. So we've got to monitor, look at edge cases, find problems at bottlenecks, and our days need to be about continually improving the system, not simply executing.

When you manage large teams like a factory manager does, you get much more concerned with explicit delegation, formal roles, responsibilities. You don't have a vague understanding that who's gonna work on what is really clear. This particular task is gonna be done by Tim by Thursday.

And the bit which tends to be surprisingly hard in market research companies is ensuring expertise is encoded. Many people who know a lot like to see themselves as the expert, but if you've got a person who's the expert, you can't move quickly. You've got to codify this expertise, build it into your software and your processes.

In terms of general search skills, it's really easy to say what we all need to do. We need to be focusing on the top of the reasoning hierarchy. The bottom of the reasoning hierarchy is what gets automated away first and has always been automated away pretty quickly in market research. So we wanna focus at the top, not the bottom. The higher we are, the safer our careers are. And so if you're down here, you've really gotta be building these higher level reasoning skills.

This new subskill, and I alluded this alluded to this before in the discussion of fragmentation, which is research system architecture, which is becoming good at building new systems that continually are always on that do data analysis or data capture data analysis and allow other agents to talk to them. Now most market researchers have pretty broad skills across the whole research process, but this new subskill requires both that general market research skill and being really good at problem definition. So it's not a skill that a junior researcher is gonna move into quickly anyway.

Just to really make the point, which I've made at least three or four times already, the execution skill is in decline. It's still a valuable skill. People who get stuff done are always gonna be useful. But if your whole identity is tell me what to do and I'll get it done, hey. I'm gonna talk to an AI. I'm not gonna talk to you. Also, being the expert or an expert on a topic is just not so useful anymore.

When I started, I was the advanced analysis guy in the corner. People came and talked to me. I was kinda difficult to deal with. A lot of us were because we're allowed to be. We were special.

But today, people just go to the LM. They don't even come and visit me anymore. LM ain't difficult. Right? This whole being an expert disappears. I'm not saying expertise ceases to be useful, but being the expert, people come to ask a question. Pretty much gone.

Skills on the rise. Your ability to reason about what causes what is more important now than ever because it's something AI is still bad at.

Working out when AI is wrong, being able to look at it and go, no. It's wrong is a massive massive skill.

The ability to ask questions that is define what are the interesting things to know becomes extraordinarily important as well. And this is really positive for all of us who have been market researchers or are still market researchers because the schools on the rise are actually always the thing that have marked good market researchers.

Right?

Probing and asking questions, that's all market researches on one level. Cause of raising is the entire analysis of market research. So I'm pretty bullish about what market researchers can do, even if the way they work, I think, is being changed massively.

So this is what I've taken you through. They're my two cents, my thoughts on what it should be and my colleagues' thoughts on what it should be. What thoughts do you have? What questions do you have? Please type them into the GoToWebinar app. Let's see what questions we have.

Brenda asks, can you show me the data preparation agent? Sure.

So I'm going to create a new document.

We're looking at display now, in case you don't know.

This is the new user interface. You have to opt in to this, but it'll be turned on for everybody soon. Starts off in a conversation, tells us to add data.

Then it tells me to write a report. I'm gonna do that, and it's very first thing it's gonna ask me about is the data preparation agent. And so what you can see is this is that flow that I described before. You can opt in and out of whichever bits of it you want. They all have to happen in this particular sequence. You can customize it by manually overriding things later.

And now it's gonna go through and do what I was describing before.

Maeve asks, can clients ask the chat strategy questions? You sure can. Go into it just like I showed you before, click in it, ask it some questions. Like with all AI, sometimes you need to prompt it a bit carefully to get the right type of answer, but you can definitely do it. At the moment, our research agent is much better at strategy than the answer question tool, but we'll take that pretty quickly. It's one of the things that is holding our release. Surprising how tricky AI can sometimes be to work with.

And Edward asks, I know Q Despires focus on quant. However, how do you see AI impacting qual in the next few years? Yeah. It's a great question.

We frequently get kind of asked this question, and we're really doubled down on quant research at this stage.

You know, it's I haven't been a market researcher really for a decade, and I haven't actually done any qual interviewing since nineteen ninety five no. Nineteen ninety six. Nineteen ninety six. So my insight into qual is not great. One of our product managers, Dawn, she's got a strong qualitative background. So we we might move into that. But AI will continue to massively impact qual.

Right? And the reason it will is because it's vastly better at interpreting large quantities of quantitative data than humans are already. It's just able to read a lot of it. Vastly faster, so that translates a lot.

And it's vastly cheaper. So I think that the future of AI is gonna be heavily correlated.

Elise says, you talked of always on research via syndication. Do you see this as via traditional surveys, omnibus style, or deploying synthetic data? Both.

So it's I I do see when people have and I would stress two things. Well, not stress. Just to be a bit clearer about what I was trying to say before. I think always on research will both be done proprietary for companies, such as with customer satisfaction, and it will also be syndicated. But I think things like the brand trackers and ad trackers, I think they'll move to largely being syndicated just due to the economics, and they'll also be always on.

I think they will contain a mix of traditional interviews, but I think they'll also contain synthetic data. And I think the synthetic data is gonna get vastly better.

And I think synthetic data will become quite amazing. But synthetic data will never get us to the stage where it can answer a novel question that's not been asked before. So I don't see any world in which market research questions cease to be asked in traditional surveys. I'm not as confident that traditional panels will be about in the long term, but I do think we'll get to a situation where but but I don't think we'll leave to lead, for example, to situations where people won't interview their customers. I think that will be around forever.

Laura. Laura says, will the answer question tool just be in Displayr, or there are plans to roll it out versions of in queue? Now it can only be in Displayr because it's an online tool. Queue as a desktop program can't be used online.

All of the AI features are really in display, not because we're mean. It's technically impossible to do them in a desktop product or to do them well in a desktop product, and that's why they're in display. But if you wanna know more about display, go to our website, book a demo, download a trial. Anyway, for all of our customers out there, thank you so much for being our customers.

If you're not a customer, maybe become a customer.

Thank you, everybody. Bye now.

Bye now.