How long are they going to take? As I've said, about four years, but time's a pretty hard thing to predict despite being so predictable in itself.
All of these predictions come about from a few basic things. The first is intelligence has now become a commodity. That's the real big implication of AI.
Intelligence used to be slow and expensive. It's now cheap, and it's fast. How is that going to change our world?
1. Adaptive quant will replace quant and qual
The first prediction is that adaptive quant is going to replace all of current quant and qual.
Qual's the tool we have when we have unknown unknowns. Iterative, flexible, creative researchers figure out what's going on in the market while they do the research. It's a very agile approach. But in practice, it's a slow and it's a very expensive approach.
In practice, quality is going talk to really small numbers of peep numbers of people, which means that qualitative research has always given us depth without scale.
Quant, on the other hand, is there to quantify variation.
It helps us reduce uncertainty. It's about the known unknowns.
But it's also got its problems. The only way we can do quant is with large samples with standardized questionnaires.
And the bottleneck we have is humans can't adapt fast enough in real time. They have to lock down on standard questionnaires, and I think that's what I meant when I wrote lice answers here.
So the compromise is we get scale without nuance. Sometimes we get garbage in and garbage out.
If you think about it, the reason we have qual and quant is due to human limitations, but AI doesn't have these limitations.
It's going to be able to iterate and provide depth at scale. Adaptive quant is going to replace what we have today, and it'll be much better.
2. Synthetic populations will replace online panels
Unless you've been living under a rock, you know that synthetic respondents, digital twins, and the like are turning up as an alternative to online panels. At the moment, there's some pretty scary things happening, but the disappointment that some of you are feeling today will be replaced. The challenges that people have at the moment really relate to poor execution rather than the intrinsic problems with the technology, in my view.
For the last few decades, whenever a census has been conducted by the US or any of the other world's statistical agencies, you've actually been able to purchase these data files of maybe, you know, a few million people representative of the world at large or the population.
Now with these files, they've anonymized certain people of them. So you might buy a file which has got, you know, two and a half million respondents in it or something like that, not really respondents, but you effectively purchase a little population that you can draw conclusions from. And I think this is what we're going to start to see. But rather than the census bureaus of the world providing them, it's going to be companies like Google. You see, Google already knows all about us. Not just Google, all of the large online advertisers, they know an awful lot of information about us, and it's not that hard to hybridize that with AI and turn them into synthetic populations.
Initially, I think they're going to be for super fast turn around studies, quick insights. But as the quality grows and the desire for low cost locks in, they're going to grow and grow in popularity, and I think they'll just become the norm.
3. Validity will shift from representativeness to explanatory power
And this is going to lead to a shift in our idea of validity.
Historically, our main focus in market research regarding validity has been with the idea of representativeness.
Do the people we talk to really represent the world at large?
If our sample is good, our conclusions can be good.
That's always how we've thought about it.
And this is why the idea of AI bots answering surveys instead of people terrifies the industry because it undermines the fundamental premise by which so much market research has been done.
But if we have these simulated populations, the actual issue disappears. The synthetic populations can be created to be largely representative, vastly more representative than samples have been for at least thirty years.
Google and the like, they just know just about everything about all of us already. They've got a much better understanding of our populations and our variations than any online panel has today.
The new need is going to be for better explanatory power.
We'll need to conduct research in a way that allows us to build predictive models so that we can work out how each person in the synthetic population will react. React.
This is the same idea that's been around for a while in a quite technical area of statistics known as multilevel regression and post stratification, known as MR P for short. You can look up the work of Andrew Gellman for any of you who are particularly interested in this.
In this world, you don't worry too much about whether your sample is representative. You instead focus on understanding how different types of people think differently and behave, and then you can use that to project onto the synthetic population.
Research starts to be conducted so that rather than the research trying to tell us the truth, it provides conclusions that can be transported or spliced onto the synthetic population, and then we can learn the truth.
This is a kind of complicated idea. Back to some simpler ones.
4. Research will speed up massively, and AI chat will replace most researcher meetings
The fourth prediction is that everything is going to be sped up massively, and I'm sure nobody is surprised by this. Fifty years ago, a research study of any importance took months. Twenty years ago, it took weeks. Today, we can do them in a few days. Soon, we can do them in hours.
Chatting with AI is going to replace most meetings, certainly most readings with researchers.
Today, when a stakeholder talks to a researcher, they've got a couple of goals. They might be trying to brief the researcher on a particular study or analysis, or they're trying to ask questions to get deeper insights into a study that they've already done.
But in the world that's coming, people will ask AI, both of these, is the AI is just going to be better at it than a person. What do I mean by better? Well, it's going to be much, much faster because AI is faster. It doesn't have to go away and think and do the research. It can do it like that.
It's also able to just know so much more. If you don't believe me, just ask chat GP Tech question about just about anything.
It's going to be better than us because it can connect all the dots from all of the other studies that are in our company databases instantly.
And it's going to be better because it's just so much easier to click the chat button than to try and catch up with somebody else.
Let's try this out.
Okay. So I've got a study.
going to add some data.
And what I should have done is tell you a little bit about the study first.
Some of you will have seen this before. I'll give you a moment to read the concept board.
So we did a standard little concept test. Nothing particularly unusual.
I've loaded the dataset. Little chat asked me if I'd like to write a report.
I do.
It offers to clean all the data for me. I would normally tell it to do that, but I'm going to click skip just in the interest of time.
It wants to know what I wanna learn about. Okay. Well, I'm going to tell it a bit about the study.
And I'll give it a couple of objectives.
So I wanna firstly know and I know some of you have seen this before. Don't worry. You're going to see something even newer today.
Should I launch the iLocke?
Give me a very clear goal in terms of what I wanna learn. I'll ask some more general questions.
K. Now I'll let this AI agent go off and do its work. So there's a couple of agents that play. There was the data preparation agent, which I talked which I skipped, and then we've got the research agent.
It's going to ask us which data we wanna use. I'm just going to leave it with default, and I'm going to let it go away and do its thing. And it's going to take a few minutes. So rather than watch us do it, all that thinking, we'll save a few minutes and see the report that's written.
So the first thing to note is it's answered the questions. It takes a couple of minutes. It'll give you a chance to read the answers.
Now in this case here, it's written as traditional report. I've got little hyperlinks as an exec summary. There's all the charts and all that stuff. But I told you before, people want a chat. So let's see what chat looks like.
We've got the research. What are the questions? We're not going to wait around to listen to somebody lecture us or tell us the results. We're just going to ask it some questions.
So here's the result it's telling us. Well, you can read it for yourself, or we can just follow the link.
So there is a slide in the report which answers it directly.
I might change a few things on the thing just to make it a little bit easier to read.
Okay. And so now we can really clearly see but the summary results here, we've got a whole slide, tells us the relationship. We can see, for example, the definitely biased score amongst the twenty five to thirty four year olds is much higher, and it declines with age as it describes in the text.
5. Procedural and operational roles will vanish
Since research began, specialists like me had safe and well paid careers. I was an advanced quant guy before I became a software guy. You can remember a lot of the other specialties, and some of them are still around. There's qualities qualitative recruiters, text coders, field work rooms well, not field work rooms, field workers there off in the field, phone room supervisors, survey programmers, data processors, ops teams.
Now most of these operational roles are really entirely about procedural expertise, the ability to know what buttons or what actions to do in what order. Guess what? AI is better at this than humans already.
Without any further advances in AI, these procedure roles are going to keep vanishing.
Software just gets hooked up to the AI.
6. The in-housing tsunami will accelerate as middlemen disappear
The middleman economy will largely die out. The future is all about end to end solutions. The middle people, better than middlemen, I suppose, will be replaced by an in housing tsunami, and it's already happening.
Ten years ago, two thirds of our company's clients were market research consultants. Today, two thirds of people doing the work in house.
7. A new skill will emerge: creating AI-digestible research
A new skill is coming, the ability to create AI digestible research. It's going to replace humans telling stories.
For my whole career, the most successful researchers were the ones with presence, the ones with control of analogy and metaphor. They were the great storytellers. But if we're asking AI the question in chat, we're not going to be meeting with people, and these skills become less useful.
What do we do instead? We need to learn how to provide data that AI can easily interrogate and then serve up to its end users. What does this mean in practice? Well, I think the PowerPoint debriefs are going to massively shrink from the fifty or a hundred page reports to three or four pages strategic question. If people have follow ups, rather than there being fifty slides prepared in advance just in case they answer a question, people will instead just ask the AI.
The new focus is going to be on creating things that are easy to read by AI, easy to be ingested into corporate data repositories, easy to be found and used by rag models, validate the throated research that can be relied upon by the human and the AI.
And the examples that we're going to be seeing are, as I said before, short strategic recommendation packs, small interactive dashboards, analytical frameworks such segmentation on pages, and the kind of things we get from qual at the moment, FoxPop, case studies, testimonials, and quotes. And, of course, it's not on the slide, but it should be. The raw data files are going to get even more useful.
8. Market researchers will shift toward strategic, syndicated, and tool-building roles
Forever market researchers have dreamt of annuity like revenues, the kinds of flows generated by products where they sell them once, and they keep getting them renewed and purchased again and again and again. But clients always kill this off. The researchers have come up with the products, then the client wants to tinker. They want something just for them.
But this is going to change.
Because with the AI doing all of the tinkering, there's going to be a little more opportunity, a little more of a need for market researchers to do something else. What will that something else be? Well, one thing is many market researchers will become in house researchers.
Many market researchers are going to become a lot more strategic in their focus, and that's really been what great qualitative researchers have always been. They've been strategy consultants.
And I think one of the big changes is going to be that market researchers will start producing syndicated data products. Syndicated data products have always been around, but clients often buy them and then don't get around to using them. But AI is great at using syndicated products, so I think there should be a big growth in them.
And lastly, some of us will spend our efforts building tools for AI to use, which is what I do.
9. Data visualization skills will decline in importance
For most of my career, I've been obsessed with data visualization.
It's one of my absolute all time favorites. I love it. Poppy Field. If you've never seen it before, it's so pretty.
I absolutely love almost everything about this magnificent, magnificent visualization. But it's from a different time, and I think stuff like this is just going to disappear pretty quickly.
In market research, we used to work hard to create visualizations, and many people still do. Maybe not as hard as the puppy visualization we just saw, but we work really hard because we're preparing results so they're easy to digest by humans.
We're trying to make it so the results just pop out of the page.
But in the AI world, the AI is going to be selecting the results for us, and it's just going to be a waste of time creating these pretty visualizations that maybe never get looked at.
We're going to shift from creating tables and visualizations designed for people to large amounts of data than AI can easily use.
10. Conjoint and MaxDiff will disappear
And the last of my predictions is I think we're going to see the end of conjoint and max diff. These were things I specialized in for much of my career. Why are they going to go? Well, these have always been techniques invented to tease precise answers out of imprecise humans. But when we're dealing with synthetic populations, we can ask quite complicated questions, and they kind of they'll be able to do that choice modeling type thinking inside their little AI brains with much more precision than a human being ever could.
So there you have it, my ten predictions. What do you think?
What questions have you got? Please type them into the questions field in GoToWebinar.
But I have to fess up. For those of you that have been coming to these webinars before, you'll know that I always do them live. But sadly, I'm currently in Thailand because one of my family members is sick, and so this is a recording.
So but do type your questions in, and I will answer them. We'll put together a blog post, or I'll answer you individually if they're very specific kind of questions. Anyway, thank you for all of you for using Display, if you're using it. And if you're not, we'll book a Display demo.
Bye now.
