Market Research in the Gen AI Era

Daryush

When I look at Gen AI usage in market research, most folks are somewhere between curious and committed, which means they're all over the spectrum. When I double click on it, I see the same thing I see in other disciplines, and market research is honestly not that different. There's a lot of experimentation happening. People are testing, being cautious, and it's contained: small teams, teams of one or a couple of people, definitely one-off projects. It's not yet fully built into the underlying market research discipline and function.

It may sound a little grim, but it's almost the same thing you'll see everywhere. Some disciplines are ahead, like software engineering, which is way ahead in utilizing AI. But aside from that, everybody is still experimenting, getting up to speed, kicking the tires.

The real gap isn't desire or interest. The real gap is trust. Can I actually trust the output of the AI in my market research? Can I trust it enough that I'll bet my market research on it? It's ultimately about trust and ROI.

Daryush

If I break market research into building blocks, everything starts with design: what is my research objective, what do I actually want to figure out, and how do I design my research? That's the first place I see most folks adopting Gen AI, and it makes sense, because it's the first place you start.

Today, when you come up with a research objective and want to start the design, you pretty much start with a blank piece of paper. One of the benefits of Gen AI is that it helps with that cold start. It helps you initiate things a lot faster, which makes it a natural entry point.

It's also the best place given that the real blocker is trust. If you start in the design stage, you can build it yourself, utilize AI, compare the output side by side, and validate it. That's where I see the world getting started.

Daryush

Without AI, you start thinking through your research objective, the right method, the right questions, and then you spend a considerable amount of time actually doing the doing. Then you take a step back, squint at it, and evaluate: is this right, is this wrong?

With Gen AI, you're still doing the thinking. But when it gets to the doing part, what you're really doing is creating a very good prompt. The interesting thing is that as you write your prompt, it helps you clarify your research objective in your own mind. Am I framing the problem correctly? You naturally refine your thinking as you go.

Gen AI then generates results, and you're back to evaluation. You're looking at what the AI provided and applying your judgment. The thinking doesn't go away. The judgment doesn't go away. You just do less grunt work, AI does the groundwork faster, and you have more time to evaluate. Quality isn't compromised. Your throughput goes up, and you have a copilot helping you do the work better and faster.

Daryush

I'd argue AI actually makes the strategic value of the market researcher pop even more. Imagine a world without AI: you spend X percent of your time thinking and evaluating, and Y percent actually doing stuff, writing, going back and forth. That Y percentage is a lot bigger than the X. So from the outside, the researcher's role looks like someone who just does a bunch of work.

With AI, you increase your throughput and you're in the arena a lot more. You're providing your point of view, your expertise. That's the strategic value a researcher brings to the table: how you think about the objective and how you get the data you need. It's not doing the work, it's the thinking behind it. With AI, you come up with insights faster, and the strategic value you bring to the organization pops a lot more.

Daryush

Let's make the numbers simple. If you're doing 500 interviews and each moderator can talk to two people at a time, you logically need 250 moderators. Those 250 folks jump on conversations at different times of day, with different amounts of espresso in their veins. Maybe some happen after lunch. The quality of moderation goes up and down as a function of the time of day and who the moderator is.

Now, if AI is doing that moderation: AI doesn't need coffee. AI functions the same way before lunch or after lunch. You spend time tweaking it and making sure it's doing the work correctly, but once it's in place, the consistency is maintained. You can scale almost indefinitely without compromising quality. AI can run 500 interviews at the same time if needed. Transcripts come in real time, and you can adjust in real time.

By the way, Victoria, this question is extremely timely, because we're working on an AI moderator that hits general availability soon, and we have another webinar scheduled just for that.

Victoria

Yes, we'll have that webinar at the beginning of September, and we'll share the registration link in the chat for anyone who'd like to sign up.

Daryush

The consistency you build into the AI scales with how much you deploy it. Every interview, every conversation runs against the same guidelines, the same logic, the same quality bar. And you don't have to wait until fieldwork closes. You see results coming in real time: the analysis, the transcripts, everything.

If you need to tighten something, you tighten it right away. You don't have to set up a meeting with moderators to adjust. Your time to resolution shrinks because you're seeing real-time results, and once you adjust, you can redeploy that AI at scale in a matter of seconds.

Daryush

Traditionally, once fieldwork wraps up, you wait for the data to make it to your data lake. That takes a few days. Then you wait for your data team or data science team to jump on it and run analysis. The whole thing, from fieldwork close to usable insights, is measured in days or weeks, because humans have to do the work, with plenty of iteration along the way. Nothing wrong with that, it's just how it works.

With AI doing most of that analysis, it works on data as it comes in. It doesn't have to wait. The moment fieldwork closes, almost immediately, you have the themes, the pain points, the delighting moments. If everything is on video, you can generate highlight reels, all in real time. Same level of consistency, same level of quality, you just get to the insights a lot quicker.

If there's one thing to walk away with from this conversation: with Gen AI you can get started a lot faster without compromising quality, and the depth of the insights doesn't change. It's all about increasing throughput without compromising quality.

Daryush

Let's talk about a parallel world. I like airplanes. Think about building the wing of a Boeing aircraft. Before automatic screwdrivers, somebody had to manually drive every screw. Automatic screwdrivers just made it faster, but the screw is the same screw, the wing is the same wing, the quality is the same. Higher speed and higher throughput don't mean shallowness.

Put differently: say an analysis takes 10 days and has 10 independent steps. What if you had 10 interns, and you gave each one a task? You're still in charge of the quality. You still make sure the results they bring back are precise and correct. You're just doing things faster.

I sometimes hear the argument, "if I do things faster, how do I make sure the quality is there?" I think that's the wrong question. The quality is built into how you construct your AI. AI just lets you apply the same consistency and quality faster and at bigger scale.

Daryush

Pick one place, one spot, and start there. Don't try to apply it to 100 things simultaneously. Pick one area, whether it's design, moderation, any part of the workflow. Then quantitatively prove you were able to do the work at the same level of quality as humans, maybe better, but faster, cheaper, with more confidence.

Remember, the barrier to going big on AI right now is trust, in every discipline, market research included. You lose trust in buckets, but you build it drop by drop. So start in one area, show the value, prove it, and then socialize that value and build enthusiasm around it.

Daryush

In the beginning it might look scary. But the empirical evidence we've had over the past few years applying AI to business workflows is that the transition is much smaller than people think. You start small, pick one workflow, apply AI, make it better and faster. Then you do it in three more, then five more. Slowly people say, this is becoming obvious, let's just do it everywhere. The barrier to adoption is smaller than we think.

Where things get hairy is when you try to apply AI to everything all at once. Those boil-the-ocean efforts tend not to succeed. The efforts that succeed are deliberate and step by step. Once you get past that inertia, things have a habit of falling into place.

One more thing: there's a lot of conversation everywhere about agentic AI. Take a step back, and you can apply generative AI in one of two approaches. There's assistive AI: I'm the human, I'm still doing the work, and AI assists me, like 10 or 100 very capable interns. I give them work, they do it, they come back, but I'm still in charge, the human over the loop. Then there's agentic AI, where the AI has some level of agency to take over decision making and complete tasks.

At Fuel Cycle, we use agentic AI quite a bit in product development, engineering, even in our finance functions. Victoria, I know you're using a lot of agentic AI in marketing. But the reason companies like us can roll out agents in big, critical workflows is that they've spent the time to build trust in the AI. If your company has a gap around trusting AI, or you're at the beginning of the maturity curve, assistive AI is where you need to start. Get your prompts right, get your AI infrastructure set up correctly, build confidence, and then move into the agentic world. Leapfrogging over assistive and jumping straight into agentic? Do that at your own risk.

Victoria

That totally makes sense. Speaking for a marketing department, we started using AI two years ago, and it really was those baby steps. Once you feel comfortable, you can really run with it, and it's pretty darn great.

Daryush

If you're talking about two or three people, then a human talking to them and an AI talking to them will probably yield the same result. Where AI moderation makes sense is when you want to scale up: hundreds of interviews, completed in a short period of time, with the same level of quality. If you have pressure on the time dimension and want the highest quality, that's where AI helps.

You also need to think about where you're deploying it. I was recently rewatching Mad Men, for the third time, don't judge, and there's an episode where researchers are interviewing Burger Chef customers to find the best tagline. The researchers were hungry, they'd already talked to hundreds of people, and the quality of the research was obviously diminishing with time since the last burger. Those are the places AI can help. We don't yet have humanoid robots doing on-the-spot research in a drive-thru, so that's a place you can't deploy physical AI today. Something to think about.

Daryush

Our data shows two big areas. Number one: figuring out, refining, and going deeper into how to get started with a particular research function. Number two: asking very interesting questions of their data. "I have this piece of data, that piece of data, and something from six months ago. AI, can you connect the dots between these for me?" Those are where clients get the most usage today, and as you'd expect, it's all assistive AI. We're also starting to see traction on AI moderation, with more to share in early September.

Where I don't see AI making a big dent yet, because we haven't matured the AI there, is around fielding: how do I field this study, how do I pick my audience, do I know enough about my audience? Those are obvious, untapped areas of potential for us. We're starting to experiment there.

And for everybody on the call: we're always looking for design partners to join us in building these experiences. If you're interested, just ping us at marketing@fuelcycle.com.

Daryush

The most important thing: AI does not change the laws of statistics. If you're doing research at scale, you still want a representative sample of your population, recruited correctly. All of those principles remain unchanged. AI just makes the work with that sample a lot faster.

To the second part of the question, how AI helps recruit and build that population better: those are the areas we're building AI for now, at Fuel Cycle and across the industry in general. Do we have perfect AI for it today? No. Will we have an awesome AI that can do that work in a few months or quarters? Yes, we will. That's probably the next frontier for AI in market research. And if you want to join us in that journey as a design partner, ping us and we'd be happy to chat.

Daryush

I'm not a cybersecurity expert, so it's hard for me to opine on that side of it. But I would say the SaaS apocalypse is real. It has to do with the fact that users and consumers are demanding usage-based pricing as opposed to seat-based pricing. Coupled with market dynamics around SaaS companies and the amount of investment that's gone in, that makes it a real phenomenon, and it's not limited to market research.

Whether you pay based on usage or seats, you still want the highest level of trust, security, and protection. That part is unchanged, it's orthogonal to the pricing model. Speaking in generalities rather than about any particular company: established players have spent a lot of time, money, and blood and sweat building well-defended infrastructure, so they're sitting well to deliver on that promise. The real question is whether they can adapt fast enough to the dynamics of the SaaS apocalypse. That's ultimately a question of execution rather than strategy. They say execution eats strategy for lunch every day of the week, and that's still true here.

Daryush

100% Fuel Cycle. That's the only tool you need. You don't need anything else.

Victoria

We would love to show you what we're working on. It's coming out within the next few weeks, and I'll follow up personally to connect you with our sales and product teams so you can see everything that's going on.

And that's a wrap. Thank you everyone for attending, and Daryush, thank you for your expertise. As mentioned, we have a webinar specifically on AI moderation coming up in early September, and the registration link is in the chat. If you have any questions, reach out to marketing@fuelcycle.com and I'll direct you to the right expert. Thanks everyone!