21st Century Market Research: Process, Part 2

The Age of Automated Market Research

What happens when AI automates market research execution?

And what will researchers actually do once this happens?

The short answer? Those who adapt will use AI to become radically more efficient. Just as we did with the Internet.

The longer answer is below.

I have written previously that the research process is transitioning from a series of handoffs to people to a series of handoffs to apps. And eventually, this will turn into a single handoff to an agent or concierge, which will in turn coordinate different apps.

But let's focus on the near term for now. What does the process of running a study look like when it is made up of "a series of handoffs to apps"?

The App Layer

First, let's define what we mean by an "app" here.

I'm thinking of an application that has been built and trained on traditional domain expertise and to which a natural language front end interface has been added (i.e., users can prompt or query it like ChatGPT, Perplexity, etc.).

There will be apps that have been trained to create, for example, surveys and discussion guides, to program surveys, to QC and clean data files, and to tabulate, analyze and report on data.

But the ability to perform technical tasks correctly is only part of the value of these apps. They will also learn preferences and improve over time, by accepting user feedback and corrections. Just like people.

This is what AI is bringing to market research: the ability to have always on, lightning fast, organically improving domain experts for all aspects of the process.

How It Will Work

Let's assume that I have access to the apps needed to support quantitative survey work. I might load all of my past surveys into the app and create examples of what I consider "standard" or "best practice" on a number of topics (e.g., how to ask demographic questions, DQ and trap questions to include, etc.) and organize my examples by methodology, industry, topic, and client.

By rigorously curating inputs (e.g., ensuring only high quality surveys are used and clearly tagging examples) I can efficiently generate high-quality draft outputs for common use cases.

If I were, for example, starting work on a transactional CX program for a chain of quick casual restaurants, following the kickoff call I would refine my notes to make sure that the methodology, objectives, key issues, and desired length are all clear.

Then I would I upload these notes to my survey creation app. In the time it takes to check my email or grab a cup of coffee, I will have a survey draft that is at least as good as I might get from an average colleague. From there, I can edit it to get it into shape so I can share with the client.

And, as I make these edits and refinements to the draft, the app learns my preferences and style. Over time it will use this feedback to produce drafts that are closer and closer to my vision. Just like a colleague I might work with over some period of time.

The Future Process

Applying this to the entire quantitative survey process, we get something that looks like:

1. Kickoff meeting to confirm objectives and scope

2. Upload notes into an app to generate a draft survey

3. Review and refine the draft before sharing with the client

4. Once approved:

   • Upload the survey and objectives into an analysis app to create the analytical plan

   • Upload the survey to an automated tool to program the survey

5. Field the survey and apply real time fraud detection and data cleaning

6. Post field: 

   1. Upload the data to the analysis app, which generates a report, including:

      1. Predefined analyses from the analytical plan.

      2. Unprompted insights that reveal unexpected findings

7. Review and refine the report before delivering results.

(I have more thoughts on insight delivery and formats that improve client impact, but I’ll save those for another post)

Takeaways

As I review this future process, there are a few observations that jump out at me:

• The researcher remains in control of both inputs and final outputs (i.e., human in the loop)

• The researcher’s role will shift to a more "up front," design-oriented function, combined with "back end" storytelling, rather than a mix of design and execution (i.e., design + storytelling > execution)

• Initial investment in customizing apps by providing rigorously curated examples and feedback will be a key driver of success (i.e., garbage in, garbage out)

Let me close with a disclaimer: I can’t predict the future.

That said, I do believe that the outlines of what is coming are becoming visible.

And, I’m as excited and curious as anyone to see how this evolves.

How do you see AI changing the research process?