Blogs

What comes after AI in consumer research? The calculator thesis

Statements
9
When the spreadsheet arrived, bookkeeping shrank and accounting grew. We think AI is doing the same to consumer research: answers get cheap, and real, observed, deeply qualitative understanding of people becomes the scarce, valuable thing.

What comes after AI in consumer research?

The calculator thesis: when answers get cheap, understanding gets valuable

Written by: Mark Nguyen, ND Research

Originally Published: November 21, 2026

Labels: Statements, AI-assisted, Continuously Maintained

Topics: Future of market research, AI in consumer research, Synthetic respondents, Qualitative research, Feed exposure intelligence

TLDR: When the spreadsheet arrived, bookkeeping shrank and accounting grew. We think AI is doing the same thing to consumer research. Writing questionnaires, tabulating answers and summarizing them are getting cheap. What becomes scarce, and valuable, is real, observed and deeply qualitative understanding of people. That is the work we are building ND Research around.

Key Ideas:

  1. New calculating tools rarely remove a profession. They remove its clerical layer and push the work up toward judgment. After VisiCalc, the US had about 400,000 fewer accounting clerks but about 600,000 more accountants (NPR Planet Money, via Tim Harford) [1].
  2. Consumer research has its own "bookkeeping": questionnaire drafting, fieldwork admin, coding open-ended answers, crosstabs and topline decks. AI is absorbing that layer quickly. In Qualtrics' 2026 trends study of 3,000+ researchers, 78% expect AI agents to run more than half of research projects by 2028 [2].
  3. When answers become cheap to generate, three things become scarce: real observed inputs, context that only comes from people, and judgment tied to a decision.
  4. Personalized feeds make observed data more important, not less. Two people can open the same app at the same time and get a very different internet. A synthetic respondent cannot tell you what an algorithm actually served someone yesterday.
  5. Our bet is on new, consented ways of observing people (studying their feeds, their digital diets and their lives on film), not on replacing them. We could be wrong about parts of this, and we list where below.

‍

Preamble

Most of the conversation about AI in market research right now is about speed: faster surveys, faster analysis, synthetic panels that answer in seconds. All of that is real, and we use a lot of it ourselves (this article was drafted with AI help, then rewritten by hand).

But speed is a strange thing to build a strategy on, because everyone gets it at roughly the same time. The more interesting question, I think, is what becomes valuable once speed is free. History has a pretty good answer, and it starts with a $99 piece of software.

‍

What happened to accounting when the spreadsheet arrived?

Before 1979, a spreadsheet was a literal sheet of paper. Changing one assumption in a financial model meant erasing and re-adding columns by hand, which could take a whole day. VisiCalc, the first electronic spreadsheet, did that work in seconds [1].

The clerical layer of accounting shrank fast. Planet Money found that in the decades after VisiCalc launched, the US had roughly 400,000 fewer bookkeeping and accounting clerks. Over the same period, it had roughly 600,000 more accountants [1]. Once arithmetic was cheap, people asked more questions of their numbers (what if we charged a little less, what if we borrowed more?), and someone had to interpret the answers.

That arc kept going. Bookkeeping became accounting, and accounting grew a strategic layer: forecasting, scenario planning and advising the people who make decisions. The calculator didn't end the profession. It moved the profession up the value chain.

‍

What is the "bookkeeping" of consumer research?

Every research project has a clerical layer. In our experience it looks like this:

  • drafting and translating questionnaires and discussion guides
  • fieldwork admin: quotas, reminders, incentives, cleaning bad responses
  • coding open-ended answers into themes
  • building crosstabs and significance tests
  • turning tables into a topline deck
  • summarizing desk research and past studies

AI is now good at most of these. Interview moderation, coding and first-draft reporting are increasingly automated, and synthetic respondents can simulate how an audience might answer a question before anyone is surveyed [3]. Qualtrics' 2026 Market Research Trends report found that 78% of researchers expect AI agents to run more than half of all research projects by 2028 [2].

If that happens, the cost of producing a "finding" falls toward zero. Which raises the question that matters for every insights team and research firm.

‍

If AI can generate the answer, what becomes scarce?

We think three things.

1. Real, observed inputs. Models are trained on what already exists. To say something new about people, you need new data about people, collected from the world rather than predicted from text. The Market Research Society's managing director put it plainly: to create synthetic data, "you still need really, really good human data" [4]. When we spoke with synthetic-data vendors this year, they said the same thing about their own products [5].

2. Context that only comes from people. Hesitation, contradiction, the thing someone says on the way out the door. Deep qualitative work (ethnography, long interviews, documentary film) captures the why behind behaviour. It's slow and expensive, which is exactly why it doesn't get commoditized.

3. Judgment tied to a decision. In 19 recorded conversations with agency, research and ad-tech people in 2026, the most common need (raised in 9 of 19) was not more data. It was insight tied to an action or a KPI [6]. "Interesting" no longer clears the bar. Someone has to decide what a finding means for this brand, this launch, this creative brief.

‍

Why does personalization make observed data more important?

Because the shared media environment that a lot of research quietly assumed no longer exists.

Think: two people open the same app at the same time. One gets gym content, debt-payoff creators and a car ad. The other gets K-beauty tutorials, comedy dog videos and the same car ad. Same app, same ad, very different internet. If you only measured the ad, those two people look identical.

Recommendation systems now choose much of what each person sees, individually. In our 2026 conversations, 6 of 19 people said feeds have become so personalized that today's research tools miss what people actually see [6]. You can ask someone what they saw (they will mostly forget), and you can ask a model what they probably saw (it will guess). Or you can look, with their consent.

‍

What does the research stack look like after AI?

Here is how we'd split the work, roughly:

  • Let AI do: drafting instruments, cleaning and coding data, crosstabs, first-draft reports, desk research, scenario exploration with synthetic respondents.
  • Keep people on: deciding what question is worth asking, deep qualitative fieldwork, interpreting results against a real decision, and saying when the data doesn't support the story.
  • Invest in new observed data: consented, privacy-preserving ways of seeing what people actually encounter and do, especially inside personalized feeds where traditional tools can't reach.

The accounting parallel holds here too. The firms that did well after the spreadsheet weren't the ones with the fastest clerks. They were the ones who used cheap arithmetic to ask better questions.

‍

Where does ND Research fit?

ND Research is a Toronto consumer research company. Our work sits on the "new observed data" and "deep qualitative" side of that stack:

  • Tapestry (feed exposure intelligence). Consented, paid participants share screenshots of their own signed-in social feeds. AI classifies each post (platform, creator, content category, ad versus organic, format) and links it to who that person is. Social listening sees what people post. Tapestry sees what people are recommended. The method is covered by a published US patent application (US 2026/0087516 A1) [7].
  • Digital diet studies. Measuring what people actually consume across platforms, not just how long they spend there.
  • Documentary films on users. Turning persona diagrams into real people on camera.
  • System dynamics modelling. Mapping cause and effect so teams can anticipate shifts rather than react to them.

We also still do the classic work (interviews, usability tests, surveys), because good observed data is most useful next to good questions.

‍

What could make us wrong?

A few honest possibilities:

  • Synthetic models might get much better. If models gain access to large, fresh behavioural datasets, the gap between predicted and observed answers could narrow for many routine questions. (We'd still expect them to need observed data underneath, but the line could move.)
  • Platforms could open up. If social platforms gave researchers rich, audience-level exposure data, new collection methods like ours would matter less.
  • Exposure is not attention. Seeing what a feed served is not the same as knowing what someone noticed or what changed their mind. Our own methods are early and independent validation is still ahead of us.

We'd rather name these than pretend the future is settled.

‍

FAQ

Will AI replace market researchers?

We don't think so, but it will replace a lot of what market researchers currently spend time on. The spreadsheet removed many clerical accounting jobs while growing the number of accountants [1]. We expect a similar shift: fewer hours on tabulation and reporting, more on fieldwork, interpretation and new data sources.

Are synthetic respondents reliable?

They are useful for exploring ideas and boosting small segments inside a large enough real sample. In our 2026 conversations, even synthetic-data vendors said their products need real, observed data underneath and shouldn't be used to invent things nobody observed [5].

What is feed exposure intelligence?

It's our name for measuring what personalized social and recommendation feeds actually show people, item by item, linked to who those people are [8].

What should an insights team invest in now?

Automate the clerical layer, then spend the saved time on the things AI can't produce: new observed data, deep qualitative work and tight links between findings and decisions.

Who is ND Research?

A Toronto consumer research company founded by Schulich School of Business alumni, building Tapestry and offering research for brands, agencies and technology companies. Contact: mark@nguyendigital.co.

‍

How to cite this

Nguyen, M. (2026, November 21). What comes after AI in consumer research? The calculator thesis. ND Research. https://www.nguyendigital.co/blog-posts/what-comes-after-ai-consumer-research

‍

List of References

(Continuously Maintained)

[1] Harford, T. How computing's first "killer app" changed everything. BBC News, drawing on NPR Planet Money, "How The Electronic Spreadsheet Revolutionized Business" (Goldstein, J., February 27, 2015). https://www.npr.org/transcripts/389585340

[2] Qualtrics. 2026 Market Research Trends report (3,000+ researchers). Press release, November 2025. https://www.qualtrics.com/news/research-teams-not-using-ai-are-four-times-more-likely-lose-organizational-influence/

[3] Greenbook. How AI Is Changing Market Research in 2026: What Practitioners Need to Know. https://www.greenbook.org/insights/artificial-intelligence-and-machine-learning/how-ai-is-changing-market-research-in-2026-what-practitioners-need-to-know

[4] Market Research Society. Boughton, L. Synthetic respondents in market research: risk or reward? (quoting Debrah Harding, Managing Director, MRS). https://mrs.org.uk/blog/operations/synthetic-respondents-in-market-research-risk-or-reward

[5] ND Research. Nguyen, M. (2026, October 23). Will synthetic respondents replace real consumer research? https://www.nguyendigital.co/blog-posts/will-synthetic-respondents-replace-consumer-research

[6] ND Research. Nguyen, M. (2026, October 23). What does the insights industry actually struggle with? Notes from 19 recorded conversations in 2026. https://www.nguyendigital.co/blog-posts/what-strategists-and-insights-teams-struggle-with

[7] United States Patent and Trademark Office. Patent application publication US 2026/0087516 A1, published March 2026.

[8] ND Research. Nguyen, M. (2026, October 23). What is feed exposure intelligence? https://www.nguyendigital.co/blog-posts/what-is-feed-exposure-intelligence

Still curious?

Incomplete and outdated insights are a thing of the past. We give you a front seat into how your customers live, work and interact. We weave together diverse inputs into a complete picture of your product.