Decision-Bots May Finally Tell Us Which Data Is Actually Good

Insight Innovation Ventures recently wrote about the future role of agentic AI in market research and how research processes may eventually become integrated into what they call “decision infrastructure.” I say “decision-bots.”

Regardless of the terminology, there is an important idea buried in this vision: market research may stop being evaluated based on process metrics and start being evaluated based on outcomes.

That would be a huge change.

Today, market research is typically one step removed from business results. A research team conducts a study, delivers a report, and stakeholders make decisions. Months later, revenue is up or down, a product succeeds or fails, or a campaign works or doesn’t. The connection between the research and the outcome is often unclear.

As a result, we rarely know whether the data itself was actually helpful.

Decision-bots could change that.

Part of what makes this possible is the emergence of technologies like Model Context Protocol (MCP), which provides a standardized way for AI systems to interact with data sources, software platforms, and business systems. In practical terms, it becomes much easier for an AI agent to access survey data, historical reports, CRM systems, sales databases, and performance metrics all within the same workflow.

Imagine a company is considering a packaging redesign for Product A.

The decision-bot reviews survey results, past concept tests, sales data, competitor activity, pricing information, and historical research. It recommends a course of action. Six months later, the system sees whether the decision produced the desired outcome.

Did sales increase?

Did market share improve?

Did customer satisfaction rise?

The system can then adjust how much weight it gives to various data sources and research inputs in future decisions.

Of course, this gets complicated quickly. Revenue changes for many reasons. Maybe the company doubled its sales team. Maybe a competitor stumbled. Maybe macroeconomic conditions shifted. A useful decision-bot would need to account for all of these factors and attribute outcomes appropriately.

This is where some healthy skepticism is warranted.

Marketers have been trying to solve attribution for decades. Entire industries have been built around determining which advertising channels deserve credit for a purchase. Despite massive investments in analytics and measurement, attribution remains an imperfect science.

If we struggle to determine whether a Facebook ad or a television commercial caused a sale, determining how much credit a particular survey, focus group, or data provider deserves for a business outcome may be even harder.

Still, there is an important difference.

Traditional attribution systems are usually focused on understanding what happened. Decision-bots are ultimately trying to improve what happens next. Perfect attribution may not be necessary if the system can continuously learn which information sources tend to contribute to better decisions over time.

And if they work, market research may finally get something researchers have wanted for decades: a measurable connection between research activity and business outcomes.

The most interesting implication isn’t what happens to researchers. It’s what happens to data quality.

Today, data providers are mostly evaluated using process metrics:

• Cost

• Speed

• Incidence rate

• Fraud rates

• Response removal statistics

Those measures matter. But they don’t answer the question that clients actually care about:

Which data sources lead to better decisions?

In many automated survey marketplaces, buyers don’t even choose individual sample providers. The marketplace algorithm decides who fills quotas. Providers are largely rewarded for volume and fulfillment rather than demonstrated business impact.

Decision-bots could change that.

Imagine a system discovers that decisions informed by Data Provider Z consistently produce better outcomes than decisions informed by Data Provider U. The system could automatically increase spending on Provider Z and reduce spending on Provider U.

For the first time, we might have a mechanism for evaluating research quality based on outcomes rather than operational metrics.

That raises an uncomfortable possibility.

What if the system discovers that 80% of market research spending produces no measurable improvement in decision quality?

What if certain providers, methodologies, trackers, dashboards, or entire categories of research contribute little or nothing to better business outcomes?

The market research industry spends a lot of time discussing data quality. But most discussions focus on whether respondents are authentic, whether surveys are completed correctly, or whether quality checks are passed. Those are important considerations, but they are still proxy measures.

The real question is whether the information helped produce a better decision.

A future generation of decision-bots may be able to answer that question directly.

There is also a clear advantage for larger research organizations in building these systems. Individual clients make a limited number of major decisions each year. A model built on one company’s outcomes may not have enough data to learn effectively.

But a model trained across hundreds or thousands of companies, industries, and decisions could potentially learn much faster. Building a useful decision-bot will not be easy, but scale will matter.

The implications for researchers are obvious.

I’ve previously suggested that researchers move client-side if they are concerned about job security. But if decision-bots eventually manage large portions of a company’s research spend and decision-making process, client-side positions may be affected as well.

In the near term, I agree with Insight Innovation Ventures that there will likely be a transitional period. Researchers who can supervise AI workflows, identify bad data, challenge flawed outputs, and connect findings to business actions will remain valuable. The future researcher may increasingly look like a combination of analyst, data steward, AI supervisor, and decision architect.

That transition could last twenty years. It could last five. Nobody really knows.

My own view is that there are few aspects of market research that are fundamentally beyond the reach of capable AI agents. Eventually, decision-bots may be able to perform much of the work currently done by researchers, analysts, and even some business leaders.

That raises questions that are much larger than market research.

What happens to workers displaced by these systems?

Do executives deserve enormous compensation packages if models are making many of the most important decisions?

How comfortable are we allowing automated systems to influence public policy, healthcare, education, or other critical areas?

And if technology increasingly determines what actions should be taken, how do people continue to find meaning and pride in their work?

I don’t have answers to those questions.

What I do know is that the technology is advancing rapidly, and the financial incentives behind it are enormous. Decision-bots may not arrive tomorrow, but they are coming.

The possibility that they could finally tell us which data is actually good is exciting.

The possibility that they change far more than that is harder to sit with.

Thanks for reading.

CW