Research has always moved carefully. Good reason for it. Our work runs on rigour, repeatability and trust, and those instincts are the reason anyone believes a number and insight we hand them. We are trained to question methodology, validate inputs, and avoid making claims that the evidence cannot support. Those instincts remain essential – but they can also become a liability when they shape how we approach innovation.

Artificial intelligence is advancing on a very different clock from traditional research technology. Capabilities can change meaningfully in a matter of months, sometimes weeks. New interfaces emerge, client expectations shift, and entirely new ways of collecting, analysing and accessing insight become viable before most annual planning cycles have finished.

Waiting for certainty is no longer the safe option. Waiting for certainty means letting the market move first. Waiting for certainty means falling behind.

Rigour and speed are not opposites

The industry keeps framing this as a trade. Move fast and accept risk or move carefully and protect quality. Wrong frame. The work is learning to move fast inside clear guardrails.

There are places we should never lower the bar:

• Data provenance and chain of custody
• Consent and the rights to use what was collected
• Identity verification, deduplication and fraud detection at the point of entry rather than as cleanup afterwards
• Attention checks and open-end review
• Representative, census-balanced coverage instead of whoever answered first
• An audit trail that makes a finding reproducible six months later.

If AI makes information faster and less trustworthy, we have not made research better. We have made it cheaper to be wrong.

That same certainty is not required before testing a workflow, an interface or a commercial model. Protect the integrity of the output, but experiment aggressively with how you get there.

Trust is infrastructure. It exists to give teams permission to experiment, not an excuse to avoid it.

This gets more valuable, not less, as the machines get better. The more machine generated text gets put into the world, the greater the premium value of a response that can be traced to a consenting, verified, identifiable human being.

Every system that reasons about people has to be grounded in evidence from actual people, collected under known conditions, time stamped, and defensible when somebody asks where it came from: every response scored, every supplier vetted and every answer traceable to a verified human.

The interface to research is changing

For years, much of research technology has focused on improving individual stages of a familiar process: build the survey, find the audience, collect the data, analyse the results, deliver the report. AI challenges that process.

When software can help formulate questions, interact conversationally with respondents, synthesise large volumes of unstructured material and surface answers inside the tools where decisions are being made, the value of research goes beyond completing a study efficiently. It increasingly sits in making trusted human intelligence available at the moment a decision is being made.

Model context protocol (MCP) is one indication of where this is heading. By creating a standardised way for AI systems and agents to connect with external tools and data sources, protocols like MCP could make research less dependent on someone opening a dashboard, downloading a report or manually moving information from one system to another. Instead, an AI agent could potentially access the appropriate research evidence as part of the workflow in which a decision is already taking place.

That shift has significant consequences. We should expect new interfaces, new expectations around turnaround time and new business models. We should also expect humans to remain essential. As machine-generated outputs proliferate, grounding those systems in credible, representative human data becomes more important.

The opportunity is not to automate people out of research. It is to remove friction between a question, the right human evidence, and a decision.

Permission to fail

Telling a team to innovate with AI while holding every experiment to the approval standard of a mature product is not innovation. It is theatre with a budget line.

The opportunities that matter here will look small and strange at first. Researchers talking to data instead of reading dashboards. Agents starting parts of a workflow. The line blurring between a discrete study and an always-on stream of evidence. Most of these experiments will not become businesses. That is the cost of finding the ones that do.

Set the boundaries on cost, risk and quality. Then get out of the way.

The organisations that adapt fastest will not necessarily be those with the biggest arsenal of AI tools but rather those that utilise their tools most effectively to shorten the distance between an idea, a test and a real user reaction.

Research has spent decades perfecting ways to reduce uncertainty for other businesses. We now need to become more comfortable operating with some uncertainty ourselves. It’s time to stop optimising yesterday’s model; it’s time to start experimenting aggressively to build the technology of tomorrow.

Patrick Comer is chief executive at Cint