When MRS Delphi published the BEST (Boost, Expand, Shift, and Transform) framework in late 2023, it moved the conversation about generative AI beyond hype and provided practitioners with a practical framework for using it to support, extend or change research practice.

Much has changed since 2023. AI adoption is no longer limited to experimentation around isolated tools or use cases. It is becoming embedded in day-to-day research work: helping to design materials, structure analysis, summarise evidence, draft outputs and manage parts of the workflow.

As adoption has widened, AI’s function within research has also become more varied.
This suggests the framework would benefit from a second dimension: one that captures the role AI plays within the research workflow.

BEST gives us a way to identify the kinds of situations where AI can support, extend or transform research practice. A role-based dimension would describe what AI is doing functionally or structurally within that work.

Based on what we are seeing in practice across the industry, including in our own work at o-x.ai, I believe there are four broad roles AI can play in the research process:

1.  It can deliver parts of the research process, including qualitative coding, summarisation, analysis and first-draft reporting.
   
   
2. It can generate research inputs, including synthetic data, augmented samples and simulated stimuli.
   
   
3. It can manage workflows, from elements of project management to work that might previously have sat with junior or mid-level researchers.
   
   
4. It can build research tools, such as questionnaire scripters, coding aids, analysis utilities and dashboards.

These categories add a workflow layer to BEST. The original framework helps practitioners identify where AI may create value across different kinds of research problems. A role-based view brings the focus closer to the workflow itself: the function AI is performing, the kind of judgement it requires and the form of oversight researchers need to provide.

The value is insight, not process

This distinction will become more important as the technology matures. Early debates understandably focused on what AI could and could not do, but capability is a moving target. Model performance is improving, and with the right context, prompting and self- diagnostic systems, hallucination risk can be significantly contained.

Agentic systems are making multi-step workflows more realistic. Frameworks built mainly around yesterday’s strengths and weaknesses will age quickly.

Much of the discussion has focused on what AI is suited to do: which tasks it can support, where it performs reliably, and where its limitations still matter. As capability improves, that question starts to lose some of its usefulness. The question the industry will soon need to answer is what we want to keep human in research, and what skills we need to develop to do that well. This is less about what AI can do, and more about what we choose to keep human because it defines the quality and value of research.

This brings the discussion back to value. Much of the industry’s current use of AI is still focused on micro-productivity: accelerating individual tasks within a research operating model that otherwise remains largely unchanged. Faster transcript turnaround, quicker summaries, cleaner first drafts. These are useful gains, but they do not realise the full potential of AI in our sector.

The bigger opportunity is to rethink the operating model itself. That means asking what the work should become, not just how to do today’s work faster. We have historically been paid for much of the visible labour of research: interviews conducted, transcripts coded, analysis completed, reports written.

AI can now perform many of these activities quickly and competently, with the requisite supervision. This should not be seen as a threat to the sector. It is an opportunity to shift more of our time and expertise towards the work where research creates most value: providing insight and answering clients’ questions, rather than executing process.

The value of research lies in framing the right question, understanding context, interpreting evidence, applying judgement and building trust with clients. If AI reduces the burden of process, it gives researchers more room to focus on those higher-value activities. It also creates an opportunity for the sector to occupy a more advisory role, closer to decision-making , but from a distinctive position, grounded in evidence, methodological rigour and expertise in understanding people.

Building judgement, not just capability

A future in which AI takes on more of the research workflow has direct implications for people, skills and roles. The sector will need to choose which parts of research should remain expert-driven: not because AI cannot assist with them, but because they define the quality, credibility and value of the work.

The issue is not simply keeping a human in the loop, which should be standard, but deciding where human expertise should lead: in framing the problem, making methodological choices, interpreting evidence, challenging weak conclusions and building the client trust that gives research its value.

This has consequences for the skills researchers need, and for how careers develop.
Critical thinking, methodological confidence and the ability to challenge fluent, persuasive outputs are basic requirements. They will need to know when an AI-generated answer is plausible but thin, when an interpretation is over-extended, and when the evidence does not support the conclusion.

When it comes to progression, if AI absorbs more of the executional work through which junior researchers have traditionally learnt, the sector will need to be more deliberate about how people develop judgement.

The future researcher may spend less time producing every element of the process by hand, but they will need to be better at directing the work, questioning its outputs and knowing not to outsource the thinking to the machine.

The next phase of AI in research will require frameworks that are useful but provisional. The sector needs enough structure to act with confidence, and enough pragmatism to adapt as the technology, the work and the role of the researcher change.

Dr Zsolt Kiss is founder at o-x.ai

• The full Delphi report, ‘Who owns understanding? How AI is reshaping value, expertise and accountability in research’ is accessible via the MRS website.