AIRON

AI adoption in research management is moving quickly. Across universities, research institutes and funding organisations, people are experimenting with ChatGPT, Microsoft Copilot, Claude, AI agents and workflow automation. The appeal is obvious: research management is full of repetitive, information-heavy work, and AI can already help with summarising guidance, drafting communications, reviewing information and preparing reports.

But the biggest challenge is no longer whether AI can help with individual tasks. It is whether organisations are using AI to improve research support, or simply adding new technology to existing ways of working.

A 2026 EDUCAUSE study found that 94% of higher education professionals surveyed had used AI for work in the previous six months, but only 54% were aware of institutional policies or guidelines for that use. Just 13% said their institution was measuring return on investment from work-related AI tools. EDUCAUSE

That gap between experimentation and institutional readiness is becoming increasingly important.

The common mistake: starting with the tool

One of the most common mistakes is to begin with the technology. An organisation introduces Copilot, ChatGPT or an AI platform and immediately asks what can be automated.

That can produce useful improvements, but it can also mean that AI is layered onto processes that were already inefficient.

Imagine a researcher who finds a funding opportunity and wants to know whether they are eligible, when costing should begin, what approvals are required and who they need to contact. AI could help a research manager draft the response much faster, but the researcher still needs to know whom to approach and the research manager still needs to gather information from different places.

The task is quicker, but the service itself has not really changed.

A better question is: how could the funding-support service work better if AI were part of its design? Perhaps the researcher could start in one place, provide the funding call and some basic project information, and receive guidance on eligibility, internal steps and likely deadlines, while anything unusual is routed to the right person for review.

At that point, AI is no longer simply automating a task. It is helping redesign the service.

Why this matters

The OECD has identified many of the same problems in public-sector AI adoption more broadly: skills gaps, difficulty accessing high-quality data, limited practical guidance, risk aversion, legacy systems and weak measurement of results. It also notes that many AI initiatives remain stuck at pilot stage because organisations struggle to move from experimentation to implementation. OECD

Research management faces similar conditions. Our work often cuts across finance systems, grant-management platforms, HR data, contracts, policies and local spreadsheets. Automating one part of that environment does not necessarily solve the underlying problem.

A costing process, for example, may involve interpreting funder rules, gathering project information, modelling staff costs, entering information into systems, seeking approvals and resolving unusual cases. AI might help draft a budget justification, but if information still has to be entered several times and approvals still happen too late, the main problem has not been addressed.

This is why AI adoption needs to move from isolated tasks to whole workflows.

Keep human judgement at the centre

Research management also depends heavily on professional judgement. A researcher may appear eligible for a scheme until one condition changes the answer. A cost may appear allowable until the project context is considered. A collaboration may raise contractual or governance issues that are not obvious at first.

AI can help gather information, compare criteria and flag possible concerns, but that does not mean it should make every decision.

In many cases, the most useful role for AI is to bring the right information together and help a person reach a decision more quickly.

That means human escalation needs to be part of the service design from the beginning. A good AI-enabled service should be able to recognise when a case is routine, when it is uncertain and when it needs specialist review.

Sometimes the best AI response is simply: this needs human review.

That is not a limitation but a responsible design.

Sometimes the problem is the service, not the question

There is another important reason to think about redesign.

If a research office receives hundreds of questions about internal deadlines, approval routes, costing forms and responsibilities, an AI assistant could answer many of them.

But it is also worth asking why people need to ask those questions so often.

Perhaps the information is difficult to find. Perhaps responsibilities are unclear. Perhaps the service is organised around institutional structures rather than around the researcher’s experience.

In that case, the long-term solution may not be a better chatbot. It may be a simpler service.

This is where AI becomes useful not only as a tool, but as a reason to re-examine processes we may have stopped questioning.

What good service redesign looks like

Service redesign does not have to mean a major transformation programme. It can begin with one persistent problem.

A team can take a workflow that regularly causes frustration, look at how it actually works and identify where time is lost, where information is duplicated and where people become confused. AI can then be introduced where it genuinely adds value: structuring incoming requests, identifying missing information, retrieving approved guidance, flagging unusual cases or preparing information for review.

The important principle is that the technology should support the service design, not determine it.

Success should also be measured differently. Instead of counting how many people use an AI tool or how many agents have been created, organisations should look at whether researchers receive support faster, whether staff spend less time on repetitive work, whether risks are identified earlier and whether people have more time for complex work that genuinely requires expertise.

What happens if we do not redesign the work?

The risk is that organisations invest heavily in AI but achieve relatively little.

We may end up with many pilots, chatbots and agents sitting on top of fragmented processes. Staff may save time on individual tasks but continue struggling with poor systems. Researchers may receive faster answers while still finding services difficult to navigate.

Over time, organisations may conclude that AI has failed to deliver the transformation they expected. The technology may not be the problem.

The problem may be that the work around it never changed.

Why we need to talk about this now

This is the right moment because AI adoption is still taking shape.

The OECD’s 2026 Digital Government Outlook makes the same broader point: AI can only scale effectively when the foundations around it are strong, including data, digital infrastructure, governance and workforce capability. Where those foundations are fragmented, AI can amplify existing weaknesses rather than solve them. OECD

Research organisations therefore have a choice. We can use AI mainly to make existing tasks faster, or we can use this period of experimentation to rethink how research support should work.

Research managers should be central to that conversation because they understand the workflows, the exceptions and the places where professional judgement matters.

The next stage of adoption should move beyond asking “What task can AI automate?” and towards asking “How could this service work better if AI were part of its design?”

The goal should not be a research organisation full of faster emails and more AI tools. The goal should be better research support.

The real AI opportunity in research management is not simply automation. It is service redesign.