Monitoring and evaluation has always evolved with the tools available to it. We moved from paper-based surveys to digital data collection, from spreadsheets to integrated management information systems, and from periodic reporting to dashboards capable of showing results almost in real time. Artificial intelligence is the next shift. But it is likely to be more consequential than the changes that came before it..
The question for M&E professionals is no longer whether AI will enter our profession. It already has. The more relevant question is what we are going to do with it.
There are two equally unhelpful responses. One is to dismiss AI as unreliable, unethical or incapable of understanding the complexity of development. The other is to assume that because AI can process enormous quantities of information and produce sophisticated-looking analysis, it can replace professional evaluation judgement. Neither position is convincing. AI will change M&E. It should not be allowed to define it.
Where AI is already changing the work
Much of monitoring and evaluation involves work that is necessary but time-consuming: cleaning data, classifying responses, reviewing documents, comparing reports, identifying patterns, coding large volumes of qualitative material and converting findings into different formats for different audiences.
AI can substantially reduce the time required for many of these tasks. An evaluator working through hundreds of interview transcripts can use AI-assisted tools to identify recurring themes and contradictions. Large project document sets can be searched and compared far more quickly than through conventional manual review. Monitoring data can be examined for unusual patterns or emerging risks. Draft survey questions can be tested, long reports summarised, evidence organised and findings converted into briefing notes, dashboards or presentations. The real value is not simply that these tasks become faster. It is that the evaluator can spend less time processing information and more time asking what it means.
There is also considerable potential in monitoring. Traditional monitoring systems often tell us what happened after the event. AI-supported systems can increasingly identify patterns while implementation is still under way. If absenteeism is suddenly rising in a group of schools, disbursements are slowing in one region, beneficiaries are dropping out of a training programme or complaints are clustering around a particular service, these patterns can potentially be detected much earlier.
This moves monitoring closer to what it should have been all along: a management and learning function, not merely a reporting requirement.
But faster analysis is not necessarily better evaluation
This is where professional caution becomes essential. Evaluation is not simply the processing of evidence. An evaluator has to decide what questions matter, what constitutes credible evidence, whose voices are being heard, what alternative explanations need to be considered and what conclusions can reasonably be drawn from incomplete and sometimes contradictory information. These are judgement functions.
AI can identify a pattern in interview transcripts. It cannot automatically determine whether that pattern reflects reality, sampling bias, the way questions were asked, unequal power between interviewer and respondent, or an institutional narrative repeated across documents.
It can generate an elegant explanation of why a programme performed poorly. That explanation may be entirely plausible and entirely wrong. This matters particularly in international development. Context is not a footnote to evaluation. It is often the explanation. The same intervention may produce very different results in Bangladesh, Fiji, Nepal or Papua New Guinea because institutions, incentives, political relationships, administrative capacity and community expectations differ. AI may help organise the evidence, but it does not acquire decades of contextual understanding merely because it can summarise thousands of pages. That remains the evaluator's job.
The danger of false confidence
One of the more serious risks is not that AI produces obviously absurd answers. Those are relatively easy to detect. The more dangerous problem is that it can produce convincing answers that are slightly wrong. A fabricated reference, an incorrectly interpreted statistic or a causal explanation presented with unwarranted certainty can easily find its way into an evaluation report if the evaluator assumes that a fluent response is an accurate one.
M&E professionals therefore need a new professional discipline: verify before accepting. Sources must still be checked. Calculations must still be reproduced. Quotes must still be traced to interviews or documents. Statistical relationships must not quietly become causal findings simply because an AI system describes them confidently. The evaluator remains accountable for the evidence. That principle should not change.
AI also creates an ethical problem
Development programmes frequently deal with sensitive information: health records, household poverty, disability, migration status, employment, gender-based violence or information about vulnerable children and communities. Uploading such material into an AI system without understanding how the information is stored, processed or retained can create serious privacy and confidentiality risks.
There are also questions of bias. AI systems learn from existing information, and existing information reflects existing inequalities. Groups that are poorly represented in the underlying data may remain poorly represented in AI-assisted analysis.
Language creates another problem. AI capability is much stronger in some languages than others. An evaluation that relies heavily on automated interpretation may therefore unintentionally give greater analytical weight to evidence available in dominant languages. Technology can reproduce exclusion just as easily as it can overcome it.
Do not outsource the theory of change
There will be increasing temptation to ask AI to generate theories of change, evaluation matrices, indicators and even entire evaluation designs. It can certainly help. But a theory of change generated in seconds from a project document is not necessarily a theory of how change actually happens.
Good evaluation design requires discussion with programme managers, communities, government agencies and those expected to benefit from the intervention. It involves uncovering assumptions, identifying incentives and understanding why people behave as they do. AI can challenge a theory of change. It can identify missing assumptions. It can suggest indicators. It should not substitute for the process through which people themselves explain how change is expected to occur.
The profession itself will change
I suspect the greatest impact of AI on M&E will eventually be on the evaluator rather than the evaluation. Some skills that once distinguished evaluators will become less valuable because technology can perform them quickly. Producing routine descriptive tables, summarising documents or drafting standard survey instruments will increasingly be automated.
Other skills will become more important. Evaluators will need stronger critical judgement, better understanding of causality, greater ability to interrogate data, deeper contextual knowledge and stronger capacity to communicate difficult findings. They will also need enough understanding of AI to know when it is useful, when it is unreliable and when it should not be used at all. This does not diminish the profession. It could improve it.
For too long, parts of M&E have been dominated by compliance: indicators collected because a results framework required them, quarterly reports produced because somebody had to receive them, evaluations commissioned after key decisions had already been made. Used well, AI could reduce some of that administrative burden and allow M&E to return to its more important purpose: helping institutions understand whether they are making a difference and why.
What should M&E organisations do now?
The response should not be to purchase every new AI product that appears. Organisations first need clear rules about where AI can be used, what data can be shared, what outputs require independent verification and who remains accountable for the analysis.
Evaluators should start with relatively low-risk uses: evidence synthesis, document comparison, data cleaning, drafting, qualitative coding support and presentation of findings. More consequential uses, particularly those involving predictive decisions, vulnerable populations or causal conclusions, require much stronger safeguards. AI literacy should also become part of evaluator competency. That does not mean every evaluator needs to become a programmer or data scientist. But an evaluator who cannot interrogate an AI-generated analysis may eventually face the same disadvantage as an evaluator who cannot interpret a spreadsheet today.
Embrace, but do not surrender
There is a tendency when discussing technological change to choose between enthusiasm and resistance. M&E needs neither. It needs informed adoption.
Artificial intelligence can make monitoring faster, evaluation more responsive and evidence more accessible. It can help evaluators find patterns that would otherwise remain hidden and free experienced professionals from hours of routine analytical work. But AI cannot accept professional responsibility. It cannot understand political and institutional context in the way practitioners who have worked within those systems do. It cannot decide whose interests should matter most when evidence conflicts. And it cannot replace the ethical judgement at the heart of credible evaluation.
Dr Mak Khan is a Melbourne-based international development and evaluation practitioner with nearly four decades of experience across 35 countries. A former senior United Nations official and founding Director of Development Studies Program at the University of Melbourne, he advises international development organisations on monitoring, evaluation, learning, institutional reform and evidence-informed development policy, covering a range of sector, including education, public sector management and capacity building, social protection and climate change resilience. He is the Principal of RMDG Global (www.rmdg.com.au), a Melbourne based consulting practice.