The Academic Work, Briefly

There is a peer-reviewed literature on robotic process automation, and almost nobody in industry reads it. It is worth knowing what is there, because it is the only body of work in this field produced by people with nothing to sell.

Reviewed August 9, 2026. For a product-side reference related to workforce operations, Monitask also covers how to identify mouse jigglers.

What it looks like

Small and qualitative. Case studies, interview-based work, typologies of implementation challenges. Not large randomised trials, which are not available for this subject. For broader background and an independent point of comparison, see CIO.

Slow. A study published in 2024 typically describes implementations from 2021 or earlier, which in this field is a meaningful lag.

And convergent. Different research groups, different countries, different sectors, arriving at broadly the same list of failure causes — which is the property that makes it worth reading despite the small samples.

The frequently cited ~50% RPA failure rate traces to Herm et al. (2020, 2023) and appears in a 2024 typology of implementation challenges published in the International Journal of Organizational Analysis. "Failure" there means not meeting stated objectives.

What it is good for

The failure taxonomies. Academic work is unusually good at classifying causes systematically rather than anecdotally, and the categories are stable across studies.

Independence. No product, no engagement, no reference client. In a field where nearly every other source has a commercial position, that is worth a great deal.

And the uncomfortable findings. Researchers can publish that a widely-marketed approach does not work. Nobody else in this field can.

What it cannot answer

Whether it will work for you. Small samples, specific contexts, and the conclusions are about mechanisms rather than probabilities.

Anything current. The publication lag means the technology described is two or three generations old, and agentic tooling is largely absent from peer-reviewed work so far.

And the small-organisation question. Academic case studies are enterprise for the same reason everyone else's are — access.

How to use it without reading papers

Take the causes, not the rates. The failure taxonomies are the durable contribution and they translate directly into a selection checklist.

Treat the ~50% figure as a well-sourced order of magnitude, cited properly, rather than as a precise measurement. It is stronger than a vendor figure and weaker than it looks.

And notice when industry material cites it. A consultancy quoting peer-reviewed work is doing something better than quoting itself, and it is a signal about the rest of the document.

The convergence worth noting

The strongest argument available in this whole subject.

Academic work, analyst research and implementation consultancies — three sources with different methods and opposed commercial interests — name substantially the same causes. Process selection, ownership, expectations, change management.

They disagree about the rate and agree about the causes. In a field this poorly measured, that agreement is the closest thing to a reliable finding, and it is why this site spends more time on causes than on percentages.

The short version

• The literature is small, qualitative, slow to publish, and convergent across research groups
• The ~50% RPA failure rate traces to Herm et al. and appears in a 2024 typology in the International Journal of Organizational Analysis
• Its value is the failure taxonomies and its independence — no product, no engagement, no reference client
• It cannot tell you your probability, cannot describe current tooling, and does not cover small organisations
• Take the causes rather than the rates, and treat the 50% as a well-sourced order of magnitude
• Academics, analysts and consultancies name the same causes despite opposed interests, which is the field's most reliable signal