Where AI Delivers the Highest ROI (It's Document Work)

Jun 24

The highest returns on enterprise AI are not coming from the tools that get the attention. They are coming from document work: contracts, invoices, reports, and compliance files. Across the research I could find, that conclusion is surprisingly consistent. The least glamorous work in the building is the part that pays for itself.

It is still early, and not every organization is fully transparent about what worked and what didn't. But the data points the same direction, and it isn't the direction most companies are spending in.

Where does AI deliver the highest ROI?

In short: Back-office document work. MIT's NANDA initiative studied hundreds of enterprise AI deployments and found that most pilots returned nothing measurable. The ones that paid off shared a trait: the biggest measurable returns came from automating document-heavy back-office work, not the customer-facing tools soaking up the budgets.

A separate January 2026 analysis from TechTarget reached the same place independently, noting that organizations most often see the highest ROI in invoice and accounts payable automation, with measurable drops in cost per invoice, shorter cycle times, and fewer errors.

What does the research say?

Two pieces of primary research anchor the case:

MIT NANDA, State of AI in Business 2025. Most generative AI pilots showed no measurable profit-and-loss impact. The returns that did show up concentrated in back-office document automation.
Deloitte. What the firm calls "cognitive automation," software that reads contracts, invoices, and forms, delivers the highest ROI per dollar of any automation category, with payback around nine months.

The case studies point the same way. These come from secondhand reporting and vendor write-ups, so treat them as directional rather than settled:

Invoice and document processing tops the ROI benchmarks across 2025 and 2026, landing at 400 to 520% return.
A&O Shearman's rollout of the legal AI tool Harvey reportedly cut contract review time by roughly 30%, with lawyers saving two to three hours a week each.
A survey of more than 100 in-house legal teams using GC AI reported 14 hours saved per lawyer per week and a 14% drop in outside counsel spend.
Ironclad documented a 314% three-year ROI on contract workflow automation in a Forrester study.

Why do most AI projects still fail?

Because companies buy the tool and skip the work. PwC's analysis is the cleanest answer I have seen: the tool itself is only about 20% of the value, and the other 80% comes from redesigning the work around it. The software shows up, the process stays broken, and the return never arrives.

Why does document work pay back when other AI doesn't?

Because you can measure it. I research how AI systems reason, and the conclusion here is not subtle. Document work is high-volume, repetitive, and already carries a known cost. You know what an hour of contract review costs and how many invoices come through in a month. When you point AI at that work, the before-and-after is easy to measure, so the return is easy to prove.

Open-ended pilots fail for the opposite reason. A general assistant that helps a little with everything is almost impossible to quantify. Nobody can say what it was worth, so it reads as a failure even when people like using it. The much-quoted 95% failure rate is not really a story about weak technology. It is a story about companies aiming AI at work they cannot measure.

How should you decide where to apply AI?

Start with the work, not the tool. A simple filter: pick a task that is high-volume, repetitive, and has a cost you can already name. If you cannot measure it today, you will not be able to prove the return tomorrow.

For most organizations, that filter points straight at their documents. The work is constant, the cost is real, and the output is checkable. The catch is that capturing the return depends on tools that do document work reliably and keep a person in control, rather than rewriting your files or quietly breaking formatting. That gap between the promise and the tooling is exactly why we built SquarePact, and why document work is still far from a solved problem. There is a lot left to build here.

Point AI at the work you can measure. The returns follow from there.

Frequently asked questions

Which AI use case has the highest ROI?

Document-heavy back-office automation: invoice processing, contract review, and similar work. MIT's 2025 research and Deloitte both place it at the top, and Deloitte puts payback for cognitive automation at around nine months.

Why do most enterprise AI projects fail to show ROI?

Most teams buy a tool and skip the process redesign. PwC's analysis attributes only about 20% of the value to the technology and 80% to redesigning the work around it. Projects that cannot be measured also cannot prove their return.

Is sales and marketing AI a bad investment?

Not bad, just oversubscribed relative to its measured payback. Much of corporate AI budget goes to front-office tools while the clearest measured returns have shown up in back-office document work.

What makes document work a good first AI project?

It is high-volume, repetitive, and already has a known cost, which makes the return straightforward to measure. That measurability is the single biggest predictor of whether an AI project can prove its value.

What is SquarePact, and how does it provide return on investment?

SquarePact is a Microsoft Word add-in for the kind of document work that has been shown to provide return on investment. The document work with the clearest ROI is also the work generic AI tools handle worst. It fixes formatting inconsistencies, holds document structure in place, and checks defined-term and cross-reference consistency, all inside Word. It works at the document's structural level, so it doesn't break formatting the way copy-paste or general-purpose AI often does. It shows you every change before applying it, nothing runs on its own, and nothing rewrites your content, and it runs inside your own Microsoft 365 environment with zero data retention. You can find it at squarepact.com . That combination (reliable on the work you can measure and built to keep a person in control) is what turns the returns in this research into returns you can actually capture.

Subscribe to the newsletter: https://squarepact.com/mailing-list