The short version
Three 2026 studies point to a related , but not identical problem. Airwallex's commissioned Forrester research found that manual intervention remains common across finance workflows. Grant Thornton found that very few organisations permit AI agents to make high-stakes decisions without human review. Deloitte, meanwhile, describes advanced adoption as redesigning workflows so AI handles routine execution while people retain judgment, exception handling, and oversight.
Together, they suggest that automating individual tasks is progressing faster than redesigning complete workflows. For an accountant managing multiple clients, every manual handoff can reduce the time actually saved.
What the data actually says
A Forrester Consulting study commissioned by Airwallex surveyed 1,279 finance decision-makers across EMEA, APAC, and North America. According to Airwallex's June 2026 publication, 84% said their finance workflows still required manual intervention to complete , even though one-third described those workflows as fully digital.
The other studies reinforce the broader challenge, although they don't specifically measure finance-workflow completion. Grant Thornton's 2026 AI Impact Survey of 950 C-suite and senior business leaders found that only 5% of organisations permit AI agents to execute high-stakes decisions without human review, while 60% limit them to moderate-risk task automation. Deloitte's State of AI in the Enterprise describes advanced organisations as redesigning workflows so AI can execute routine steps end-to-end while humans focus on judgment, exception handling, and strategic oversight — a description of the direction of mature adoption, not a measurement of how many finance teams have reached it.
Taken together, the studies suggest not conclusively prove that AI task adoption is moving faster than end-to-end workflow redesign.
Why this matters more for you than for a big corporate finance team
A large finance organisation may distribute the remaining manual work across specialised teams. If you manage 10–20 client accounts yourself, every handoff comes back to the same desk. The AI may have changed the nature of the work without fully removing it.
A common invoice-processing example looks like this:
AI reads the invoice, extracts amount and vendor → fine
You still check it against the PO
You still confirm the goods actually arrived
You still chase down mismatched fields
You still decide if something needs supporting documents
You still push it through approval and keep the trail for audit
That's not "the AI didn't work." It's that one step got faster while five steps around it stayed exactly as manual as before.
Does it run input to output without you picking it back up halfway? Reading a document isn't the same as closing a workflow.
When something doesn't match, is there an actual next step — or just a flag? Surfacing exceptions you still resolve manually hasn't removed the bottleneck, just labelled it.
Is the saving in clicks, or in hours? Fewer clicks per transaction adds up slower than you'd think if review still takes the same total time.
A build note from Rekons
One thing this reinforced for us while building Rekons: data extraction is only the easy part. The real challenge comes after , matching PO, delivery note, and invoice together, or reconciling a marketplace settlement against what actually lands in the bank. That's where exceptions, fees, and mismatched fields create manual cross-checking. We're building toward that full matching-and-exception chain, not stopping at document extraction.
Bottom line
If a tool only automates one step, it's a smart plugin. It only becomes a workflow when the exceptions, approvals, and audit trail travel with the data instead of landing back on your desk. Worth checking which one you're actually paying for.
Sources: Airwallex/Forrester, "Building An AI-Ready Finance Function" (June 2026); Grant Thornton, 2026 AI Impact Survey; Deloitte, "The State of AI in the Enterprise" (2026).
This piece reflects our own observations from building Rekons' reconciliation logic and reviewing publicly available industry research. We're not accounting practitioners ,our background is in enterprise IT and AI, built over 15 years across tech and e-commerce.
