Product1 publisher3 min readPublished
Workiva's survey finds 84% of executives willing to skip human review of an AI-written annual report
Workiva's research finds executive appetite for unreviewed AI reporting running well ahead of the data those reports would sit on, and the interview it came from was recorded at the vendor's own event.
The Product Desk · Product desk
What happened
- The same research found that only 11% of executives considered their own data sufficient for AI to use.
- Soter gave the figures in an interview with Krista Case and Alison Kosik at Workiva's Amplify event, broadcast on theCUBE, SiliconANGLE Media's livestreaming studio.
- SiliconANGLE disclosed that theCUBE is a paid media partner for the Amplify event and that Workiva has no editorial control over the coverage.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- constraint Where a team cannot show where a figure originated or how it was reviewed, one wrong number stops being an isolated error and becomes a control failure someone has to chase backwards through the workflow.
- exposure The signature at the end of the process still belongs to a person, so the rollout question in finance is what evidence that person receives before signing, not how capable the model is.
- decision Workiva's framing puts corporate confidence ahead of operational readiness. That leaves controllers picking an order: the drafting automation first, or the lineage and approval evidence underneath it.
Somebody signs the financial statements, and that person has always needed four answers about every figure. Steve Soter, vice president and industry principal at Workiva, gave them in the language of his old job: "it was really important for me to know where the data was coming from, who touched it, what happened to it, how did it get reviewed and approved?" [5][3]
Soter said the finding that surprised him most was "that 84% of executives said that they were at least somewhat willing to trust AI to generate an annual report without human review" [1]. At least somewhat willing. That is stated tolerance collected in a questionnaire, and the published interview does not include the survey's sample size or field dates [14].
The other number is the one a controller has to plan around. Workiva's research found that only 11% of executives considered their data sufficient for AI use [2]. Take 84 and subtract 11: 73 points separate willingness to remove the reviewer from confidence in the inputs that reviewer would be checking [10]. Soter looked backwards at the smaller figure. "It makes you wonder, how bad was the data before we were even having this AI conversation?" he said [8].
He put the sequencing plainly too: "accelerating a process, if you don't have it grounded by those things that we discussed, those four things, then speed doesn't become an asset. It really becomes a liability," Soter said [7].
Workiva sells the safeguards. The company is applying its established reporting controls to AI-assisted processes, and the research behind both numbers is its own [11]. Both figures are self-reports from the same executive population, so a finance team that runs the same two questions past its own controllers gets a number that describes its own close.
The 2x2 worth drawing has two axes: whether a named person can reproduce the approval trail for a given figure inside a day, and whether the output leaves the building. Internal drafting with a reproducible trail is where AI belongs now, and the cost of being wrong is a rework cycle. External output with a reproducible trail can take AI in the drafting seat as long as the lineage travels with the review. Internal work with no trail is a cheap place to learn what the tool gets wrong. External output with no trail puts a signature on top of a chain nobody can walk back, and Soter described that case in the interview: "If that human signs off on it, but yet trusted that AI had done everything that it was supposed to do and done it correctly, again, if that's not the case, that could be a big risk," he said [13].
What to watch
- Whether Workiva publishes the sample size, field dates and question wording behind the 84% so operators can compare it with their own executive population.
- Whether the 11% data-sufficiency figure moves in the next round of the same research.
- Whether any company discloses AI-drafted content in a filing and how its auditor treats the lineage evidence.