What happened
Compliance teams are putting more money into artificial intelligence while staying sceptical about what comes out of it. That is the finding of new research from S-RM, From Screening to Decision-Making: The Next Phase of AI in Compliance, based on a global survey of 225 professionals working across compliance and ethics, procurement, supplier management and third party risk. Respondents came from the Americas, EMEA and Asia Pacific.
Seventy eight percent expect spending on AI enabled compliance and due diligence tools to rise over the next 12 to 24 months. Average annual spend is forecast to move from roughly $337,000 to $416,000, an increase of about 24 percent. Sixty eight percent have deployed AI somewhere in a compliance process already, though only 20 percent describe that adoption as extensive.
The trust numbers are less comfortable. Eighty four percent say trust and output reliability remain a challenge, half cite accuracy of outputs, and 47 percent report unpredictable or inconsistent results including hallucinations. Fifty eight percent say AI enabled due diligence still struggles to assess context, credibility and relevance well enough to replicate human led research.
Where the tools are used today is revealing. Adverse media screening leads at 48 percent adoption, with 38 percent already using AI for third party due diligence. A further 48 percent intend to add it there within 12 months, followed by customer due diligence at 42 percent, enhanced due diligence at 40 percent, and investigations at 38 percent. Only 12 percent expect AI to handle most due diligence with limited human oversight within five years, while 88 percent expect human expertise to stay central. Toby Thomas, director of research for corporate intelligence at S-RM, said the data shows a transition from screening toward judgement heavy workflows, where the context of the underlying information matters more.
Why this is a GRC story
The spend is moving toward the decision. Screening failures are visible and cheap to fix. Due diligence failures end up as an argument in front of a regulator. As budgets shift into the second category, the review gate has to shift with them.
Humans in the loop is a claim, not a control. A loop needs a named reviewer, a defined question and evidence of what that person changed. Without those three, oversight is a click.
Outputs become workpapers. If an AI summary supports a customer approval or a supplier onboarding, it is evidence in any later examination, and its provenance will be tested like a spreadsheet.
Time saved is measurable, judgement is not. Respondents report real reductions in manual research time. Far fewer report better decisions. Report the first number honestly and do not let it stand in for the second.
What to watch
Watch whether the 48 percent planning to add third party due diligence also write a review standard, since tooling without one only produces faster files. Watch, too, for regulators asking how an AI assisted decision was reached, which is the question that turns a vendor demo into a disclosure obligation.
The cheap preparation is documentation. One page per AI use case: what it reads, what a human checks, who signs, and what the file looks like if the model is wrong. Teams that write it before procurement tend to buy less and sleep better.
Attribution: Analysis based on Compliance Week and related public reporting. This article is original commentary, not a repost of the source material.