InackGuardWho this guide is for
This guide is for anyone deciding how screening, due-diligence and AI tools should fit into their compliance process: compliance leads, operations managers, founders and the people who review results every day. It explains why reports support decisions rather than making them, where human judgement is essential, how to use AI responsibly, and how to write decision reasons that stand up later. It is general information, not legal advice; your regulator and policies set the rules for your business.
Tools find; people decide
Screening and due-diligence tools are good at finding things: possible matches, news items, gaps in a contract. They are not good at judgement: whether a namesake is really the same person, whether an old story still matters, whether a risk is acceptable for your business.
That is why our reports say they are not a decision. The report gives you the evidence and the reasons; a person makes the call and records it.
Why it matters legally
In many places, decisions about people that are made solely by automated means carry extra obligations, and some uses of background information about individuals are regulated as consumer reports. Our reports are for business and compliance use, and must not be used for employment, housing, credit, lending or insurance eligibility.
Where AI fits
We use AI to read and summarise documents faster. It never sets a result. Every quote it produces is checked word for word against the source document, and anything that does not match is dropped before you see it. The result word comes from defined rules, not from a model's opinion.
What this asks of your team
A worked example
A report on a prospective distributor shows two adverse-media items: a five-year-old local dispute that was settled, and a recent regulator warning about a company with a similar name in another country. The tool cannot know which matters. The reviewer reads both, confirms the warning is about a different company, judges the settled dispute immaterial, and approves the distributor with both reasons recorded. A colleague reviewing a sample of decisions a month later sees exactly why.
How to keep human decisions consistent
Checklist
A deeper look: where judgement is needed
Identity: is this news story, listing or court case about the same person or company? Tools can show signals, but only a person can weigh them in context.
Relevance: does an old, resolved matter still affect the relationship? Policy sets the frame; judgement applies it.
Proportionality: is the risk acceptable for this product, this amount and this customer? That is a business decision, not a data lookup.
Exceptions: when you go against the usual rule, a person should approve it and record why.
A deeper look: using AI responsibly
AI helps by reading long documents quickly and pulling out the parts that matter. The risks are well known: summaries that drift from the source and confident statements that are wrong. Controls that work include quoting rather than paraphrasing, checking every quote against the source, keeping a person in the decision, and recording which version of the model and rules was used.
Questions people ask
Can we automate clear results? Many firms auto-approve results with no findings while sending anything else to a person. Make sure the clear result really means nothing was found, including no gaps in coverage.
What about automated declines? Be careful: automated adverse decisions about individuals can carry legal obligations. Keep a person in the loop.
How do we show a person decided? Record the reviewer, time and reason for every decision.
Can a tool ever decide on its own? For narrow, well-defined steps, some firms let rules approve results that show no findings and no coverage gaps, with a person reviewing a sample. Anything with a finding, a gap or a judgement about an individual should go to a person, and the reason for each automated rule should be written down and reviewed.
What to tell customers and partners
Be clear that your tools support decisions and that people make them. It builds trust, matches what regulators expect, and protects you if a decision is challenged.
Writing good decision reasons
A good reason is short, specific and checkable. Compare:
Good reasons make decisions consistent across a team, speed up reviews by managers and auditors, and protect the person who decided when a question comes months later. They also make it possible to spot patterns: if many reasons cite the same missing detail, you may need to collect it at onboarding.
Build a small library of example reasons for common situations, share it with new reviewers, and refresh it when your products or risks change.
Training new reviewers
Pair a new reviewer with an experienced one for their first weeks. Let them make decisions with a second review, then sample their decisions more often until they are consistent. Share real, anonymised examples of good and poor reasons, and explain the policy behind each. Training is part of the control: regulators expect people who make decisions to understand them.
Key terms
Summary
Tools find and people decide. Screening and review tools are good at surfacing evidence: possible matches, news items, missing clauses. They cannot judge identity, relevance or proportionality in your context, and some uses of personal information about individuals are regulated in ways that require a person in the loop. Use AI to read faster, never to set a result, and verify every quote it produces. Make decisions consistent with clear guidelines, good examples, second reviews for escalations and regular sampling. Write short, specific reasons that someone else could check months later. That combination of fast tools and accountable people is what customers, partners and regulators expect, and it is the safest way to scale a compliance programme without losing control of it.