iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
For AI used in consequential or regulated decisions, ask a vendor to reproduce a specific past decision and show the evidence and process behind it. A polished explanation generated afterward is not the same as a record of the sources and decision path that actually produced the result.
What should buyers ask about an AI decision?
Graham French, chief technology officer of UnlikelyAI, says procurement conversations he has encountered in financial services and insurance raise four practical questions:
- What explanation shows how this output was reached, and what evidence supports the answer?
- Can the system’s reasoning be reconstructed months later if the decision is challenged?
- Which source material and rules contributed to the outcome?
- Who is accountable if the result is wrong?
These are questions French reports from procurement conversations, not a statistically representative survey of all AI buyers. They are nevertheless useful prompts for evaluating a system whose decisions may need to be reviewed or defended.
Recommended Free Tools
Why a fluent explanation may not be evidence
A language model can produce a plausible account of why it gave an answer. Unless that account is tied to a contemporaneous record, it does not establish which documents the system used or what process actually led to the decision. A narrative generated after the fact may describe a rationale without proving that it was the rationale used.
For audit and challenge purposes, look for a record that connects the decision to its inputs and method: the source material considered, the rules or reasoning applied, and the resulting path to the output. The central test is not whether a vendor can explain AI in a presentation; it is whether the vendor can reconstruct a real decision made earlier.
How to test a vendor’s answer
Use a consequential case from your own workflow, with suitable privacy and security controls. Ask the vendor to reproduce the result and demonstrate the record supporting it. French recommends this as a practical evaluation question; it is not a formal standard or a published benchmark.
Rank #2
- Select a past decision. Choose a specific case that can be evaluated against the original input and known outcome.
- Request a replay. Ask the vendor to reproduce the decision and identify the source material, relevant rules or reasoning, and sequence that led to the result.
- Probe variations. Change a meaningful input or use an edge case to see whether the system behaves as expected and whether the changed path is visible.
- Ask about accountability. Clarify who investigates errors, who can correct an outcome, and how responsibilities are divided between your organization and the vendor.
- Check how change is handled. Find out how policy updates reach the system, who approves them, and how the organization can verify their effects.
Keep the evaluation grounded in your actual requirements: evidence of a decision path, performance on relevant cases, ownership of errors, and the effort required to maintain the system. A general claim of “explainability” does not answer those questions.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWhat explicit rules can add—and what they cost
French proposes a neurosymbolic approach as one possible design: a language model processes unstructured information, while an explicit rule system makes the decision and produces a traceable path. The proposal is not evidence that this architecture is best for every AI use case.
Rank #3
Explicit rules can make it easier to reconstruct how a decision was reached, test edge cases, and correct behavior by changing a rule rather than retraining a model. But domain experts must write and maintain those rules as policy changes. That can make the approach slower and more expensive than relying more heavily on a model-driven process. Buyers should weigh that maintenance burden against their need for traceability and control.
What the reported governance figures do—and don’t—show
French attributes several figures to Grant Thornton’s 2026 AI Impact Survey: 78% of senior leaders reportedly lacked strong confidence that they could pass an independent AI governance audit within 90 days; 46% reportedly named governance failures as a leading cause of AI underperformance. The article also reports that 7% of organizations still piloting AI were very confident of passing that audit, compared with 74% of organizations running AI in full production.
These figures are presented here as the article’s attribution to Grant Thornton, not as independently verified survey results. They give context for the governance concern but do not establish how often buyers ask particular questions or prove that any specific system is compliant.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsHow the EU AI Act deadline fits into a buying decision
A Grant Thornton UK legal briefing says standalone Annex III high-risk AI systems have until 2 December 2027 to comply under Regulation (EU) 2026/1744, which the briefing says entered into force on 27 July 2026. This is a secondary legal summary, not the text of the Official Journal. The date alone does not determine whether a particular organization or system falls within the Act’s scope. Buyers should establish applicability for their specific use before treating a deadline as their compliance timetable.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
