← All research

How we assess this

07 — AI, behaviorally not aspirationally

Q01

Question — Which AI tools, assistants, chatbots, document tools, coding tools, transcription tools, or automation features have people in the firm actually used for work in the last 90 days?

Uncovers — Actual AI usage, tool diversity, bottom-up adoption, recency.

Why it stays clean — It asks for recent behavior and named tools, not beliefs about AI.

Follow-up probes

Polite-answer trap — Useless/evasive: “People are experimenting with AI.” Redirect: “Which named person or role used which named tool last, and for what task?”

Q02

Question — What has the firm paid for around AI or automation so far, including subscriptions, pilots, training, consultants, integrations, or internal time?

Uncovers — Committed spend, seriousness, budget owner, AI investment history.

Why it stays clean — It asks for money and time already spent rather than AI importance.

Follow-up probes

Polite-answer trap — Useless/evasive: “We have not spent much yet.” Redirect: “Has any invoice, licence, training fee, consultant day, or internal project time been attached to AI or automation?”

Q03

Question — Tell me about the most concrete AI or automation trial the firm has run. What data went in, what output came out, and what happened to the output?

Uncovers — Pilot realism, data access, output usefulness, adoption depth.

Why it stays clean — It demands the mechanics of a real trial, not a forecast.

Follow-up probes

Polite-answer trap — Useless/evasive: “We played around with a few things.” Redirect: “Which trial had the clearest input, output, and owner?”

Q04

Question — What was the last AI-generated output someone trusted enough to use in firm work?

Uncovers — Trust threshold, output acceptance, practical use cases, risk tolerance.

Why it stays clean — It asks for a specific accepted output rather than perceived promise.

Follow-up probes

Polite-answer trap — Useless/evasive: “AI can be useful for drafts.” Redirect: “Which draft or output was actually used, and who received it?”

Q05

Question — What was the last AI-generated output someone rejected, corrected, or stopped using?

Uncovers — Failure modes, trust barriers, quality bar, risk perception.

Why it stays clean — It asks for a concrete rejection event.

Follow-up probes

Polite-answer trap — Useless/evasive: “You have to check the output.” Redirect: “What is the clearest example where checking revealed a real problem?”

Q06

Question — Which AI or automation usage is company-approved, and which usage happens informally through personal accounts or unapproved tools?

Uncovers — Governance gap, shadow AI, security risk, practical demand.

Why it stays clean — It asks about observed approval status and behavior.

Follow-up probes

Polite-answer trap — Useless/evasive: “People know to be careful.” Redirect: “What is the last example you heard of someone using a personal account for work?”

Q07

Question — Who inside the firm has built or configured a script, macro, chatbot, AI workflow, spreadsheet automation, or low-code tool for their own work?

Uncovers — Internal builder capacity, forward-leaning users, latent automation demand.

Why it stays clean — It asks for actual builders and artifacts, not interest in innovation.

Follow-up probes

Polite-answer trap — Useless/evasive: “We have a few tech-savvy people.” Redirect: “Which person has built something that others rely on?”

Q08

Question — What AI training, policy session, webinar, vendor workshop, or internal presentation has the firm attended or paid for, and what changed afterwards?

Uncovers — Learning spend, action after education, seriousness, internal momentum.

Why it stays clean — It asks for observable follow-through after an event.

Follow-up probes

Polite-answer trap — Useless/evasive: “We keep up to date with AI.” Redirect: “Which event led to the most concrete next action?”

Q09

Question — Which client, project, contract, data, or confidentiality concern has limited AI use in practice?

Uncovers — Data constraints, legal blockers, client sensitivity, adoption ceiling.

Why it stays clean — It asks for actual limits encountered rather than abstract risk views.

Follow-up probes

Polite-answer trap — Useless/evasive: “We are cautious with data.” Redirect: “Which specific dataset or document was not allowed into an AI tool?”

Q10

Question — Which AI or automation effort has been abandoned, paused, or left unused, and what happened?

Uncovers — AI churn, bandwidth limits, quality failures, implementation barriers.

Why it stays clean — It asks for a real stopped effort.

Follow-up probes

Polite-answer trap — Useless/evasive: “Some experiments naturally fade.” Redirect: “Which experiment faded most clearly, and what was the last visible activity on it?”