The Tool Suggested It. Who Checked It?

Workplace Growth Daily · 2026-09-23

LUMARA Workplace Growth Daily · 23 September 2026

Illustrative scene: A customer-support colleague opens a draft reply prepared with a new writing tool. It sounds wonderfully confident. It also promises a delivery date nobody has confirmed. The cursor hovers over “Send”; the team’s unofficial emergency alarm is one raised eyebrow.

The useful question is not whether the tool can write a smooth sentence. It is whether the team knows which claims need checking, who can approve the reply and what to do when the answer is uncertain. That applies to AI, a template or any other new technology. This is a research-backed practical briefing, not breaking news: none of the supplied research articles was published in the past seven days. The findings below are sourced; the exercises and working rules are LUMARA’s practical suggestions, not findings from those studies.

What is changing

Trends: evidence for a careful trial

21 July 2026 — Guidance matters alongside access. In Gallup’s U.S. employee research, 48% of employees who said their manager actively supported their team’s AI use were engaged, compared with 30% of those who did not report that support. This is an association, not evidence that a manager can produce an 18-point rise by introducing a tool. For a team: explain the permitted use, the standard for a finished answer and where human judgement is required. Gallup, 21 July 2026
29 April 2026 — Experimentation can outrun preparation. CIPD reported that around 45% of organisations surveyed described their AI use as ad hoc or experimental. Around 42% of those experimenting fell into its lowest organisational-readiness group. For a team: run a bounded trial with checks and a way to stop, rather than quietly turning individual experiments into standard practice. CIPD, 29 April 2026
24 June 2026 — Existing skills are easy to overlook. CIPD says 45% of UK employees feel their skills are under-used in their current roles. That UK finding should not be treated as a global rate. For a team: ask who already knows the workflow, the customer context and the common exceptions before assigning a technology trial. CIPD, 24 June 2026

These findings justify better questions, not a promise of faster work, higher engagement or growth.

Build a skill in 10 minutes

Skill: claim tracing

The 10-minute exercise: Take one draft reply that has not been sent and remove customer-identifying details. Use an existing approved example if a live case is unsuitable. Spend two minutes underlining statements a customer might rely on: dates, prices, eligibility, next steps or promises. Spend three minutes writing the source beside each statement, such as an approved policy or the current order record. Spend three minutes marking each one confirmed, unclear or unsupported. Use the final two minutes to revise the draft: keep confirmed claims, ask about unclear ones and remove unsupported ones.

Visible success criteria: Every consequential claim has a source or a clear question awaiting an answer; no invented certainty remains; one named person reviews the reply before it goes out. If information is missing, a good result may be a short holding response, not a polished guess.

If an approved AI tool produced the first draft, treat its wording as a proposal to inspect, never as a source. The same test works on a saved template or an automatically filled form. It builds AI literacy through a transferable habit: knowing what a system has said is different from knowing what the team can verify.

Staff and management: respect in both directions

Relationships: make caution safe and fair

Managers: Set the same review standard for everyone, whether they use AI, another tool or no tool at all. Explain what may be entered into a system and what must stay out; check the organisation’s rules before a trial. Listen when staff describe awkward exceptions or extra checking work. Recognise the person who catches a misleading claim, not only the person who finishes first. Give people time to learn, ask for help and meet a manageable workload; decide what other task will wait if a new check takes time. Do not turn reasonable objections into a performance fault.

Manager, say this: “Please check delivery claims against the current record before sending. I’ll review the first few with you. What part of this process adds work, and what should we pause while we test it?”

Staff: Raise concerns early and concretely. Point to the claim, explain the possible effect on the customer and offer a workable next step. You can disagree with a rollout or ask for a private conversation without having to prove the tool is always wrong. Persistent bullying or retaliation is not a normal difference of opinion: use the organisation’s reporting or support route rather than tolerating it.

Staff member, say this: “I’m concerned this draft gives a date our record doesn’t confirm. Can we send a holding reply while we check it? I’d also like to discuss how many drafts I can review safely today.”

Work better as a team

Teamwork: a 15-minute “claim-and-owner” table

Gather a small team around one anonymised, illustrative draft response containing a delivery-date claim and a suggested next step. Use a shared document or paper; no new software is needed. Appoint a facilitator to keep time, a customer voice to ask what a recipient would understand, a checker to identify approved evidence, and a decision owner who can approve or hold the final message. If the team is larger, pair people in those roles.

Minutes 0–3: Read the draft silently. Each person highlights one claim that could change a customer’s decision. Do not debate whether the tool is “good”; stay with this reply.

Minutes 3–7: The checker locates the authorised source for each highlighted claim. The customer voice flags wording that sounds more certain than the evidence. Put “unknown” where a source cannot be found; nobody wins points for filling the gap with confidence.

Minutes 7–11: Agree which claims to keep, change or remove. The decision owner names who will resolve unknowns and who will approve the reply. The facilitator asks whether a manual draft would need exactly the same checks.

Minutes 11–15: Write a one-page claim-and-owner table with four columns: claim, evidence or unknown, safe wording, and reviewer. Finish with a clear hold-or-send decision. That shared output can guide the next trial without pretending one exercise has settled every case.

Turn teamwork into measurable growth

Growth: test the behaviour, not the hype

LUMARA’s suggested low-cost experiment is to try that claim-and-owner table on one narrow type of routine customer reply for seven days. The support team lead owns the trial and agrees in advance which replies qualify. A reviewer remains accountable for every outgoing message; staff may draft manually or use an approved tool where policy permits. Do not paste sensitive customer information into an unapproved system, and stop the trial if the approved route is unclear.

Before starting, the lead records a baseline from comparable recent replies: how many required a correction after sending, and how long the team spent checking each one. Record the same measures during the trial. Also note the balancing measure: review time and unfinished work per person, so a possible quality improvement does not hide an unmanageable workload. Where feasible, ask a colleague whether the final wording clearly tells the customer what happens next.

At the seven-day review, compare like with like and ask staff what the table missed. Keep, revise or stop the practice based on the evidence and workload, not on how impressive the draft sounded. A small change may help quality or customer clarity; it is not a guaranteed productivity or revenue gain.