What training do staff need to use AI well? Five things everyone needs, taught on their own tasks, in short sessions.
Short answer. Staff need short, hands-on training on five things: what the office AI is for and not for, what data may go into it, how to ask for what they want, how to check the answer, and how to report a problem. Teach it on real tasks, not generic demos, and add deeper sessions for roles that depend on it. Evidence from two studies suggests gains are uneven and self-assessment can be unreliable. A study of 5,179 customer support agents found a 14% average productivity gain, 34% for novices and little for experts. A randomised trial found experienced developers 19% slower while they believed they were faster.
01
Five things everyone needs
What it is for. The tasks the office has approved, and the tasks it is not for, such as decisions about people or advice nobody will review.
What may go in. The data classes allowed, and what must never be pasted. OWASP advises educating users to avoid entering sensitive information into language models.
How to ask. Give context, the goal and the format; ask for sources; break big jobs into steps; keep the prompts that work.
How to check. Read the cited source passage; check names, numbers and dates against the original; treat confident wording as no evidence of accuracy.
How to report. Errors, near misses and odd behaviour go to a named person, without blame.
02
A three-session plan
Session
Length
What happens
1. Rules and basics
About 60 minutes
The policy in plain words, what the tool does, a live demonstration on a real task, do’s and don’ts
2. Hands-on in teams
About 90 minutes
Each team tries its own three tasks, compares outputs with the source and writes down prompts that work
3. Checking and mistakes
About 60 minutes
Spot errors in sample outputs, trace each to its source, practise reporting
03
Different roles, different depth
The GOV.UK AI Playbook, written for civil servants, uses the same idea. It separates beginners, technical staff outside digital teams, data professionals, digital professionals and leaders, and says the learning should differ for each group.
Accounts and roles, logs, updates, backups and incident steps
Partners or directors
Risk, client disclosure, accountability and the owner’s decisions
New joiners
A short induction in the first week and acknowledgement of the policy
04
Why checking is the skill
Brynjolfsson, Li and Raymond studied 5,179 customer support agents given a generative AI assistant. Productivity, measured as issues resolved per hour, rose 14% on average, 34% for novice and low-skilled workers, with minimal impact on experienced and highly skilled workers.
METR ran a randomised trial with 16 experienced developers on 246 tasks. Before starting they forecast that AI would cut completion time by 24%, and afterwards they estimated 20%; measured completion time rose by 19%. That study concerned software developers and early-2025 tools, not office staff, but it shows why training should teach measuring and checking, not only prompting.
05
Checking that it worked
A short quiz on the policy, repeated for new joiners.
A sample review of real outputs for source checking.
The number and quality of error reports.
Usage from the logs, by team.
A question to staff: what did you stop doing, and why?
Repeat the training when the tool or model changes. NIST’s guidance on learning programmes asks for role-based content and for metrics and evaluation to improve it.
06
What ANVE says
ANVE is designed so that the checking step has something to check against. Every answer is cited, every action item in meeting minutes quotes the transcript line it came from, and an owner must be confirmed by a person before an action is approved.
The demo call is free. What training, if any, comes with a paid pilot is not stated here; ask for it to be written into the pilot agreement along with the fee.
07
Questions people ask
How long does it take staff to become useful with AI?
It varies by person and task. The customer-support study found novices gained most. Measure on your own tasks rather than relying on a general figure.
Do we need a prompt-writing course?
A short one helps. Checking answers and knowing what data may go in matter more.
Should partners and senior staff be trained too?
Yes. They approve the work that goes out and set the tone. NIST’s AI framework asks for executive leadership to take responsibility for decisions about AI risk and for personnel to receive training for their duties.
08
What is not real yet
No sealed ANVE unit exists; every image of the appliance on this site is a labelled render. The software runs today on a development machine. No customers yet, no certifications, no delivery date, and the final price is not announced.
Reservations are ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing (terms). Offices that cannot use cloud AI can apply to be one of ten design partners.
