AI Literacy Workshop for SMEs: Teach People Not to Invent Missing Data

For many SMEs, the real question is not whether to use AI, but how to use it without turning a helpful draft into an operational mistake. A short AI literacy workshop is most useful when it trains people to recognize the limits of an output: when data is missing, when human review is mandatory, and when the answer must be rewritten from scratch.

The NIST AI RMF and Playbook are voluntary resources for AI risk management. This workshop and scoring rubric are Dedaloop teaching choices, not a certification. (NIST)

Why this matters for SMEs

For benefit corporations, cooperatives, and SMEs, the main challenge is often not access to AI tools, but the quality of daily use. A 90-minute workshop is valuable if it leaves people with one simple habit: do not fill gaps with assumptions.

The practical value is threefold:

• it shows that AI can speed up drafting without replacing human control;
• it makes missing information visible;
• it gives teams a simple rubric for deciding whether an output is usable or needs rework.

This proposed workshop uses fictional data; it does not describe a delivered customer program.

Workshop flow: 90 minutes

1) Objectives and rules — 10 minutes

Start with three rules:

• AI may help write a draft, but it must not invent missing data;
• final control remains human;
• if a key piece of information is missing, the response should flag it and ask for it.

This is not about measuring “how intelligent” the AI is. It is about governing its use responsibly, in line with NIST’s emphasis on continuous risk management across the AI lifecycle. NIST AI RMF Core

2) Exercise on a fictional support ticket — 20 minutes

Use a short, intentionally incomplete case.

Simulated case data

• Request: “The customer says the form does not send the confirmation.”
• Declared urgency: “Blocking for today.”
• Missing data: error code or a screenshot of the message.

Working prompt

Act as a support assistant. Write a short, professional draft reply to the request below. Use only the information provided. Do not invent missing data. If an essential detail is missing, include one clear question to obtain it. If useful, mention possible next steps, but do not attribute unverified causes.

Request: the customer says the form does not send the confirmation; treat the situation as blocking for today; the error code or a screenshot of the message is missing.

Expected example output

Thank you for reporting this. To understand where the flow is stopping, we need the error code or a screenshot of the message shown. Please also indicate when the problem occurs. Redact personal or confidential details in attachments: this information will help the owner assess the next checks without assuming a diagnosis.

This is a simulated teaching example, not the result of a real model test.

3) Error and missing-information review — 20 minutes

Compare the AI draft against the original ticket and mark:

• any detail added that was not present in the source;
• any assumption made without evidence;
• any point where the text correctly avoided inventing information.

This is often the most useful part for SMEs, because it reduces the risk of a polished draft entering an operational process with hidden errors.

4) Pair review — 20 minutes

One person checks faithfulness; the other checks clarity. Both verify whether the output really asks for the missing data instead of filling the gap.

5) Individual check — 15 minutes

Each participant receives a new short case and applies the rubric independently.

6) Next action — 5 minutes

Close by deciding where the method can be used in real workflows: email, tickets, internal notes, sales requests, support, or operational documentation. The goal is not to multiply exercises, but to choose one habit to bring into the team within a week.

A simple rubric: 3 criteria, 0–2 points each

Use a small evaluation grid:

• Faithfulness: does the text use only information present in the ticket?
• Recognition of gaps: does it clearly flag the missing detail instead of inventing it?
• Clarity: is the response understandable, brief, and useful?

Proposed threshold: 4/6, plus one hard stop rule: no critical invention. This threshold is a teaching choice for the workshop. It is not a certification and not a NIST standard.

A no-account alternative with readable materials

To lower the barrier to entry, the workshop can also be offered without registration, with printable or downloadable plain-text materials: the case sheet, the prompt, the evaluation grid, and a notes area. For a small organization with limited time, this makes internal reuse easier. That is an editorial recommendation, not a regulatory requirement.

What to check before publishing the workshop

Before considering the workshop ready, verify that:

• the simulated case really contains one missing piece of information;
• the prompt explicitly forbids inventing data;
• the expected output shows a clear question for the missing detail;
• the rubric is understandable to non-technical participants;
• the text states clearly that the method is educational, not certifying.

What an SME can do next

If the goal is to improve the quality of emails, tickets, or reply drafts, this workshop gives people a practical way to use AI with caution and consistency. The point is not to produce longer answers, but more reliable ones.

If you want to adapt the workshop to a real internal process, describe the process you want to improve or let’s evaluate a training path together.