AI Skills for Employees: Separate Forecasts From Scenarios

AI skills for employees should include a simple analytical habit: label a number before anyone acts on it. A polished AI answer may present 12,500 units as if it were a forecast. In reality, the number could be a scenario under optimistic assumptions, a sales target, or an unsupported guess. The arithmetic is not the only risk. The label changes what the decision-maker is entitled to believe.

Teach employees to separate four things: a forecast estimates an outcome under stated data and assumptions; a scenario explores what could happen under one coherent set of conditions; a target states what the organisation wants to achieve; a guess has no traceable method or evidence. Different professions use these terms differently, so the training must also require the local method and owner—not just the label.

Why AI skills for employees need number labels

The UK Department for Transport’s forecasting guidance gives a precise domain example: it defines a forecast as one model run for a year under a set of assumptions, and a scenario as a set of forecasts under one set of assumptions. The same document uses core and alternative scenarios to make uncertainty visible. These definitions are transport-specific, but the transferable lesson is clear: assumptions and purpose travel with the number.

NIST’s AI Risk Management Framework Playbook similarly recommends measuring performance against ground truth in the intended context, documenting differences between the test setting and deployment, and checking whether data still represent the operating environment. UK guidance on evaluating AI interventions also asks teams to define the baseline and document business-as-usual before comparing outcomes. A number without its context cannot carry that evidence.

This is an AI-literacy issue, not only an analyst’s issue. The European Commission’s current guidance says AI-literacy action should reflect staff knowledge, the system, its risks and the context of use. An employee who uses generative AI to draft a budget note, demand plan or project brief needs enough literacy to recognise when a fluent number lacks the evidence required for its label.

Use SCOPE before a number enters a decision

SCOPE does not make every employee a forecaster. It gives non-specialists a stopping rule. If the method, assumptions or owner are missing, the person should not strengthen the wording or add decimal places. They should ask for the missing evidence or route the number to the person responsible for the analysis.

Worked example: the 12,500-unit demand figure

A category manager asks an AI tool to estimate next quarter’s demand. The prompt includes last year’s quarterly sales—9,600, 10,100, 9,900 and 10,300 units—and mentions a planned promotion. The answer says: “Expected demand is 12,500 units.” It does not show a model, promotion uplift, stock constraints, forecast horizon detail or error range.

The AI tool can still help. It can organise assumptions, calculate transparent alternatives from supplied figures, draft a sensitivity table and list missing inputs. It should not be allowed to turn a target into a forecast or a scenario into a probability merely because the prose sounds decisive.

Four common training failures

First, exercises reward the “right” number without checking how it was produced. Second, scenarios are presented as high, medium and low but the assumptions do not differ. Third, the target appears in the same table as forecasts without a different label. Fourth, the learner asks the same AI system to generate the number and certify that it is reliable. Each failure hides the decision trail.

Use fictional or approved data in practice. Score whether the learner labels the output, exposes the method, names assumptions, communicates uncertainty and routes approval correctly. Avoid scoring closeness to a future outcome that nobody yet knows. The training evidence is the quality of the analytical handoff.

Teach people what the number permits

The goal is not to ban AI from planning. It is to stop presentation quality from outrunning analytical meaning. When employees can distinguish a forecast from a scenario, a target and a guess, they know whether a number supports planning, stress-testing, coordination or further investigation.

Connect this skill with Bokili’s guides on independent evidence, assumption logs, error-cost thresholds and verification before use. Bokili’s HR and L&D approach turns distinctions like these into short, role-relevant practice with visible feedback.