Outlook
Part of What the evidence can and cannot show about community management in 2027
Five training exercises for English community team skills
Five practical exercises to train community team skills in England: paired cases, incomplete files, data views, AI summaries and incident drills.
Skills grow fastest when training copies the work. An English community team can build the capabilities it needs from five short exercises, each using a realistic case, scored against a stated rule and repeated with a fresh case whenever the service changes.
What to take away
- Five exercises cover contextual judgement, evidence handling, data interpretation, controlled AI use and incident coordination.
- Good judgement shows when an operator changes a decision after new information without rewriting the record to hide the change.
- Data literacy means refusing an attractive conclusion when the measure does not support it.
- Practical AI literacy means defining a bounded use, protecting information, checking output and stopping when needed.
- Map critical tasks to named owners and deputies, then reassess with a fresh case after any serious change.
The five exercises at a glance
Each exercise tests one skill and asks the trainee to produce something that a colleague can mark. The product matters more than the confidence of the answer, because a marker can check a product and cannot check a feeling.
| Exercise | Skill it tests | What the trainee produces |
|---|---|---|
| Paired cases | Contextual judgement | The rule, evidence, uncertainty and proportionate next step |
| Incomplete case file | Evidence handling | What is known, what is missing, what to preserve |
| One week, three views | Data interpretation | The decision each view supports and what cannot be inferred |
| Machine-written summary | Controlled AI use | The faults found, the data the tool received, who approved it |
| Timed incident drill | Incident coordination | Preserved evidence, restricted access, known facts, the right owner |
Paired cases for judgement
Paired cases train judgement because one small detail changes the right outcome. Operators must distinguish disagreement from abuse, urgency from volume and a policy breach from conduct that merely feels uncomfortable. Ask the trainee to state the rule, evidence, uncertainty and proportionate next step.
Judgement: one detail changes outcome
Disagreement or abuse?
abuse: restrict and record
disagreement: no action
Add an appeal to the set. Good judgement is visible when a person can change a decision after new information without rewriting the record to hide the change. Mark the trainee on the record they leave behind as well as on the outcome they chose.
An incomplete case file for evidence
An incomplete file shows whether an operator can work from what is missing. The trainee should identify what is known, what is missing and what must be preserved before taking action. Moderation records need dates, source material, policy versions, actions and outcomes that another authorised person can reconstruct.
Reconstructable moderation record
- Dates of report and action
- Source material preserved
- Policy version cited
- Action taken stated
- Outcome recorded
- Another authorised person can reconstruct
Where the Online Safety Act applies, Ofcom's guidance for services likely to be accessed by children covers risk assessment, protections, record-keeping and review. Scope belongs to the service provider and its advisers, not to a community manager guessing from the brand's audience.
One week, three views for data
Show the same week through counts, rates and cohort outcomes, then ask which decision each view supports and what it cannot show. The trainee should define a denominator, recognise a changed reporting window and separate a rise in reports from a rise in confirmed harm.
Spreadsheet confidence helps, but the decisive skill is refusing an attractive conclusion when the measure does not support it. A trainee who writes "this view cannot tell us" has passed the exercise as surely as one who finds the trend.
A machine-written summary for AI control
Give the trainee a machine-generated case summary with a wrong speaker, a lost qualification and an invented resolution. They mark each fault, then say what data the tool received and who approved that use. Practical AI literacy means a bounded use, protected information, checked output and knowing when to stop.
The ICO's AI transparency page connects personal-data processing with clear information about purpose, retention and sharing, while noting that its guidance is under review. Treat it as a prompt for the question about approval, not as a substitute for the organisation's own data-protection advice.
A timed incident drill
A timed drill shows whether a team can coordinate under pressure. Pick an account compromise, a disclosure problem or a surge in harmful posts, and run it with security, communications and subject experts. Score handoffs and decisions, not theatrical confidence.
During the drill the operator needs to preserve evidence, restrict access, communicate known facts and bring in the right owner. Afterwards, ask each participant which decision was theirs and which belonged to someone else.
Supplier knowledge also matters. The NCSC's secure AI development guidelines include supply chain security and documentation in their development stage. A non-technical lead should be able to ask who provides each component, what changed and how the service falls back.
Keep the training map current
Map critical tasks to named owners and deputies. Observe real or simulated work, record the gap, practise to match it and reassess with a fresh case. Operators need fluency to spot a boundary and escalate with usable evidence, not a law or data-science degree.
Review the map after a serious case, policy revision, new tool, supplier change or altered member group. That cadence keeps capability responsive to the service rather than to annual training catalogues, and a ninety-day plan can hold the first round.
Include language and access needs in the exercise set. A person who performs well on short standard-English posts may need different support for translated material, voice notes or a distressed member using an unfamiliar format.
Before you act
- Test judgement with paired cases where one detail changes the outcome.
- Hand out an incomplete case file and mark what the trainee preserves.
- Show one week in three views and ask what each cannot support.
- Review a machine-written summary for faults and for data sent to the tool.
- Run a timed incident drill and score handoffs, not confidence.
- Review the training map after serious cases, policy revisions or supplier changes.
Common questions
How should a team mark a paired-case exercise?
Check that the trainee states the rule, evidence, uncertainty and proportionate next step for each case. Include an appeal, and credit a changed decision when the record shows what changed and why.
What should an operator do with an incomplete case file?
Identify what is known, what is missing and what must be preserved before taking action. The record should let another authorised person reconstruct dates, source material, policy versions, actions and outcomes.
What belongs in an AI summary review?
The faults found in the summary, such as a wrong speaker, a lost qualification or an invented resolution. Add what data the tool received and who approved that use.
