Don't Hand Over the Keys Yet: A Village Hall Bootcamp for AI Helpers

There is a particular kind of optimism that arrives when a community group discovers AI chatbots. Someone in the WhatsApp group says they have built a helper for the website, it answers questions about bin day and hall hire, and suddenly everyone is picturing fewer repetitive emails. That optimism is understandable. It is also exactly the moment to slow down.

An AI helper is not a replacement for the volunteer who knows that the side door sticks, the caretaker is on holiday, and the accessible toilet key lives in the blue tin behind the bar. It is a front desk, not the whole building. The useful question is not “can it answer?” but “can it hand someone over to the right person, service or room before it guesses?”

Start with the questions people actually ask

A stress test should sound like a normal Tuesday, not a product demo. Pull twenty real messages from your inbox, Facebook group or WhatsApp chat, then anonymise them. Mix the mundane with the sensitive. A decent starter set for a UK community group might include:

• “When is the next bin collection after the bank holiday?”
• “Can I book the small hall for a birthday party on 14 March, and is there a kitchen?”
• “Is the Thursday coffee morning step-free? My dad uses a walker.”
• “Who do I speak to about a safeguarding concern at the youth club?”
• “I’ve lost the link to the parish council meeting papers.”
• “Can my support worker come with me to the craft group?”
• “Is the car park free after 6pm?”
• “I need to cancel my table for the quiz night.”

Run each one through the helper and score the response. Did it answer directly, or did it invent a detail? Did it point to the right page on the council website, the booking secretary, the venue access page or the safeguarding lead? Did it ask a clarifying question when the query was ambiguous? Did it offer a phone number or an email address for a human?

Watch for confident nonsense

The most dangerous AI answer is not the one that says “I don’t know.” It is the one that says “Yes, the hall is available” with total confidence when the calendar has not been checked. Community groups run on trust, and trust is quickly dented by a cheerful bot sending someone to a locked door.

Look for these warning signs:

• It invents opening times, prices or room capacities.
• It answers safeguarding or medical questions without escalating.
• It gives a postcode or phone number that does not match your website.
• It treats a general council query as if your group runs the service.
• It promises a booking, refund or accessibility arrangement that no volunteer has agreed.
• It cannot say where its information came from.

A useful rule is simple: if the helper cannot link to a source, it should link to a person. That link can be a named role, an email address, a phone number or a short form. It does not need to be a dramatic handover. It just needs to be reliable.

Build a handover protocol before you need it

A handover protocol is less bureaucratic than it sounds. It is a short agreement about when the bot stops and a human starts. Write it down, keep it visible, and review it after the first month.

At minimum, define:

1. Escalation triggers. Safeguarding, complaints, accessibility requests, bookings, payments, data protection and anything involving a named person should go to a human.
2. Named roles, not just “contact us”. A generic inbox is where queries go to nap. Point to the booking secretary, the access lead, the safeguarding officer or the moderator rota.
3. Response expectations. If the bot says a volunteer will reply, say when. “Within two working days” is better than “shortly”.
4. A fallback path. If the human does not reply, where does the query go next? A phone number, a drop-in session or a noticeboard can stop someone slipping through the net.
5. A log. Keep a simple record of escalated queries and gaps. That log is how you improve both the bot and the group’s information.

This is where the village hall test helps. Ask the helper: “I’m coming to the film night on Friday. I use a wheelchair. Which entrance should I use and who will be there to meet me?” A good helper does not just say “yes, accessible”. It points to the accessible entrance, names the volunteer on the door, and gives a number to call if plans change. If it cannot do that, it should hand over to someone who can.

Keep the volunteer in the loop

AI helpers work best when they are treated like a new committee member: useful, supervised, and never left alone with the keys. That means a human reviews the answers, updates the source material, and can switch the helper off if it starts freelancing. It also means being honest with your community about what the bot is and is not. A short line on the website — “This helper answers common questions and passes anything tricky to a volunteer” — does more for trust than a polished avatar.

The point of a bootcamp is not to catch the bot failing. It is to find the edges of its usefulness before a real person does. Run the questions, watch the handovers, and keep a volunteer in the loop. Then the helper can do what good community tech should do: make the link to a human easier, not harder.