The draft for a Tuesday seminar still carries a wedding clause pasted from a Saturday file.
In event venue management, the work that changes the week is narrower than a catalogue. The request gets read, the fields that are actually there get filled and the rest stays blank, while a person still sets the rate, picks the wine and decides whether Saturday is even free. Two jobs matter: putting the overnight pile in an order someone can work, and getting a quote ready to review.
Most briefs we see go to more than one room, so the first reply has to be fast. Speed comes from not retyping the brief. Not from a score out of 100.
What the request actually contains
A request arrives messy, by email, web form or marketplace message, with the date and headcount sometimes there and the budget often missing. Format buried in a sentence about 'something more standing than seated'. The work is to pull those fields into one place so nobody copies them into a spreadsheet by hand.
Leave gaps as gaps. If the sender did not name a budget, the system should not invent one from last June. If the month is March and the day is not, the date field stays empty. A blank field does no harm, whereas a guessed one turns into a wrong quote.
Dietary notes and access needs sometimes sit in the same paragraph as the headcount. Those are personal data. They belong on the function sheet later, not in a prompt pasted into a public chatbot.
Sort the pile before anyone writes
In most inboxes, a large product launch and a birthday that will never book carry the same New label. Opening them in the order they arrived wastes the first hour on the note that will never sign.
A model can group by type, flag Saturday against Tuesday, and put a complete brief on top. That is sorting, nothing more. A low signature rate usually means the list is late or the qualification is sloppy.
If the system starts refusing dates on its own, the legal position changes. The ICO's guidance on automated decision-making, including profiling, treats a decision as ADM when it is based solely on automated processing and has a legal or similarly significant effect. A human who reads the ordered list and decides who to call first stays on the safe side of that line. A silent refusal sent before anyone is in does not.
- Complete brief (date, headcount, format) goes to the top.
- Incomplete brief gets a short ask for the missing field, not a full package.
- Saturday wedding and Tuesday seminar do not share a reply template.
- A regular client is marked as such before anyone writes 'Dear Sir or Madam'.
A draft quote is still a draft
The useful output is a pre-filled quote: room, format, last similar event's food and drink, the seasonal rate if that rate already exists in the grid. The salesperson opens it, changes the wine, adds the late finish, deletes the package that does not exist this year.
Do not send the first draft. Models invent rooms. They put last year's minimum spend on a client who already negotiated it out. Joinways can leave that draft on the file, but someone still rereads it, and sending stays a human action.
| Ready in the draft | Still a human decision |
|---|---|
| Room name and listed capacity | Whether that capacity still holds after a stage or a buffet |
| Format guessed from the wording | The actual layout the fire plan allows |
| Last similar package and its lines | This year's price, the wine, the late finish |
| A first cover letter in the house tone | The sentence that names their product, not 'your event' |
Replies that stay in draft
Capacity, whether Saturday is free, a price range already on the site: a draft saves the first ten minutes. High-stakes briefs stay with a person, because buyers notice a letter that never uses their product name, and the send button stays off the model.
Personal data and the ICO
A request is personal data: name, email, company, sometimes a diet or an access need. The ICO is plain: whenever you process personal data, whether to train a system or to make a prediction with one you already have, you need an appropriate lawful basis. Development and deployment are different purposes. You choose the basis before you start, you write it down, and you put it in the privacy notice.
Transparency is a separate duty. In its guidance on transparency in AI, the ICO asks you to tell people the purposes, the retention period, and who you share the data with. If you collect the request directly, that notice goes out at the time of collection, before you train on it or run a model on that person.
Do not paste the whole thread into a public chatbot. The CNIL's first answers on deploying generative AI tell organisations to start from a concrete need, to list allowed and forbidden uses (for example not feeding personal data into the system, or not handing it a decision), and to check whether the vendor may reuse what you type. A venue tool that extracts fields inside your own account is a narrower treatment than a paste into a consumer chat window.
What we do not put on the machine
The European Commission's page on the AI Act is useful for seeing where venue software falls. High-risk cases on that page include exam scoring, CV-sorting for recruitment, credit scoring, biometrics. Sorting venue inquiries and drafting a quote is not on that list. The vast majority of AI systems currently used in the EU, the Commission says, fall into minimal or no risk. GDPR still applies. If a chatbot answers the organiser without a person in the thread, the Act's transparency layer asks that the human know they are talking to a machine.
The visit and the negotiation on a long brief stay with people, not the model. Keep the request list clean enough that a person can see what arrived, what is waiting, and what already has a draft. That is enough to start with; forecasting can wait, and most rooms never need it.
Questions venues actually ask
- Does the system send the quote on its own?
- No. A draft that already has the room and last year's package still needs a person to check the date, the rate and the clauses. Models invent rooms and copy wedding language onto a weekday seminar. Keeping a person on the send button is also the safer side of the ICO line on solely automated decisions with a significant effect.
- Where should a venue start with AI?
- Sort the overnight pile, then open a draft quote. Those two jobs remove the retyping. Do not begin with seasonal yield or contact enrichment scraped from the web. A quick first reply matters more than a forecast nobody rereads.
- If we order requests with a model, is that an automated decision?
- Not if a person still chooses who to call and what to offer. The ICO treats ADM as a decision based solely on automated processing that has a legal or similarly significant effect. Ordering a list is not a refusal, whereas a silent rejection sent without review could be. Keep a human on the outcome.
