Key takeaways
- The tasks AI handles well in a venue are repetitive information jobs: answering the same questions, capturing requests and writing summaries.
- Hospitality itself, judgement calls, allergen decisions, complaints and licensing duties stay with people.
- Run a two-week interruption log before buying anything, so you know which hours you are trying to win back.
- Few businesses using AI report cutting staff; the realistic gain is manager attention back on the floor.
Where a manager's week actually goes
Ask a general manager what they did yesterday and the answer is rarely 'managed'. It is a list of small jobs that arrive at the worst moment: the phone during the lunch rush, a supplier chasing a delivery slot, three emails about a Christmas party, a rota gap, a review that needs a reply. Each takes two minutes. Together they fill the gaps where coaching, cost control and guest contact should happen.
The national data points the same way. Among UK businesses with 10 or more employees that use AI, 38% said administrative and clerical roles were the most affected[1]. In accommodation and food, the most common reason for using AI was to improve business operations, cited by 39.2% of those using or unsure about it[2]. The practical opportunity is in the paperwork and the phone, not in the kitchen or on the floor.
Pay data shows why that time is worth protecting. Median hourly pay for restaurant and catering managers and proprietors was £15.75 in April 2026, and £18.51 for publicans and managers of licensed premises[3]. An hour of a manager answering 'what time do you close on Sunday?' is an expensive hour, and it is usually taken from the floor at peak.
Hand off, share or keep
Sort the recurring jobs into three groups. The test is simple: does the job need judgement about a person, or does it need the same information delivered accurately every time?
| Task | Decision | Why |
|---|---|---|
| Answering hours, parking, dogs, menus and directions | Hand off | Same facts every time; accuracy depends on a maintained facts list |
| Taking table booking requests by phone | Hand off | Capture is routine; confirmation follows your rules |
| Capturing private dining and party enquiries | Hand off the capture, keep the proposal | Details are standard; the menu and price conversation is a sale |
| Call summaries and a daily enquiry digest | Hand off | Removes note-taking and missed messages |
| Drafting review replies and supplier emails | Share | AI drafts, a manager edits and sends |
| Rota drafts from availability | Share | Useful first draft; fairness and hours need a person |
| Allergen questions | Keep | Safety decision based on today's kitchen |
| Complaints, refunds and difficult guests | Keep | Judgement and goodwill |
| Refusing service, licensing and safeguarding | Keep | Legal duties sit with licence holders and trained staff |
| Coaching, hiring and discipline | Keep | People decisions with legal weight |
What 'hand off' looks like in practice
An AI employee is software that answers calls and messages, handles routine requests within rules you set, and passes everything else to the right person with a written summary. The difference from a basic AI receptionist is that it carries the job to a usable outcome, such as a complete booking request in the team's queue, rather than only taking a message.
- Phone calls during service: answered on the first ring, routine questions handled, booking requests captured with party size, date, time and contact details.
- Party and private dining enquiries: date, numbers, occasion and menu interest captured and passed to the events lead, who replies with a proposal.
- Escalation: allergen questions, complaints and anything off-script transferred or passed to the manager on duty, with a backup if they cannot answer.
- Records: every call logged with a summary and transcript, so the morning handover is a two-minute read rather than a scrawled note by the till.
Ava works this way today: it answers everyday questions from information you approve, takes booking requests for your team to confirm, captures private dining enquiries, passes allergen and complaint calls to the manager you choose and logs every call. Direct booking into your reservation system is a planned feature, not something to rely on yet.
What stays human, and why that is the point
Government research on AI adoption found that face-to-face roles in sectors such as hospitality were widely seen as resilient to AI[4]. That matches what operators see: guests come for the welcome, the recommendation and the recovery when something goes wrong. The goal of handing off phone and admin work is to put the manager back where those things happen.
Some decisions must stay with people for legal reasons too. Licensing duties, such as refusing service to someone who is drunk, sit with the licence holder and staff. Allergen decisions depend on today's ingredients and kitchen set-up. Employment decisions carry rights for staff. None of these should be delegated to software, however good its summaries.
6%
of UK businesses with 10+ employees using AI reported reduced headcount as a result (September 2026)[5]
Run a two-week interruption log first
Before choosing a tool, measure the problem. A simple log tells you which tasks to hand off and gives you a baseline to judge the result.
Make a tally sheet
Columns for time, channel (phone, email, social, walk-in), topic, minutes spent and whether it pulled someone off the floor.
Log every interruption for two weeks
Include a weekend and a quiet weekday. Ask whoever answers the phone to tally too, not only the manager.
Group by topic
Hours and directions, bookings, parties, suppliers, staff, complaints, other.
Mark the repeatable ones
Anything where the answer is the same each time, or where the job is to collect the same five details, is a candidate to hand off.
Price the time
Multiply minutes by the hourly cost of the person who handled them, and note which ones happened during peak service.
Set a target
For example: no manager time on hours and directions calls, every booking request logged within five minutes, every party enquiry answered within one working day.
Example: a manager's log before and after
Example. The general manager of a fictional 110-cover pub in Leeds logged 212 interruptions over two weeks. Phone calls about hours, menus and directions made up 71 of them; booking requests 64; party enquiries 18; supplier calls 22; staff and rota issues 25; complaints 12. Roughly half arrived between 12pm and 2pm or 6pm and 9pm.
Example, continued. After routing calls through an AI employee with clear handover rules, the next two-week log showed 19 manager interruptions for bookings and information, mostly allergen questions and large groups that were rightly passed on. Party enquiries arrived as written summaries. The manager used the recovered time for two extra floor walks per service and a weekly stock count that had been slipping.
The numbers are illustrative, but the method is the useful part: count, hand off the repeatable work, count again.
Worries managers raise, and honest answers
Managers who have run a busy service are rightly sceptical. These are the objections that come up most, with the answers that hold up in practice.
- 'Regulars will hate it.' Some will, at first. Telling callers up front that they are speaking to an assistant, and putting them through to a person when they ask, takes most of the sting out. Regulars who call to chat with the landlord should still be able to.
- 'It will get things wrong.' It will repeat whatever is in the facts list, so the list has to be right. Wrong answers usually trace back to a bank holiday change nobody updated. Give one person ownership of the list.
- 'It will promise tables we haven't got.' Configure it to take requests, not confirmations, unless it is connected to live availability. The acknowledgement wording matters: 'request received', never 'see you Saturday'.
- 'Staff will stop answering the phone.' Decide when people pick up first, for example outside service, and when the assistant does. Write it on the rota.
- 'It is one more system to check.' Only if handovers land in a separate inbox. Send summaries to the place the team already looks, and review the log once a day.
None of these is a reason not to start, but each is a reason to start small: one channel, clear rules, and a review after the first fortnight.
Questions to settle before you start
- Who owns the facts list (hours, menus, policies), and how quickly is it updated when something changes?
- Who receives handovers on each shift, and who is the backup?
- What is the reply time for booking requests and party enquiries, and how is it checked?
- How will guests be told they are speaking to an AI assistant, and how do they reach a person?
- Where are call recordings and transcripts stored, and for how long?
Hiring remains hard in the sector: 12.7% of accommodation and food businesses reported difficulty recruiting in September 2026, against 7.1% of all businesses[6][7]. Taking routine phone and admin work away from managers will not fix that, but it makes the existing team's shifts more manageable. For the numbers side, see building a business case for AI call handling; for buying, choosing an AI phone provider; and for the guest side, what guests ask in website chat.
Frequently asked questions
What is the difference between an AI employee and an AI receptionist?
A basic AI receptionist answers and takes a message. An AI employee completes routine jobs within your rules, such as capturing a full booking request or party enquiry, and passes the rest to the right person with a summary.
Will an AI employee replace front-of-house staff?
Not in a well-run venue. It takes repetitive phone and admin work so staff can focus on guests. National survey data shows very few businesses using AI have reduced headcount as a result.
Which jobs should never be handed to AI in a restaurant?
Allergen safety decisions, complaints and refunds, refusing service, licensing and safeguarding duties, and decisions about staff such as hiring, hours and discipline.
How do I know which tasks to hand off first?
Log every interruption for two weeks, group them by topic and pick the ones where the answer or the details collected are the same every time.
References
Office for National Statistics (2026). Business insights and impact on the UK economy: 8 October 2026 (opens in a new tab) ↑
Office for National Statistics (2026). Business insights and impact on the UK economy, wave 165 dataset (AI sheets) (opens in a new tab) ↑
Office for National Statistics (2026). ASHE 2026 provisional, Table 14.6a: hourly pay excluding overtime by occupation (SOC 2020), all employees (opens in a new tab) ↑
Department for Science, Innovation and Technology (2026). AI Adoption Research (opens in a new tab) ↑
Office for National Statistics (2026). Business insights and impact on the UK economy: 8 October 2026 (opens in a new tab) ↑
Office for National Statistics (2026). Business insights and impact on the UK economy, wave 166 dataset (opens in a new tab) ↑
Office for National Statistics (2026). Business insights and impact on the UK economy, wave 166 dataset (opens in a new tab) ↑