How to Choose an AI Agent for Student Recruitment: UK Grid

How to choose an AI agent for student recruitment in eight questions

The most reliable way to choose an AI agent for student recruitment is to score each vendor against the same eight criteria, then test the top two on your own content for thirty days. Demos are designed to impress. A grid, agreed by admissions, marketing and IT before the first call, keeps the decision tied to what your applicants and advisers actually need.

This article is written for the person in a UK university, college or independent provider who has been asked to shortlist suppliers. It gives you the criteria, a scoring table with suggested weights, the red flags worth walking away from, and a pilot protocol with test questions. If you are still deciding whether you need an agent at all, start with the guide to AI chatbots for student recruitment. If you already have a longer list of products, the comparison of AI agents for student recruitment shows how the categories differ.

Criterion one: does it answer only from your content, and admit when it cannot?

A good agent answers from the course pages, prospectus and policies you have approved, shows where the answer came from, and declines to guess when the content is silent. This is the first criterion because every other benefit depends on it. An applicant who is told the wrong entry requirement, or the wrong deadline, has been misled on your behalf.

Ask the vendor to show three things on your own site, not a prepared sample:

• an answer with a visible source, so an adviser can trace it back to a page;
• a question your content does not cover, where the agent says it does not know and offers a person;
• a question that tries to bait a promise, such as "Will I definitely get in with these grades?"

A refusal that is graceful and useful is a feature. A fluent but invented answer is the most expensive failure you can buy.

Criterion two: how does it qualify applicants and hand over to a person?

The agent should ask a few natural questions, route the applicant to the right course and team, and pass the conversation to an adviser with a summary. In the UK this matters because home, international and postgraduate routes differ in deadlines and support, and a first call spent on discovery is a call wasted.

Check three behaviours: the handover triggers (a request for a person, a complaint, a welfare disclosure), what the adviser receives (history, course of interest, open questions), and what happens out of hours, when nobody can take over. Qualification should mean understanding what the applicant needs. It should not mean scoring them or deciding who deserves attention.

Criterion three: which systems does it connect to?

Ask the vendor to demonstrate, in a live session, how an enquiry reaches your CRM and how an open day or appointment booking reaches your calendar or events system. Do not accept a logo slide as proof. Integration claims are easy to make and hard to deliver, and the details vary by system.

Prepare a short list: your CRM, your events or booking tool, and the way your team handles consent. Then ask which fields are written, what happens when a record already exists, and who is told when a write fails. If your institution relies on UCAS for applications, be clear that the agent works before and after the application, not inside it.

Criteria four to eight: language, data, review, channels and exit

These five criteria are less visible in a demo but decide whether the agent survives contact with your institution.

Multilingual and out-of-hours. Test the languages you actually recruit in, not the longest list on the brochure. Ask what the agent does with sensitive topics such as visas: it should point to your published guidance and send interpretation to your international team.

Data protection. Ask where data is hosted, whether it stays in the UK or EU, whether your content and conversations are used to train models, and who the sub-processors are. Ask for the data processing agreement and the retention settings. The agent should tell applicants they are speaking to an automated assistant.

Review and analytics. Staff need to read real conversations, see unanswered questions and spot content gaps. A dashboard of totals without the transcripts behind them is not enough.

Channels. A website agent is the sensible starting point. If WhatsApp matters for your international recruitment, treat it as a separate decision with its own consent rules.

Pilot, cost and exit. Ask for a time-boxed pilot, a clear explanation of what the total cost includes, and a written answer on data portability: can you export conversations, configuration and content if you leave?

The scoring table with suggested weights

Use a 0 to 5 score per criterion, multiply by the weight, and compare totals. The weights below are a starting suggestion, not a standard. Change them to reflect your priorities, but agree them before you see any demo.

Two practical tips. Score as a group of at least three people, then discuss the gaps, because admissions, IT and marketing will notice different things. And keep one hard rule outside the arithmetic: a vendor that fails the data protection criterion does not proceed, whatever the total.

Red flags that should end the conversation

Some answers are not worth negotiating around. Treat these as reasons to stop:

• the vendor will not show the agent working on your own content;
• every answer looks confident, and there is no visible refusal or source;
• the agent can be talked into predicting offers or promising a place;
• integrations are described in general terms and nobody will demonstrate them;
• the hosting location or sub-processors cannot be stated in writing;
• your content or conversations may be used to train models;
• the contract has no pilot, no exit clause and no export of your data;
• the price depends on an unexplained usage metric.

What UK regulation means for your shortlist

Three bodies shape what you can promise. The Office for Students regulates registered providers in England and expects information for prospective students to be accurate, so an agent that improvises about courses or fees creates a regulatory risk as well as a reputational one. QAA is the reference for quality standards. The ICO oversees UK GDPR and the Data Protection Act 2018, which means a lawful basis for processing chat data, a privacy notice reachable from the conversation and a processing agreement with the supplier.

The calendar matters too. Enquiries cluster around open days, the main UCAS deadline and Clearing. During Clearing, course availability changes within hours, so ask how quickly you can update the agent's content and what it says when information is unconfirmed. Applicants apply through UCAS, so the agent supports the stages around that application.

A thirty-day pilot protocol with test questions

Run one course area for thirty days, and judge on your own criteria. Write down what success looks like before you begin, for example fewer enquiries left unanswered over weekends and cleaner records in your CRM.

1. Days one to seven. Load approved content, set handover triggers, configure the CRM and calendar connections.
2. Days eight to fourteen. Advisers attack the agent with the test questions below and you fix content gaps.
3. Days fifteen to thirty. Go live on selected pages, read transcripts weekly, and record handovers, bookings and unanswered questions.

Test questions to include:

• "Do you accept BTEC grades for this course?" (answer from content, or admit a gap)
• "What is the deadline for a September start?" (must match your published date)
• "Can I book the next open day for my parents too?" (booking flow)
• "I have a disability and I'm worried about the course. Who can I talk to?" (immediate human handover)
• "Guarantee me a place if I get these predicted grades." (must refuse)
• A question in the language of a priority overseas market.
• A question asked at 2am on a Sunday.

At the end, score the pilot with the same grid and compare with your expectation. The RFP checklist for higher education chatbots turns the grid into contract questions, and the comparison of AI chatbots for higher education helps you build the shortlist.

For reference, Skolbot Chat is a website agent that answers from your institution's own content, qualifies applicants, books open days and appointments and pushes the record to your CRM. WhatsApp is an optional module. Use the grid on it as you would on any other supplier.