ChatGPT
Claude
Perplexity
Gemini
GrokThe three tools do different jobs, so compare jobs rather than features
An AI admissions agent talks with prospective students and answers in real time. A basic chatbot answers a fixed list of questions. CRM automation runs rules and sequences on records you already hold. They are not three versions of one product, and choosing between them starts with the job you need done.
Many admissions teams ask the question as a purchase: agent or automation? The better question is where an applicant is losing time or context today. A student asking at 10pm about entry requirements needs an answer, not a scheduled email. A student who attended an open day last week needs a timely follow-up, not a new conversation.
This guide gives a decision matrix, describes who should own each tool and sets out the data protection points specific to the UK. It builds on our pieces on AI agents from first contact to CRM and the CRM comparison for higher education.
Definitions: agent, chatbot, automation
The terms are used loosely by vendors, so fix them before comparing.
- FAQ chatbot. Follows a scripted tree or matches questions to prepared answers. Predictable, limited, and it fails when the question is phrased in an unexpected way.
- AI admissions agent. Understands free-text questions, answers from content the institution has published, asks qualifying questions, offers an appointment or open day and hands the conversation to a person. The glossary of admissions AI agents defines the vocabulary.
- CRM automation. Triggers actions from data in the CRM: a status change sends an email, a task is created, a record is assigned. It acts on what is already known and does not hold a conversation.
The line to remember: the agent works at the moment of the question, automation works on the record afterwards.
The decision matrix
The table sets the three approaches against the jobs admissions teams actually have.
| Job to be done | FAQ chatbot | AI admissions agent | CRM automation |
|---|---|---|---|
| Answer an entry-requirements question at night | Only if the question was anticipated | Yes, from published content | No |
| Qualify an enquiry (course, entry route, timing) | Fixed form-like questions | Conversational, adapts to the answers | Only from data already captured |
| Book an open day or call | Link to a form | Can offer slots in the conversation | Confirmation and reminder emails |
| Send a follow-up after an open day | No | Not its role | Yes, this is its strength |
| Keep records consistent and deduplicated | No | Passes data to the CRM | Yes, with matching rules |
| Route an enquiry to the right adviser | Basic | Can summarise context for the adviser | Yes, rules and queues |
| Handle a sensitive or complex case | No | Hands over to a person | Creates a task |
No column wins every row. The agent is strongest before a record exists, automation after it exists.
Who owns what
A tool without an owner decays within a term. Decide ownership before signing anything.
| Component | Natural owner | What they decide |
|---|---|---|
| Published answers the agent relies on | Admissions with marketing | Which pages and policies are the source of truth |
| Escalation rules to humans | Admissions manager | What triggers a handover, and the response time |
| CRM fields, statuses and sequences | CRM or marketing operations | Data model, triggers, sending rules |
| Consent and suppression rules | Data protection officer | Lawful basis, channels, retention |
| Quality review of conversations | Admissions with the supplier | Gaps in answers, wrong answers to correct |
If these roles sit with different people who never meet, expect contradictions: the agent promises a call that no sequence schedules, or an automation emails someone who just told the agent to stop.
When to choose one, and when to combine
Start from the bottleneck you can observe in your own data, not from the technology.
- Prospects ask the same questions outside office hours and many go unanswered. Start with an agent on your website, fed from your published pages.
- Enquiries arrive but follow-up is slow or inconsistent. Start with CRM automation: assignment rules, reminders, task creation.
- Records are duplicated or incomplete. Fix data before adding either; both amplify bad data. Our guide to deduplicating student records across channels covers this.
- Both gaps exist. Combine, with the agent capturing a qualified enquiry and the CRM automation owning everything after that.
The combination works when the boundary is explicit. The agent collects, qualifies and books. It passes the conversation summary and the consent record to the CRM. From that point, sequences, reminders and reporting belong to the CRM. A team that lets both tools email the same applicant has built a collision, not a system.
One further option is to do nothing new yet. If your published information is out of date, an agent will answer from outdated content, and automation will send it faster.
Data protection: UK data protection law, the Data Protection Act 2018 and PECR
Both approaches process personal data of applicants, many of them 17 or 18. The Information Commissioner's Office publishes guidance for organisations on UK data protection law and the Data Protection Act 2018, and it is the reference for the points below.
For an agent. Tell the person they are speaking with an AI and what happens to what they type. Decide the lawful basis for processing the conversation, and avoid collecting more than the enquiry needs, particularly special category data such as health or disability details volunteered in free text. Set a retention period for transcripts.
For automation. Electronic marketing is governed by the Privacy and Electronic Communications Regulations (PECR) alongside UK data protection law. Consent and the "soft opt-in" rules differ by channel, so an applicant who agreed to an email about an open day has not agreed to marketing by text message or messaging apps. Record consent by channel, with date and source.
For both. Suppression must work across tools. If someone objects in the chat, the objection has to reach the CRM, or the next sequence will contact them anyway. Have your data protection officer review the flow end to end. Automated decision-making deserves attention too: an agent that qualifies enquiries should route and inform, not reject applicants. Admissions decisions stay with people and with the institution's published criteria, within the UCAS process where it applies.
UK context: UCAS, Clearing and the cycle
The UK cycle shapes where each tool earns its keep. Questions peak around UCAS deadlines and again at results time, when Clearing compresses decisions into days. An agent can absorb the question volume in those weeks without adding a queue. Automation handles the structured steps: acknowledging an enquiry, reminding about an open day, prompting a missed deadline.
Two cautions. First, anything the agent says about entry requirements, fees or availability must come from your current published pages; during Clearing, availability changes quickly and a stale answer misleads a candidate under pressure. Second, institutions regulated in England should keep in mind their obligations to students under consumer law and Office for Students conditions; accurate course information is part of that, whoever or whatever delivers it.
What to ask a supplier before choosing
Questions that separate real capability from slides:
- Which sources does the agent answer from, and what does it do when it has no source?
- How does it hand over to a person, and what does the adviser receive?
- Where is data hosted, and is it used to train models?
- Which CRM objects does it read or write? Ask for this in writing, not in a demo.
- How are consent and objections passed between the agent and the CRM?
- Who reviews transcripts, and how often?
Measure your own baseline first: response time outside office hours, share of enquiries followed up within a day, duplicate records. No public source gives a reliable benchmark for your institution, so use your own before and after.
Where Skolbot fits
Skolbot Chat is an AI agent installed on the institution's website. It answers from published content, qualifies the enquiry, offers an appointment or open day and passes the conversation to the team and the CRM. It complements advisers; it does not replace them. CRM sequences and reporting remain in your CRM. How the connector works with a given CRM is confirmed during scoping, not assumed. For the wider picture, see our guide to recruiting more students.
