How The College of West Anglia Turned Website Chat Into Enrolments and Insight

The challenge: CWA's rule-based website chatbot gave wrong answers, buried any real insight into what prospective students were asking behind hours of manual review, and had no way to capture who was asking or hand off what it couldn't answer.

The achievement: BabelQuest deployed HubSpot's Breeze Customer Agent inside a strict budget, live in time for CWA's September intake, so each enquiry it handled became both an answer for the prospective student and a data point for the business.

Standout numbers:

• 473 conversations handled since go-live in late August 2025
• 323 resolved autonomously - a 68% resolution rate, with no added headcount
• ~1 day average time to resolution
• Comfortably within budget, months into a fixed, governor-approved credit allowance

About The College of West Anglia: The College of West Anglia is one of the largest education and training providers in Norfolk and Cambridgeshire, supporting prospective and current students through every stage of their journey. With marketing, admissions, and employer engagement teams all working in HubSpot, CWA needs a shared view of each prospect’s interactions to deliver joined-up, timely communication.

Further education is a conversion business, and the window in which a prospective student is actively deciding is short - and rarely falls within office hours. A 19-year-old working out course eligibility and funding is just as likely to be doing it on a Sunday evening as a Tuesday afternoon - and the old chatbot being switched on for that moment meant little if it couldn't answer, and couldn't even capture the question for someone to follow up later.

CWA's existing website chatbot wasn't equipped to meet that moment:

• No control over its answers. The bot pulled from uncontrolled sources and would confidently give wrong information - including, memorably, telling prospective students the college had a swimming pool. It doesn't.
• Visibility that didn't scale. The team could review what people were asking, but only by opening each conversation individually - there was no reporting to spot patterns, and no way to see who had asked, so the same question showing up fifty times looked no different to one showing up once.
• No data capture. Conversations weren't tied to a known contact, resolved or not - so even the enquiries the team could see carried no way to follow up, and any query the bot couldn't answer was simply lost rather than handed off.

As Natalie Metcalfe, Digital Marketing Coordinator at CWA, put it:

The result was missed opportunity at both ends of the funnel: prospective students left without answers, and a team with no insight into how to close the gap.

BabelQuest's approach started with a principle, not a product install: the Breeze Customer Agent would be onboarded like a new team member - with guardrails and coaching - not switched on and left alone.

A workshop before deployment. BabelQuest consultants Gem and Laura ran a session attended by CWA showing how a chatflow and the Customer Agent work together, setting expectations that the agent would need ongoing management rather than a one-off setup.

A tiered, credit-aware resolution flow, built around a fixed budget. BabelQuest made that constraint the centre of the design: the agent answers first from CWA's own trusted content - website pages, knowledge articles, campus documents, FAQs - with stray sources deliberately excluded. Where the knowledge base can't fully answer, the AI agent reasons over the same material; where it hits its confidence limit, it captures the student's email and hands it off to a human. A non-AI chatflow sits behind it as a failsafe, ready to take over if usage ever nears the credit limit. Months after go-live, CWA remains comfortably within budget.

Deliberate hand-off, not failure. A college can't staff live chat around the clock, so anything the agent can't resolve is captured cleanly, with the question and contact already logged, ready for a human to pick up rather than lost overnight.

Timed for peak intent. Go-live was set for late August 2025, immediately ahead of the September intake, so the agent was live at the exact moment enquiry volume spiked.

A second, custom agent closing the loop. BabelQuest built a further custom Breeze assistant that mines chat transcripts and support tickets for where the main agent falls short, then outputs a prioritised list of content gaps to fill - feeding straight back into the knowledge base to lift the resolution rate over time.

Everything runs inside HubSpot: the agent answers from HubSpot-held content, writes contacts back against the correct lifecycle stage, and is fully reportable alongside CWA's other data, with no separate system to maintain.

Since going live in late August 2025, the Breeze Customer Agent has handled 473 conversations, resolving 323 of them - a 68% resolution rate - entirely autonomously and with no added headcount. Average time to resolution has come down to roughly one day, and usage has stayed comfortably within CWA's fixed credit budget throughout.

The deflected enquiries are only half the impact. The custom reporting BabelQuest built alongside the agent - something most Customer Agent deployments never attempt - turns every conversation into intelligence CWA can act on:

• Who's chatting: volumes segmented by lifecycle stage, so content and routing can be tailored to prospective students, current students, and employers differently.
• Where the website falls short: reporting on which pages trigger chats surfaced heavy demand around exam booking, traced to a specific gap in the website - letting CWA fix the page itself and head off future chats at the source, rather than just answering them one by one.
• What content to prioritise: where the same question comes up repeatedly - such as how much a specific course costs - the team now uses that signal to make the answer more prominent on the relevant page, rather than leaving it buried for prospective students to have to ask.
• Turning misses into improvements: every "unhelpful" rating is reviewed for patterns, feeding directly into agent coaching or new knowledge-base content.

For the team, that means a lean marketing and admissions function has gained what is effectively a 24/7 front-line colleague, and any conversation that does need a human arrives with the question and contact already captured - nothing to chase, nothing lost overnight. For prospective and current students, it means accurate, instant answers at any hour, replacing a bot that was technically always on but left them stuck with an unresolved query and no way for anyone to follow up when it couldn't help.

CWA's confidence in the results extended beyond their own use of it: the college co-presented the work with BabelQuest at their AI Engine workshop, sharing the approach with peers facing the same conversion and capacity pressures.

Looking ahead, CWA and BabelQuest are closing the content gaps the reporting surfaces to raise the resolution rate further and conserve credits; turning chat demand into website and Answer Engine Optimisation (AEO) content; building lifecycle-aware routing that alerts the right team when a known prospect, employer or even a parent starts a chat; and quantifying the time and cost saved so far to extend agentic automation across the rest of the college.

If your team is trying to answer prospective students, customers, or members around the clock without the headcount to do it - and without visibility into what they're actually asking - this is the kind of transformation available to you within your own budget constraints.

Get in touch with BabelQuest to talk through what an AI-driven agent inside HubSpot could look like for your organisation.