From the live discussion and the questions sent in at registration. As promised on the call, every question is answered. Pick a topic, then click any question to read the answer.
Do I really need to map my process?
Honestly, no, and please don't let it hold you back. It's one of the most common reasons we see projects stall before they've even started. Teams feel they need a perfect map of everything first, and months later they're still mapping.
Anglia Ruskin University (ARU) is a great example of the alternative. They got on with it, started with a real process, and improved it as they went. That's now become continuous improvement, and by their own admission, if they'd waited until every process was mapped, they'd never have got started.
And if you do want the map, you don't have to build it by hand. Our process discovery agent, NiicoVU, observes how the work actually gets done, then gives you the process map, a standard operating procedure, and a view of which steps are suitable for automation. So you can see how things really work, not how everyone remembers them working, without taking your team away from the day job to document it.Alistair Sergeant
We're implementing our first agent this month or next. Any ideas?
Brilliant, and good luck. My tips: pick one narrow, high volume process, measure how it works today before you start so you can show the difference, and bring your team in early. Happy to have a chat if it would help.Alistair Sergeant
What practical first steps would you recommend for implementing agents in core systems like SITS?
Pick one process, talk to your SITS and IT teams early to understand your integration options, do your DPIA, then run a pilot in the background before anyone relies on it.
Alistair Sergeant & Josh ConlanHow can AI do the initial assessments?
The agent reads the application, checks it against your entry requirements, confirms the right documents are there, and produces a recommendation with its reasons. A person confirms it. So your team starts from a prepared case rather than a blank screen.Alistair Sergeant
How do you decide what can be automated and what needs a human, and how does the handover work?
I think of it in three groups. Clear rules and clear evidence: automate. Evidence that needs interpreting: the agent prepares and a person decides. Judgement or sensitivity: a person from the start. The handover happens when the agent flags a case with its reason, and it lands in a queue your team already uses, with a summary and the evidence attached, so nobody has to start from scratch.Alistair Sergeant
How do you choose the right AI system when so many look similar and are still developing?
A few questions sort the field quickly. Can it show you why it made a recommendation? Where is your data hosted, and is it used to train anyone's models? Does it work with your existing record system? Can you speak to a similar institution already using it? What happens to your data if you leave? And is the pricing predictable? Then run a small pilot with clear success measures. Buy the evidence, not the demo.Alistair Sergeant
How can an agency improve its approach to recruiting high quality students?
Help students qualify themselves earlier. AI can help applicants check their eligibility and understand requirements before they apply, which means more complete applications, fewer wasted ones, and better conversion for everyone.Alistair Sergeant
How does it integrate with standard platforms? We use SITS and eVision.
Lots of you are on SITS, so you're in good company. Our agents sit alongside your student record system rather than replacing it. We connect through web services and APIs, or RPA where no API exists, so your team keeps working in the systems they know and SITS stays the source of truth.Alistair Sergeant
Would it work alongside our CRM, or would applications go through your system?
Alongside. Your existing systems remain the source of truth, and the agent works on top of them. Nothing needs to be replaced, so you can benefit from this technology and see the impact very quickly.Alistair Sergeant
Could we see how staff review the agent's recommendations, and where this sits if it overlays existing systems?
Completely agree, it's much easier to show than describe. We'll share a short walkthrough.Alistair Sergeant
What's the distinction between AI and automation?
Automation follows the rules you give it, the same way every time. AI handles the messy stuff automation can't, like reading a transcript, making sense of an email, or summarising a personal statement. The sweet spot is using both together: AI reads and understands, automation carries out the steps, and your people make the decisions. For the basics, download the
AI and automation guide Alistair mentioned on the call.
Alistair SergeantWhat does RPA stand for?
Robotic Process Automation. It's software that does the clicks and keystrokes a person would do in a system. It's really useful where a system doesn't have an API to connect to.
Alistair SergeantDo students need to give grades in structured fields for the agent to compare them with entry requirements?
It helps, but it isn't essential. The agent can read uploaded transcripts and certificates too. Structured data gives the most reliable results, and unstructured documents get extra checks.
Alistair SergeantIs this just for UK qualifications, or overseas ones too?
Both. For international qualifications, the agent works from your own equivalency guidance, so it's applying your rules, not inventing its own. Anything unusual or unfamiliar gets flagged to a person.Alistair Sergeant
Will the session cover applications from international students?
We touched on it, and the answer above covers the core of it. Happy to go deeper on a call, as international applications tend to be where teams feel the most pressure.Alistair Sergeant
Will it set up online interviews, identify key positives and negatives, and record outcomes?
Scheduling and recording outcomes into your system, yes. On judging the interview itself, I'd be cautious. It can help summarise the interviewer's notes, but I wouldn't recommend AI assessing candidates in an interview.Alistair Sergeant
Which admissions decision would you never hand entirely to AI, and why?
Rejections, particularly anywhere judgement is involved. Also anything touching safeguarding, disability, mitigating circumstances or conviction declarations. These affect someone's life in a real way, and the applicant deserves to know a person looked at it.
Alistair Sergeant, with Rachel Reeds in the clipIf a client said "just automate the whole thing and push the offers out", would you?
No. The technology can do it, but just because we can does not mean we should. Agents should take away the laborious manual work institutions have inherited through technical debt and poor processes, so admissions officers can spend their skill on complex cases. The platform does the heavy lifting, and a person always stays in the loop for the final call.
Alistair SergeantWhere does the decisions agent add most value?
At pinch points like the weeks after the UCAS January deadline. Volumes are huge, the pressure is on to make conditional offers, and most applicants are school and college leavers with tariff qualifications that fit neatly into boxes. You cannot borrow the recruitment team for two weeks while they are out at events. Taking the straightforward cases away gives time back for the non standard ones, which are the ones that get rushed now.
Rachel ReedsHow does the agent handle complex qualitative entry criteria, and what's the error rate?
We work with your team to turn the criteria into clear rules, then test the agent against your past decisions before it goes live. It has to show where it found the evidence for each criterion, and when something isn't clear, it refers rather than guesses. On error rate, I'd rather give you a real number from your own data than a headline one, which is why we run a pilot comparing the agent's recommendations with your team's decisions.Alistair Sergeant
How do you measure and demonstrate accuracy, and if an applicant challenges a decision, how can an officer see how the agent reached its recommendation?
We run the agent in the background first, comparing its recommendations with your team's decisions before anyone relies on it. Every recommendation comes with its reasoning and the evidence it used, so an officer can see which criteria were met and where. If a decision is challenged, there's a clear audit trail, and the final decision was made by a person.Alistair Sergeant
How does this work with significant PSRB regulated provision, where requirements can't be contextualised?
Those requirements are set up as non negotiables that the agent checks for and can never contextualise away. Exceptions only happen through a person. We configure this with your team, based on your PSRB rules.Alistair Sergeant
How do you balance complexity in offer making with conversion, personalisation and automation?
Let the standard cases move fast, because speed itself helps conversion. For conditional or contextual offers, the agent does the preparation and a person adds the judgement and the personal touch. Automation buys your team the time to be more personal where it matters, not less.Alistair Sergeant
How would an AI agent handle portfolios, for example for art courses?
It can check the portfolio is complete, in the right format and matches the brief, and organise it for review. The artistic judgement stays firmly with your academics. It's actually a lovely example of AI clearing the admin so academics spend their time on the work itself.Alistair Sergeant
Honestly, how many personal statements are actually read?
Nobody likes to admit it. There is little research, but a HEPI study of Russell Group institutions found anywhere from 10% to over 90% being read. They are read more in competitive institutions where academics review before offer making, and for professional courses. Some institutions say openly that they do not read them.
Rachel ReedsWhat is the risk of not reading them?
Something materially important may be disclosed: a need for extra support, a safeguarding issue, or missing qualification information from an applicant filling in the form without a school's help. Rachel has made Prevent referrals from what she read in personal statements. You hold the data, so if something was disclosed and nobody acted, the applicant will say they told you. An agent can raise the flags without the team reading every statement for straightforward courses.
Rachel ReedsHow accurate is sentiment analysis in personal statement review, and how does it link to other checks?
Personal statements deserve to be read and understood properly, but at high volumes they often get a quick skim at best. That's a missed opportunity, because it's where applicants tell you who they really are.
AI reads every statement in full and picks out what matters: hobbies, motivations, experiences, and circumstances or support needs they may not have declared elsewhere. That sits alongside your other checks, giving your team a richer picture of each applicant. So you can personalise every conversation around what each applicant genuinely cares about, and that's what drives conversion.
On accuracy, the best test is your own applicants, which is why we check it against your statements during a pilot. It's there to understand applicants, not score them, so your team keeps the judgement and gains the insight.Alistair Sergeant
Does the agent detect whether a personal statement was written by AI?
It can, but honestly, AI detectors aren't reliable enough to base decisions on. They produce false positives, and research from Stanford in 2023 found they're more likely to wrongly flag writing by non native English speakers, which is a real fairness risk in admissions. The better question is probably what you want the personal statement to do, and weighting it accordingly.Alistair Sergeant
How do you balance efficiency with applicants' expectations of human judgement and empathy?
Use the time you save for the human bits. Let the agent handle "have you received my transcript?" so your team has time for "I'm really worried about my results." And be open that AI is part of the process. People are generally fine with it when they know a human is making the decisions.
Alistair SergeantWhy is fee status such a bugbear?
There are around 50 categories, with multiple criteria under each. The hard part is matching what the applicant says with the evidence. Someone may say they are British, then provide indefinite leave to remain, and being British does not make you home if you have lived overseas. At post 92 and alternative providers, a UK born 18 year old with A levels is the exception, so most applicants need reviewing to some degree. And the rules change constantly.
Rachel ReedsHow and when do you approach fee status assessments?
As early as possible, ideally when the application arrives, so there are no surprises later for the applicant or your team. The agent gathers and organises the information, and anything uncertain goes to a specialist.Alistair Sergeant
How do you distinguish automating an administrative assessment from automating professional judgement? For fee status, what triggers a referral to a specialist?
Great question. The agent's job on fee status is to gather, organise and highlight, not to decide borderline cases. Referral triggers include incomplete or conflicting evidence, gaps in residence history, categories where discretion is required, and anything the applicant disputes. If it isn't clear cut, a specialist makes the call.
Alistair Sergeant, with Rachel Reeds in the clipCan the agent tell if a document such as a passport is fraudulent?
It can spot obvious inconsistencies, like names or dates that don't match across documents. Proper fraud detection needs specialist identity verification tools, and the agent can hand over to those. I'd never want to overclaim here.
Alistair SergeantCan it verify references from overseas companies?
Our agents can help by checking the reference came from a genuine company domain, cross checking the company against public registers and flagging inconsistencies. Final verification should stay with a person, especially where something looks off.Alistair Sergeant
What checks should we carry out to make sure we're compliant with data protection when choosing and working with AI?
Involve your DPO early. The key checks are: a DPIA, your lawful basis, where data is stored and processed, whether your data is used to train models (it shouldn't be), a proper data processing agreement, the supplier's sub processors, retention periods, security certifications, and how you're handling Article 22 on automated decision making.Alistair Sergeant
Can applicants opt out of agentic processing? How do you manage consent or lawful basis, and is it transparent?
It should always be transparent, set out clearly in your privacy notice. Consent is usually a tricky basis in admissions, so institutions more often rely on public task or legitimate interests, but that's for your DPO to determine. Where a person makes the final decision, and applicants can ask for human review, you're on much firmer ground. [Confirm our position on opt out.]Alistair Sergeant
What guardrails are in place, and what CRM integration is available?
The agent only works from your approved content and rules, can't make final decisions you haven't signed off, refers when confidence is low, and logs everything for audit. We integrate with all systems, so there is no limit on the type of system you need to benefit from this.Alistair Sergeant
In special education, with tight eligibility criteria and sensitive documents like psychometric and clinical reports, how can AI help while meeting privacy regulations?
This is exactly where I'd be most careful. Health and psychological reports are special category data under UK GDPR (or your local equivalent), so the bar is high. I'd keep AI on the administrative side: checking documents are complete, organising them for your panel, scheduling, and keeping families updated. The selection decision stays with your professionals. Start with a Data Protection Impact Assessment, and only let the AI see what it genuinely needs to see.Alistair Sergeant
Can we use AI to process consultation placements, aligned with our internal systems, and how secure is it?
Yes, matching, scheduling and document checks are good fits, connected to your systems through integration. On security, we'd work through a Data Protection Impact Assessment with you, and data is [confirm: hosted in the UK/EU, encrypted, with role based access].Alistair Sergeant
What are the set up costs, staff time and risks, including if AI becomes too expensive, running locally, security and errors?
Costs depend on use cases and volume. On staff time, the main commitment is a few mapping and testing sessions, and training tends to be light because staff review recommendations in familiar tools. On risks: errors are managed through human review, audit logs and starting narrow. Running models locally is possible with open models, though there are trade offs in capability and upkeep. Security comes down to hosting, access controls and a proper data processing agreement, and we're happy to go through ours in detail.Alistair Sergeant
Do you expect costs to stay stable given token pricing?
Nobody can honestly promise that. What we've seen so far is that the cost of a given level of capability has generally come down over time, but pricing at the top end moves around. Our costs are fixed for contracted clients.Alistair Sergeant
What are admissions teams telling you about AI?
They ask sharp questions about balancing compliance, efficiency and risk, at a time when they have no capacity left for new projects. The biggest fear is putting sensitive applicant data somewhere they do not understand. There is also tension between efficiency drives from above and what makes the work better on the ground. Mostly, people are overwhelmed and want specifics rather than the abstract.
Rachel ReedsIs the fear and anxiety getting any better?
It depends on the institution: its applicants, finances and IT strategy. Some teams run admissions technology over 20 years old with countless workarounds. Admissions often has little say in new systems, but tools that work across the student journey are an easier case to make. Rachel's advice is to ask what you can do with what you have now, rather than trying to fix everything at once.
Rachel ReedsSchools are terrified of AI. How do we reassure them it's safe?
Start with what it isn't doing: it isn't making decisions about children. Then show them. Walk them through what data is used, where it's held, and where the human sits in the process. Fear usually comes from the unknown, and a short demo does more than any policy document.Alistair Sergeant
What has worked in building confidence with hesitant stakeholders?
Show, don't tell. Bring the sceptics in early, because they'll spot the real risks. Start with a process everyone dislikes, be clear about what AI won't do, and share pilot results honestly, including what didn't work. And give your team a say in where the line between AI and human sits.
Alistair Sergeant, with Rachel Reeds in the clipHow can AI help manage high volume, repetitive enquiries while keeping human oversight? Any examples of reduced workload?
The agents answer the repetitive questions around the clock using only your approved content, and hand over to a person when the question is sensitive or they aren't confident. That means your team starts the day with the conversations that genuinely need them, not a backlog of the same ten questions. At Anglia Ruskin University, admissions inbox emails fell from 700 a day to 50, with 93% of enquiries resolved automatically.
Alistair Sergeant, with Rachel Reeds in the clip 
