I improved AskINTO support chatbot so Regional Officers (ROs) and Enrolment Support Officers (ESOs) find answers 80% faster, which gives them more time to convert students before a competitor does.
Role
UX Designer & Writer
Company
INTO University Partnerships
Timeline
3 months
Team
Developers, project managers, stakeholders
Platform
Inside Salesforce
BeforeSaid 50, showed none
AskINTO
My student from Nigeria wants to study a bachelor's in mechanical engineering in the UK. What are the options?
I have found 50 courses related to mechanical engineering in the UK. To narrow down the options, could you share more details about the student? Academic grades, English level, budget, preferred university, intake.
AfterAnswers first, then offers next steps
AskINTO
My student from Nigeria wants to study bachelors in mechanical engineering in the UK. What would be the options for the student?
I have found 50 courses related to mechanical engineering in the UK across the following universities. These include options for direct entry, BEng pathways, international foundation years, international year one and courses with a year in industry:
Newcastle University
University of Exeter
University of East Anglia
Queen's University Belfast
University of Stirling
Would you like me to: A) Show detailed information about the 5 programs listed above, such as tuition fees, duration, entry requirements etc? B) Refine the list by budget and intake dates C) Explore direct entry options instead? D) Explore pathway options instead?
01
In short
Problem
AskINTO was built to save ROs and ESOs from answering the same questions every day. The first MVP version of the chatbot gave answers they couldn't trust, so they kept checking Excel sheets instead. A slow or wrong answer can send students and agents to a competitor which affected the revenue.
What I did
I joined after the MVP was launched, I audited the chatbot, tested my assumptions with users and led a phased plan: conversation principles, rewritten conversations, answers tied to SharePoint documents and a weekly feedback loop.
Outcome
About 80% less time to find an answer. Daily users went from close to zero to 36. Drop-off is down, completion is up, and answers are more accurate.
02
Context
AskINTO is a support chatbot that answers common admission related questions, helps users make decisions, and reduces the need for manual support.
Built for now
ROs and ESOs
Manage student applications and support agents.
IN FUTURE
Agents
Counsel students and need quick guidance while they handle applications.
Both groups often work under time pressure. They need clear, reliable answers without switching tools or digging through documents.
User story
One simple question, one lost student
What happened when Regional Officers couldn't trust the first version of AskINTO.
“Quick one about this student…”
An agent sends a question
A student asks their agent, who messages an RO.
AskINTO: Here’s your answer
The RO asks AskINTO
An answer comes back in seconds.
“Is this actually correct?”
The RO doesn't trust it
A wrong answer could cost the student their place.
ExcelShared docsWebsites
The RO goes back to spreadsheets
They check Excel and shared docs themselves.
Competitor ✓INTO
The agent moves on
Agents earn commission when a student enrols, so they go with whoever answers first.
Enrolled elsewhere
Competitor
INTO loses the student
They enrol with a competitor, and the revenue goes with them.
Built toGive faster answersTake repetitive questions off ROs.
What happened insteadDouble the workEvery answer had to be double-checked.
The real problemNot speed, it's trust.Officers won't rely on an answer they can't trust.
03
Where I came in
AskINTO was already live when I joined. The first version, an MVP, had gone out to internal users. Answers weren't accurate, they were hard to trust and people weren't using it.
I started with a full audit. Then I tested my assumptions with internal users. I took both back to the team and we agreed a plan to fix it one phase at a time.
Already liveMVP with internal users
Step 1Full audit of the bot
Step 2Assumptions tested with users
Step 3Phased plan, agreed with the team
04
The problem
In testing, I kept seeing the same thing. A participant would ask AskINTO a question, read the answer, then open an Excel sheet, documents and website to check it.
Some said it plainly: it was quicker to find the answer themselves. They were using the chatbot only because they'd been asked to.
Why it matters
Agents earn a commission when a student enrols, so they ask several providers the same question and go with whoever answers first. Students do the same. A slow or wrong answer from an officer can cost INTO the student.
Who I designed for
The time-pressed Regional Officer or ESO. They handle many students and applications at once, especially in peak enrolment season.
12 students waitingPeak season
Regional Officer / ESO
Needs
A quick, correct answer to pass on. Fewer repetitive questions, so they can nurture and convert leads.
Today
Checks Excel sheets, shared documents and websites to confirm anything.
Frustrated by
Wrong or generic answers, a chatbot that asks before it helps, and dead ends.
Would trust
A solid first answer, honesty about what it doesn't know, and a link to the source.
05
What I found
Before jumping into conclusions, I looked at how a RO answers the same question by email. They answer the main question first, say clearly what's not possible, offer realistic alternatives and give everything in one go.
I audited the bot across three themes:
Program fit & eligibilityCounselling questionsFinance
The example below is from Program fit & eligibility.
The old botFour turns, no answer
AskINTO · MVP
Which biochemistry foundation programmes do we have with Combined CAS? And what IELTS score does the student need?
Sure! Which country is the student looking to study in?
1Combined CAS is UK only. It didn't need to ask.
UK
Great. What is the student's IELTS score?
2Asked for the very thing the agent was asking about.
These are the closest foundation options we do have, with the IELTS requirement for each.
C
If your student is set on biochemistry, I can check direct entry for you.
ASays clearly what's not possible
BOffers realistic alternatives and answers the IELTS question, all at once
CEnds with a next step
The usability sessions and weekly check-ins showed the real issue. It wasn't the look or the tone. The bot didn't understand the question, and people couldn't lean on the information behind it.
INTO term
“Pathway”
The bot gave a generic dictionary definition.
INTO term
“Foundation”
The bot confused foundation programmes with make-up.
Then I tested it with the people using it
Usability testing with 6 Regional Officers
To validate the audit, I interviewed six ROs and ran moderated tests with each of them. First I watched how they answer a query today. Then observed how they are using AskINTO to ask real questions.
How RO answers a query the old way
01Clarify the question with the agent
02Search Partner Portal website with filters
03Open the university cards
04Cross-check SharePoint, spreadsheets and documents
05Send an answer to the agent
Their mental model is visiting website, structured filters, a list of options, then narrowing down. AskINTO worked the other way round. It asked questions first and showed results last.
6 of 6 ROs
“Show me the options first. Then let me narrow it down.”
What I saw
The chatbot said it had found 50 courses but didn’t list them. It asked about budget, grades and preferred university before showing a single result.
So we decided
Answer first with a list. Offer refinement after, never before.
6 of 6 ROs
“If it’s wrong once, I go back to Partner Portal website.”
What I saw
It said 53 standalone Commerce programs when that count included double degrees. It showed Masters for a Bachelors query and the wrong subjects for Mechanical Engineering. Every RO re-checked answers before replying.
So we decided
Fix accuracy before adding features. Link every answer to its source so it can be checked in one click.
5 of 6 ROs
“Don’t ask me what I already told you.”
What I saw
It asked for the country twice, asked for the IELTS score that was the actual question, and showed a 9-field form after a recommendation link.
So we decided
Carry context through the chat. Ask only for what’s missing, in a short form.
5 of 6 ROs
“Speak our language.”
What I saw
It read “foundation” as a subject. It confused IGCSE with Nigerian requirements. ROs said qualification matters more than home country.
So we decided
Teach the chatbot about INTO terminologies. Ask about the student’s qualification, not just where they’re from.
4 of 6 ROs
“Budget is a guide, not a filter.”
What I saw
It dropped options outside the budget. When nothing fit £10,000, it didn’t say so or show what was close.
So we decided
Sort by fee instead of excluding. If nothing fits, say so and show the nearest options.
Fix the foundation
Entry requirements were the most common question for four of six ROs. So we started with program fit and eligibility: list-first answers, memory, INTO terms, and a source for every answer, country by country. That became the plan.
06
The plan
We built the plan around the student's journey, in four phases. Then came the country-specific requirements work. For each type of question, we mapped whether AskINTO could answer for the UK, US and Australia, where the answer would come from, and what was blocking it.
PHASE 1
Explore
PHASE 2
Apply
PHASE 3
Accept
PHASE 4
Arrival
Can it answer?
Checked for each type of question, separately for the UK, US and Australia.
Where from?
The source the answer would come from.
What's blocking it?
Where there was no reliable source, the chatbot shouldn't guess.
07
What I changed
Rules to talk by
Eight conversation design principles
01Understand why they're asking first.
02Answer part of it now, then ask.
03Remember what was said.
04Sound like a person who knows the job.
05Offer the next best option.
06Share in small pieces.
07End with a next step.
08Never guess.
Rewritten conversations
The biochemistry question, again
The chatbot now says early on that Biochemistry isn't offered under Combined CAS. It suggests close alternatives, gives the IELTS requirement in the same reply, and points to what to do next.
AskINTO
Which biochemistry foundation programmes do we have with Combined CAS? And what IELTS score does the student need?
The program search, as a tree
In a program search, the old chatbot said "50 courses" and showed none. Now it opens with five universities and four clear next steps. Each step is its own branch, and each branch ends with a next step. Pick a branch to see its reply.
Agent asksMy student from Nigeria wants to study bachelors in mechanical engineering in the UK. What would be the options for the student?
AskINTO answersI have found 50 courses related to mechanical engineering in the UK across 5 universities: Newcastle, Exeter, East Anglia, Queen's Belfast and Stirling. Would you like me to:
AskINTO · Branch C
Want to go deeper? Here's the full conversation flow, with my notes.
Conversational flow · design boardScroll or pinch to zoom · drag to move
Answers people can check · A way to push back
Built for trust
AskINTO pulls from SharePoint documents that ESOs maintain themselves. It links to the source and confirms what it understood before it answers. Try the feedback buttons.
AskINTO
MBA programmes we offer?
1
Just to check: you're looking for MBA programmes across all INTO partner universities. Is that right?
YesOnly in the UK
Here are the MBA programmes we offer, grouped by university.2
↗ Source: programme list · SharePoint
3
Was this helpful?
1
Confirm before answering
“Is this what you mean?” catches a misunderstanding before it turns into a wrong answer.
2
Link to the source
Officers can check an answer in one click instead of opening Excel. The documents are ones ESOs maintain themselves.
3
Push back, with a reason
Thumbs up or down with a reason, plus quick replies. Each week I grouped the feedback into themes and fed it into the next round.
What I couldn't change
AskINTO lives inside Salesforce, so I couldn't redesign the interface as I'd have liked. I worked on what I could control: the user experience, the order of information, the wording, the length of replies, the next steps and the sources.
08
What changed
After improving the MVP80%less time to answer an agent's questionCompared with the old way of checking Excel, shared docs and websites manually.Now ROs can spend that time on nurturing leads and converting them.
Two ways to get the same answer
Old wayExcel → shared docs → websites → reply
Excel sheets
Shared docs
Websites
Reply
With AskINTOAsk → check the linked source → reply
20%
80% saved
~0 → 36
daily active users
↓
Drop-off, a statistically significant decrease
↑
Completion: more conversations reach an answer
↑
Accuracy: answers are far more accurate
“It gave me a list of MBA programmes we offer. Pretty good.”
TammiRegional Officer
“Gave me the right answer, hoorah!”
SammiRegional Officer
“So seeing improvement already!!”
OliviaRegional Officer
09
What I took from it
01
Small gaps add up fast
One unnecessary question, one forgotten detail, one generic answer, and a person decides it's quicker to do it themselves.
02
Understanding beats more logic
The fix wasn't adding more logic. It was understanding what someone is trying to do, being honest about what's possible, and showing where the answer came from.
03
Beat the spreadsheet, not the RO
The bot was never going to beat a good RO. But it could stop being worse than a spreadsheet.
