A Tuesday in November. Maya, 17, a high school senior.
She has just found your college: a story, a friend, a search. She is going through the program pages on her phone.
Two programs interest her. She hesitates. She has one question, only one: “Is the co-op paid, and does it start sophomore year or only junior year?”
The answer exists, somewhere in a forty-page viewbook. The request-info form asks for twelve fields. The office opens in eleven hours.
Maya closes the tab. Tomorrow she will look at the college across town, and your team will never know she came.
Maya is a character. The visitors who do exactly what she does, every evening, are not: this document counts them.
It all starts with one number
While the admissions office is closed, your future students compare, hesitate and decide. Everything else in this document follows from that gap.
Every visitor you lose is a visitor
you have already paid for.
Acquisition has never
cost this much
Colleges have never spent more to bring visitors in. Higher education invested more than $2.4 billion in advertising in the first half of 2025 alone.2 And every click bought returns less: in 2025, the cost of an education lead rose faster than in any other industry.1
2024
2025
2025
So every visitor costs more than last year, and a click on its own does not enroll: between the paid visit and the completed form, everything happens on the website. That is exactly where the loss is least measured, and the most expensive.
Every visitor to your website has been paid for.
The question is no longer how to bring in more:
it is how to stop letting them leave without asking.
The response window
lasts five minutes
Research on lead conversion has been unambiguous for fifteen years: response speed is the first factor in qualification, far ahead of what the response actually says.
rather than within 30. And 100× more likely to reach them.
Foundational study on 15,000 leads and 100,000 calls.3
What the student expects
What institutions deliver
The problem is not specific to education (across all sectors, the average response time to a web lead is 42 hours and 23% of companies never reply4), but education pays for it more dearly than anyone, at the lead price seen in chapter 01. Nobody will hold a five-minute window on a Sunday at 11 p.m. An assistant will: it holds the first minute, and your team takes over the next morning on a lead that is still warm.
Between what the student expects and what institutions deliver,
the gap is measured in hours. Conversion is decided in minutes.
Your applicants write to you
after the office closes
Across more than 25,000 interactions analyzed in 350 institutions, three quarters of the questions put to university assistants arrive outside business hours.11
The stealth applicant changes what a website is: for one prospective student in two, it is now the only point of contact with your college before the application.10 They compare, research and decide without ever calling or writing. If the website does not answer their questions at the moment they ask them (in the evening, at night, from another time zone), nobody will.
The admissions team’s day ends at 5 p.m.
The student’s begins after class, after work,
or on the other side of the world.
What changes when an assistant
actually answers
The effect of a conversation on conversion is one of the best documented results in digital marketing. What has changed: the assistant holding it has nothing in common with yesterday’s scripted chatbots.
Those three figures measure the same mechanism: a conversation lowers the cost of entry. Asking a question in a chat commits you to less than a twelve-field form, and once the answer is there, the form follows naturally. Chat does not replace the form: it triggers it. That is the exact opposite of a brochure website, where the form is the only door, and every unanswered question is a reason to close it. The next chapter shows this mechanism measured in real conditions, on private college websites in Europe, with a current-generation AI assistant.
The scripted chatbots of the last decade frustrated more than they helped, and many institutions still remember them that way. Today’s AI assistants understand the question and answer from your own content. And the shift goes well beyond education:
Half a million visitors, four campuses,
three months
A French private higher education group (four institutions, not named here) deployed a conversational AI assistant across all of its websites. Real figures, cumulative over less than three months, rounded to preserve anonymity.15
on mobile
abroad
business hours*
on the weekend*
Over the window analyzed, the group’s annual growth in online leads went from +52% to +71% after the assistants were switched on: around 500 additional leads attributable to the channel.15
* Measured on conversations that generated a lead. ** More than one theme is possible per conversation. † Conversations with more than one exchange; half of all conversations stop at the first message. Before / after attribution partial (internal analysis).
Fifteen questions in one evening
This is what the conversations in the deployment actually look like: the same questions your team receives, asked at the hour when nobody can answer them.*
Every unanswered question is a tab that closes. Every answer is a potential lead, and one less email to deal with on Monday morning.
* Typical wordings, faithfully reconstructed from the themes actually measured in chapter 05 (programs, admissions, tuition, international, start dates, work placements, scholarships). No conversation is quoted verbatim, out of respect for anonymity.
Three minutes, one lead
The same mechanism, seen up close: a reconstructed conversation, representative of the ones in the deployment.*
The 10:48 p.m. question is not waiting for Monday. It is waiting forty seconds.
* Reconstructed conversation, representative of the exchanges in the chapter 05 deployment; no verbatim quotes, figures for illustration.
What it returns,
what it costs
Apply the rates of an optimized deployment, where the assistant is given prominence, to a website with 100,000 visitors a year (the size of one institution, not a system).
The stance is deliberately cautious: only 6% lead → enrollment conversion, revenue limited to tuition, a full first-year cost. Halve the open rate again: the return stays above ×25. And converting those leads is still your admissions team’s job: the five-minute window from chapter 02 finally works in its favor.
funds the assistant for three years.
Assumptions: tuition $20,000 / year, 4-year degree; open and exchange rates = vendor benchmark, widget given prominence; engagement and leads = rates observed in chapter 05; lead → enrollment 6% (internal benchmark); first-year cost = US pilot rate, indicative.
A campus chatbot is not
a generic chatbot
The results in chapter 05 do not come from a widget dropped onto a website. Ten requirements make the difference.
Trained on your content
Programs, tuition, calendar, admissions process. No generic answers.
Qualification built in
Every lead arrives scored: hot, warm, cold. Never in bulk.
Spam filtered before the CRM
On the chapter 05 deployment, 28% of raw submissions were spam. Your team sees none of them.
Multilingual
64% of the leads observed come from abroad. The assistant answers in their language.
Mobile first
79% of conversations come from a phone. The experience is built for that.
Available 24/7
Because 75% of questions arrive after the office closes.11
Wired into your CRM
Leads land where your team already works, in real time.
Measurable end to end
The full funnel, from visitor to enrollment, campus by campus.
A controlled scope
The assistant answers on your institution only, and hands over to your team when it should.
No website rebuild
One script to add, no technical dependency. Live in a few days.
Those ten requirements are the brief. The chapter 05 figures are what it produces once the brief is met.
What Skolbot does,
exactly
Six links in a chain, no change for your team, other than qualified leads arriving where it already works.
Live in a few days. One script to add, no website rebuild.
Another Tuesday, another Maya. The same question, at the same hour: “Is the co-op paid, and does it start sophomore year or only junior year?”
This time, an answer. In forty seconds, from your own content: the schedule, the year it starts, and whether it is paid.
Three questions later, she downloads the brochure and leaves her email address.
Monday, 9:02 a.m.: your team calls back a qualified lead, still warm, instead of sorting through empty forms.
See it answer on your own content
The demonstration is built in a few days from your public website.
Your programs, your tuition, your dates: judge it on the evidence.
Where these figures come from
The market data quoted in this document comes from the public sources listed below, and is US data unless stated otherwise; amounts are shown in dollars as published, rounded. The figures in chapters 05 and 06 come from a live Skolbot deployment with a private higher education group in France (2026): they are anonymized and rounded to protect the institution’s confidentiality, and internal measurements not validated by a third party are flagged as such. That deployment is used because it is the one measured end to end; the behavior it documents (evening traffic, mobile, stealth applicants) is the same behavior the US sources describe. The “typical questions” and the reconstructed conversation in chapter 05 are faithful to the themes measured, never verbatim quotes.