AI for education providers

AI for education providers

Instructors teach, the admin runs itself, every learner gets a tutor's attention. 5x ROI in 30 days, or we work free.

  • Private colleges
  • Tutoring networks
  • Corporate training
  • Continuing education
  • Language schools

Teams we build for

  • Hoyes Michalos
  • Nurse Next Door
  • Fedi
  • UBC Sauder
  • Merchant House Capital
  • Picton Investments
  • Campbell Froh May & Rice LLP
  • Barnakl
  • Hungerford
  • Breez

Education providers, before and after

FIG. 01

The manual path is dashed: Grading stacks up nightly, Inquiries answered Monday, Content updated by hand. The system path replaces it, and a person approves before anything ships: First-pass marks to finalize, Every inquiry answered fast, Materials refresh themselves.

Before: by hand

  1. 01Grading stacks up nightlyhuman
  2. 02Inquiries answered Mondayhuman
  3. 03Content updated by handhuman

After: the system

  1. 01First-pass marks to finalize
  2. 02Every inquiry answered fast
  3. 03Your approvalhuman
  4. 04Materials refresh themselves
Education providers, before and after.REV 2026.08

The research

Full-time lower secondary teachers average 41 hours a week, 15 of them on lesson prep, marking and admin rather than teaching (OECD TALIS 2024, OECD average, self-reported).

OECD · 2025 · About 280,000 lower secondary teachers and school leaders across 55 education systems, fielded 2024, results published October 7, 2025

Proof, honestly

No published education case study yet, and we say so. Founding clients get priority scheduling, direct founder involvement, and a co-published case study once the numbers are real. The 5x ROI guarantee applies from day one.

Book a call to claim the founding slot

We guarantee 5x ROI inside 30 days of deployment, in writing, measured against a baseline you sign before we build. If the system misses the bar, we keep working for free until it clears.

01

Instructors burn out on everything but teaching.

The paid hour is the classroom. The unpaid ones are marking, lesson prep, differentiation and the parent inbox, stacked on top of it. The people quitting are not tired of students. They are tired of everything else.

02

Enrollment inquiries go cold.

Prospective students ask at 9pm. The reply comes Monday. They enrolled somewhere else Friday.

03

Course content costs too much to keep current.

Every cohort, client and delivery format forks the materials, and the master copy stops being the master by week three. Updating one module means finding every place it was ever pasted.

04

Feedback returns after the next assignment is due.

Marking at scale means feedback lands two assignments late, when it can no longer change how the student works. The instructor is not slow. The pile is just bigger than the evening.

05

The withdrawal was visible in week two.

The student who stopped logging in after the second module withdrew six weeks later. The LMS held the signal the whole time. Watching it was nobody's job.

06

Applications stall half-complete.

The transcript is in, the English score is not, the deposit is a maybe. A seat is only real when the file is complete, and the chase belongs to whoever remembers. Nobody remembers in intake season.

Run your numbers.

Your operations

7
8
$50

Savings use the low end of our hours-reclaimed range. The math is conservative on purpose.

The math

Cost of manual work / yr$134,400
Recovered / yr$33,600 - $67,200
Hours back / yr672+
Hours back / wk14+

Calculated at the low end of every range.

Every first build is covered in writing: 5x ROI in 30 days. Or we work for free.

The economics

Before and after economics

Line

Marking turnaround

Before

Feedback returns days late, after the next assignment is already due

After

First-pass drafts wait for the instructor, feedback goes back the same week, qualitative by design

Admin hours in scope

Before

At the page defaults, 7 staff spending 8 hours a week on admin is 56 hours, $2,800 a week at $50 loaded cost

After

The calculator above prices your own roster, and the 5x ROI guarantee is measured against a baseline you sign before we build

The 9pm inquiry

Before

Answered Monday, enrolled somewhere else Friday

After

Answered the same evening, the application chased to complete

Course refresh per cohort

Before

Materials forked by hand for every cohort, client and format

After

Updates drafted from the master course, an instructor approves the fork

No education client outcome numbers exist yet, and this table does not pretend otherwise. The only figures in it are arithmetic on this page's calculator defaults, 7 staff at 8 manual hours each per week at a $50 loaded hourly cost, which is 56 hours or $2,800 a week in scope. Every other row is a qualitative before and after state. The founding cohort's measured outcomes get co-published when they are real, which is what the founding offer above commits to.

The numbers in education providers

Full-time lower secondary teachers average 41 hours a week, 15 of them on lesson prep, marking and admin rather than teaching (OECD TALIS 2024, OECD average, self-reported).

OECD · 2025 · About 280,000 lower secondary teachers and school leaders across 55 education systems, fielded 2024, results published October 7, 2025

Teachers who use AI weekly save an average of 5.9 hours per week, about six weeks per school year (Gallup, 2025).

Gallup (via Engageli) · 2025

94% of UK undergraduates have used generative AI to help with assessed work, which includes explaining a concept or summarising a reading. 12% say they submit AI-generated text directly (HEPI, 2026, UK undergraduates).

Higher Education Policy Institute and Kortext · 2026 · 1,054 full-time UK undergraduates, fielded by Savanta in December 2025, published March 12, 2026

Who this is built for

College director

The start date arrives with the cohort full.

Tuition is the whole revenue line, and tuition follows completed files, not inquiries. The enrolment agent answers the evening inquiry, chases the transcript and the English score, and shows you which files stalled and why. You watch a pipeline instead of forwarding emails to admissions. The seat count on day one is the number the whole term runs on.

Lead instructor

Feedback lands while the work is still warm.

First-pass marking is drafted against your rubric, each comment pointing at the passage in the student's own work. You override where judgment beats the rubric and release the rest. Feedback goes back the same week, when it can still change how the student works, and every grade in the gradebook carries your name.

Admissions coordinator

No applicant waits for office hours.

The 9pm inquiry gets answered at 9pm, in the applicant's own terms, from the program pages you actually publish. Incomplete applications get a chase instead of a checklist. Your morning starts with a list of files that moved overnight and the three that need a human call, not with an inbox of duplicates.

Training program manager

One master course, every client version current.

The master curriculum lives in one place. Client cohorts, rebrands and format conversions are drafted from it, and you approve the fork instead of rebuilding it. When the software version changes in module four, the update propagates as drafts, not as a fortnight of copy-paste across nine client decks.

What stays human

  • The 8am block is just teaching
  • Grades leave under her name
  • She picks the interventions

The steps the day below leaves to a person, by design.

A lead instructor's Thursday

01

The 8am block is just teaching.

human

She teaches. No dashboard, no queue, no tabs. The system's whole job before noon is to make the classroom the only place her morning goes.

02

Draft feedback is waiting at the break.

Last night's submissions carry first-pass marks against her rubric, every comment pointing at the line in the student's work it refers to. Nothing has a grade yet. Drafts are proposals, and the gradebook does not know they exist.

03

Grades leave under her name.

human

Over lunch she reviews the drafts, overrides two marks where the reasoning was better than the rubric anticipated, and releases the rest. The gradebook writes on her click and never otherwise.

04

The course tutor holds the afternoon.

Students in the practice lab ask it questions. It scaffolds from her own materials, refuses to hand over finished answers, and flags the one question that fell outside the syllabus for her Friday session. Every exchange is logged to the integrity file beside the Turnitin reports.

05

The quiet students get loud.

Three students stalled on the same module surface on her list with their attempts and their last login. The signal sat in the LMS all term. Now it arrives before the withdrawal deadline instead of in the withdrawal form.

06

She picks the interventions.

human

Two get a seat in tomorrow's lab hour. One gets a call from the program head. The system flags, she decides. Nothing about a student's standing moves without her.

07

Friday's materials fork from the master.

The updated lab manual drafts itself out of the master course with this cohort's schedule and the new software version, and parks for her morning look. She will read it with coffee, not build it at midnight.

What we build for education providers

The teaching load

01

Tutoring and practice support.

Course-grounded practice help that scaffolds instead of answering for the learner, with instructor oversight built in.

02

Grading and feedback copilots.

First-pass marking against your rubric with draft feedback. The instructor finalizes every grade.

03

Lesson and curriculum prep.

Lesson plans, differentiated materials and assessments drafted from your curriculum, not the open internet.

The front office

04

Enrollment funnel automation.

Inquiries answered fast at any hour, applications chased to complete, no prospect left waiting on Monday.

05

Parent and student communications.

Progress updates and program communications assembled and sent on schedule, consent-tracked.

06

Attendance and completion reporting.

Enrolment status, attendance and completions assembled from the SIS into the reports the registrar and the regulator need. A person reviews and files. Nothing auto-submits.

The catalog

07

Course content production engine.

Materials updated, reformatted and versioned across cohorts and delivery formats, from the classroom workbook to the SCORM package a client loads into their Docebo.

Where the data comes from

SIS

The student information system holds enrolment, attendance, contracts, grades and transcripts, whether that is Populi, Orbund or the spreadsheet the registrar keeps beside the official one. The enrolment agent reads inquiry and application state from it, the chase runs off incomplete files, and attendance and completion reporting assembles from its records.

Where it stops. Nothing writes to a student record on its own. Enrolment changes, grades and anything that touches a transcript leave the system as proposals a person releases, and the record itself stays in your accounts.

LMS

Moodle, Canvas or Brightspace holds the course shells, submission queues, quiz banks and completion data. First-pass marking reads the submission queue, the course tutor runs beside the shell, and the stalled-student signal comes from the logins and submission gaps the LMS already records and nobody watches.

Where it stops. The gradebook writes on an instructor's click and never otherwise. Draft marks live outside the gradebook until released, and tutor transcripts are logged to the integrity file, visible to the instructor who owns the course.

Course materials

The curriculum itself. Syllabi, slide decks, lab manuals, workbooks, rubrics, past exams and the SCORM packages corporate clients receive. This is the corpus the tutor and the lesson-prep agent answer from, and when a question falls outside it the system says so instead of improvising an answer the school never taught.

Where it stops. Your IP stays yours. The index lives in your infrastructure under no-training API terms, tutor access is scoped to enrolled students, and licensed third-party content is indexed only as far as its license permits.

FIG. 02

How an answer is found in your curriculum

A question runs against your curriculum two ways at once: vector search for meaning and full-text search for exact wording. Reciprocal rank fusion merges both result sets, and the answer carries the source it came from.

  1. 01A question
  2. 02Vector search (meaning)
  3. 03Full-text search (exact wording)
  4. 04Rank fusion
  5. 05Answer, with its source
How an answer is found in your curriculum.REV 2026.08

Built around your rules

The regimes that govern education providers, what each demands, and how the system complies

Regime

PIPEDA and BC PIPA

What it demands here

Learner and family personal information collected with consent, used for the stated purpose, and protected wherever it flows.

How the system complies

Minimal collection by design, consent tracked in the data layer, and records stay in your accounts under no-training API terms. What the workflow needs is documented, and nothing beyond it is collected.

FIPPA

What it demands here

Where the provider is a BC public body. Public-sector privacy and residency duties attach to the institution, and the privacy office signs before systems touch student data.

How the system complies

Builds for public bodies are designed around FIPPA from the start, data flows are mapped in writing before anything is built, and your privacy office is in the loop from the first session.

Minors' data

What it demands here

Where learners are under 18, explicit consent flows and minimal collection are duties, not preferences.

How the system complies

Consent flows are explicit, collection is minimal, access is scoped per cohort and per role at the database layer, and nothing about a minor leaves your accounts.

CASL

What it demands here

Enrollment marketing is sent with consent, identifies the sender, and can be left with one click.

How the system complies

The enrolment agent works a consent-tracked list, every message identifies the school, and unsubscribes propagate to every sequence the same day.

Private Training Act, BC

What it demands here

Where the college offers regulated career training. Certification with the Private Training Institutions Regulatory Unit, student records kept and producible, enrolment contracts and tuition refunds handled to the standard.

How the system complies

Enrolment steps are logged in the SIS where the regulated records already live, refund-relevant dates are tracked rather than remembered, and nothing in the enrolment path is invisible to an auditor.

IRCC compliance reporting

What it demands here

Where the school hosts study-permit students as a designated learning institution. Enrolment status reported accurately and on the regulator's schedule.

How the system complies

The report assembles from SIS enrolment and attendance records into a draft the registrar reviews and files. Nothing is submitted to IRCC by an agent.

Learner data is handled to respect PIPEDA, BC PIPA and public-sector rules where they apply, with minors' data held to minimal-collection design.

Read our full security posture

The objections

Our instructors are contractors paid by the class hour. They will not adopt new software.

They are not asked to. Draft feedback lands inside the course shell they already mark in, and the tutor runs beside the course, not in a new app. Your tutors keep living in their Teachworks schedules and group chats. Adoption follows one rule in every build we run: usage climbs when the system removes unpaid work, not when people are trained into a new tool. For hourly contractors, marking drafts are the first unpaid hours to disappear, which is why they are the wedge.

A student will claim an AI marked them unfairly, and then what?

Then you have a better file than you have today. Every draft comment cites the rubric line and the passage in the student's own submission, the instructor who released the grade is named on it, and marker agreement against your own instructors' past marking is measured continuously in the eval harness. An appeal meets a documented human decision with its reasoning attached. That is easier to defend than a tired instructor's 11pm margin note.

Our curriculum is our IP. Indexing it feels like packaging it for a departing tutor to take.

The index never leaves your accounts. It lives in your infrastructure, access is scoped per role and per cohort at the database layer, and the API terms are no-training. A departing tutor could always photocopy a binder. What they cannot do is take the index, because it was never on their laptop.

Half our catalog is licensed courseware. We do not own it.

Then half your catalog is scoped, not indexed. We map ownership publisher by publisher before indexing starts, index what you own, and keep licensed content inside the terms of its license. Where a license does not permit reproduction, the tutor points the student at the licensed page instead of reciting it. That scoping document is written before the build, and you keep it.

Our accreditor and our regulator will have questions.

They will, and the answers are the system's exhaust. Every tutor exchange is logged, every grade carries a named instructor and a rubric trail, every enrolment step is recorded in the SIS where your records already live, and nothing files itself to anyone. An auditor's first request is show me the records. Here the records are the mechanism, not a reconstruction.

How we build it for education providers

Step 01

Ground everything in your curriculum.

Tutoring and feedback are indexed on your own materials and marking standards. Outside the syllabus, the system says so rather than guessing.

Step 02

Grade the grader against your instructors.

Feedback quality is measured as agreement with the standard your markers already apply, not against a generic rubric. Drift here shows up as unfair marking.

Step 03

Automate the chase, never the assessment.

Enrolment steps and routine questions are automated first. Anything reaching a learner's record stays a proposal an instructor releases.

What we will not automate

Grades and assessment decisions. Learner records are personal information, often for minors, so nothing is written without an instructor releasing it.

How the system is built for education providers

See the full capability map

Retrieval

Your curriculum, your materials and your marking standards indexed, so tutoring and feedback are grounded in what you actually teach. When a question falls outside the syllabus, the system says so instead of guessing.

  • pgvector
  • Full-text BM25
  • Abstention on low confidence

Agents and orchestration

Agents draft feedback, chase enrolment steps and handle routine learner questions on a schedule. Grades and anything on a learner's record are proposals an instructor releases. No autonomous assessment.

  • agent-worker
  • Autonomy guard
  • Proposal queue

Evaluation

Feedback quality is graded against work your instructors have already marked, checking agreement with their standard rather than a generic rubric. Drift here shows up as unfair marking, so it is measured continuously.

  • Eval graders
  • Marker agreement rate
  • quality-worker

Models

Frontier models for tutoring and feedback grounded in your materials. Embedding models for curriculum retrieval. Where a model is tuned to a house marking standard, that is fine-tuning on your data and it stays yours.

  • Frontier models, one gateway, routed per task
  • Embedding models via the same gateway
  • Fine-tuning on your rubric

Data boundary

Learner records are personal information, often for minors. They stay in your accounts, scoped per cohort and per role, with consent tracked in the data layer, under no-training API terms.

  • Supabase row-level security
  • Consent ledger
  • Per-cohort scoping
FIG. 03

SIS, LMS, Course materials feed a hybrid index. The agent runtime works from that index, and every consequential action passes a human approval before it reaches Ask your brain, Ask your team, Workflow builder.

Your systems

  1. 01SIS
  2. 02LMS
  3. 03Course materials

The system

  1. 01Hybrid index
  2. 02Agent runtime
  3. 03Your approvalhuman

Where your team works

  1. 01Ask your brain
  2. 02Ask your team
  3. 03Workflow builder
Education providers: how the system fits together.REV 2026.08

What that means in practice

Retrieval and RAG

Connecting AI to your actual documents so it answers from your knowledge, accurately and with citations, instead of making things up.

Where we stop. A dedicated vector database is justified by scale, not by default. Most of RAG quality is won or lost in chunking and indexing strategy, not in the model choice, and we have walked clients back from RAG to plain search when that was the honest answer.

How we use it

Agents and orchestration

AI that does the work instead of just answering: looks things up, calls your systems, completes multi-step tasks, and knows when to hand off to a human.

Where we stop. Multi-agent swarms are oversold; most jobs need one well-guarded loop. If a cron job and a script solve it, that is what we build, because 90 percent per-step accuracy compounds to 59 percent over five chained steps and no framework changes that arithmetic.

How we use it

Fine-tuning and customization

Teaching an existing model your domain's style, vocabulary or task. Far cheaper than building one from scratch.

Where we stop. Fine-tuning is a cost, latency and consistency optimization, not a knowledge store. If you want the model to know your documents, that is retrieval, not training, and we will tell you which one you actually need before you pay for either.

How we use it

Where your team works

Tour the platform

Ask your brain

Ask a question. Get the answer and where it came from.

Every answer cites the document behind it, so you can check the source yourself. When the brain does not have the answer, it says so instead of guessing.

Ask your team

Chat with a team that already knows your business.

Every agent is briefed on your documents, your data, and your preferences. Ask for the number, the draft, or the plan and cite where it came from.

Workflow builder

Describe the workflow. Watch it assemble.

Say what should happen in plain language and the builder assembles the automation on a canvas you can read, run, and change.

Approval queue

The AI proposes. You approve. Nothing sends itself.

Every consequential action arrives as a proposal with the risk, reversibility, and expiry spelled out before you say yes. Control stays in the room.

The hard questions

Free · 3-5 days

Know your number in five days.

We map your operations, find the highest-ROI automations, and hand you a ranked plan with the payback math attached. Yours to keep, whoever builds it.

Prefer to talk first? Book 15 minutes with James. No pitch deck.

Free · no obligation · five minutes

The plan is yours to keep, whoever builds it.

5x ROI in 30 days. Or we work for free.