When a tutoring or training business outgrows per-student software

We have built two education platforms: PathMed, a GAMSAT preparation platform, and Skyhigh Academy, a tutoring management system. So this is a topic we sell into, and you should read the recommendation below with that in mind. It is still that most centres should stay where they are.

All prices quoted here are the vendors’ own published rates, checked on 28 August 2026 and linked so you can confirm them. They change. Several bill in US dollars; conversions below use approximately US$1 = A$1.39, which also moves.

The question is not what it costs. It is what the meter is attached to

There are four pricing metrics in this market, and they behave completely differently as you grow.

Only one of those has no upper bound. A percentage of revenue is the metric that gets genuinely uncomfortable, because the bill grows precisely when the business is working. Per-delivery pricing is next, and it is the one most tutoring centres are actually on.

The instruction that follows is simple and almost nobody does it: find the meter on your invoice, then project it to where you expect to be in three years — not where you are now. That single calculation resolves most of these decisions without any further analysis.

One detail on TutorCruncher’s Australian pricing worth reading carefully, because it changes the sum: the percentage applies to charges not paid through its integrated online payment methods. And on the Australian page, companies contract with Stripe directly and pay Stripe’s fees themselves — unlike its US pricing, which shows card processing at 3.85% and 3.5% through the vendor.

Per-student pricing charges by head, not by class

This is the structural detail that matters most to Australian centres, and it is stated plainly on Teachworks’ own pricing page: “Lessons are counted based on the number of students, so a lesson with 1 student is 1 student lesson, a lesson with 2 students is two student lessons.” Cancelled lessons are excluded.

Now put that against how Australian centres actually teach. Group tuition is common, and it is priced lower per head than one-to-one — published Australian rates commonly sit around $35 to $75 per student per hour for group work against $55 to $130 for one-to-one.

So a group-model centre earns less per head but is metered per head. As a share of revenue, it pays roughly double what a one-to-one centre pays for the same software.

Two honest qualifications. The absolute numbers are still small — this is a structural observation, not an alarm. And per-head metering is not unreasonable: it does track the load the system carries. But if you run groups, it is worth knowing that the pricing model is working against your delivery model, and worth checking against a per-seat product like TutorBird, where students are unlimited and you pay per staff member instead.

The break-even framework, run properly

Almost everyone does this sum wrong, in a way that flatters building.

The naive version: annual subscription divided into build cost. “We pay $6,000 a year, the build is $30,000, so it pays back in five years.”

That is wrong because it treats a custom system as though it costs nothing to run. It does not.

The running cost nobody includes

A custom system has a floor. On the stack we use, published rates as at August 2026 put Vercel Pro at US$20 per month per seat and Supabase Pro from US$25 per month. Free tiers exist but are not viable for a business system — Supabase pauses free projects after a week of inactivity. Call the infrastructure floor roughly A$750 a year before anything else.

Then maintenance. Australian agencies commonly suggest budgeting 10–20% of build cost per year for updates, security patches and small additions. I have not been able to trace that convention to a primary study, so treat it as a widely repeated rule of thumb rather than a measured figure — but treat it as something, because zero is definitely wrong. Most post-launch spend on software is adapting it to change, not fixing defects.

Add payment processing, which does not disappear when you build. And add your own staff time, which is the cost that never appears in anybody’s comparison.

Putting a build number in

We do not publish our prices, and this article is not a quote. For an order-of-magnitude anchor you can check yourself: the Australian agency Bocati publishes internal tool pricing at $10,000–$20,000 for two to three weeks of work, $20,000–$45,000 for four to six weeks, and $45,000–$75,000 for six to ten weeks. Substitute a real quote when you have one.

Now derive the threshold rather than asserting one. For a $20,000 build to pay back inside five years you need a net annual saving of about $4,000. Add the running cost back on — roughly $750 hosting, plus maintenance at 10–20% of build, so $2,000 to $4,000 — and the subscription spend you need to be replacing is somewhere around $6,750 to $8,750 a year.

That is the honest number, and it is lower than the one most suppliers imply. It is also still well above what a typical single-site centre spends. Note how sensitive it is to the maintenance assumption: at 10% the case opens up considerably, at 20% it nearly closes. Anyone quoting you a payback period without stating their maintenance assumption has not done the sum.

Three worked examples

Every input is shown so you can substitute your own. All vendor rates are published rates as at 28 August 2026.

Case 1 — a 150-student single-site centre

On Teachworks Growth: US$47.99 a month plus US$0.189 per student lesson. At 150 students averaging one lesson a week across 40 teaching weeks, that is 6,000 student lessons a year — about US$1,134 in usage plus US$576 in subscription. Call it A$2,000 to A$2,500 a year all in.

Verdict: never build. The entire annual subscription is a fraction of the running cost of a custom system, before you have paid for the build at all. The correct action here is to close this article and get on with teaching.

Case 2 — a 600-student multi-tutor centre running four separate tools

Scheduling and billing, a separate parent portal, a separate reporting tool, and a payments add-on. Individually modest; together they land near A$10,000 a year, plus the staff time spent re-entering data between them.

Against a build in the A$30,000 range with roughly A$5,000 a year to run and maintain, the annual saving is about A$5,000 and payback lands around six years. Verdict: marginal, and the subscriptions are not the real argument. If this case works, it works because of the manual re-entry between four systems, not because of the licence fees. Cost that re-entry properly and the decision changes; leave it out and building loses.

Case 3 — a training provider on per-registration pricing

High-volume providers metered per paid registration are the case where the meter genuinely runs away, because the fee scales with throughput rather than with the size of the business. At ten thousand paid registrations a year, per-registration fees plus administrator licences reach a level where a custom system becomes a serious question rather than a vanity one.

Verdict: worth modelling properly. With one large caveat in the next section.

Before you build anything, check these four things

In our experience most “our software costs too much” conversations end here, and the answer costs nothing.

1. Are you on the wrong plan?

Usage-priced products have a crossover point, and businesses routinely sit on the wrong side of it for years. On Teachworks’ published rates the crossover between Growth and Premium sits somewhere around 13,500 student lessons a year — above that, the higher monthly fee with the lower per-lesson rate is cheaper. Work out your own crossover before you conclude the product is expensive.

2. Can you bring your own payment account?

The difference between contracting with a payment processor directly and paying a platform’s marked-up rate is often larger than the software subscription itself. TutorCruncher’s Australian pricing has companies contracting with Stripe directly; its US pricing shows processing through the vendor at 3.85% and 3.5%. Find out which arrangement you are on.

3. Are you paying for add-ons you could drop?

Published add-on pricing stacks quickly. On TutorCruncher’s Australian page: priority chat support A$200/month, extra branches A$75 each per month, a custom domain A$150/month, plus per-minute charges for video calls, recording and storage. Some of that is worth it. Some of it was switched on once and never reviewed.

4. Are you comparing US prices against an Australian decision?

Several major products in this market bill in US dollars with no AUD option, so your real cost includes the exchange rate and your card’s foreign transaction fee, and it moves month to month. Also worth knowing: aXcelerate, the Australian RTO system, publishes no prices at all — its tiers are quoted on application, so a like-for-like comparison requires actually asking.

If you are an RTO, the answer leans harder towards buying

Registered training organisations have a reporting obligation that changes this calculation, and it is the strongest single argument in this article for buying rather than building.

RTOs must report Total VET Activity data to NCVER — a summary of the obligation rather than advice on your own reporting duties, which you should confirm with NCVER or your regulator. The sector is transitioning from AVETMISS to the VET Information Standard under the VET Data Streamlining program, with the new standard in place from 1 January 2029.

Here is what that means commercially. On vendor software, that migration is absorbed by the provider. It is in the subscription you are already paying. On a system you own, it is a project you scope, fund, build and test — on a deadline set by someone else, for no competitive benefit whatsoever.

NCVER does maintain a register of AVETMISS-compliant software, and it is worth being precise about what that is: entries are supplied by the vendors themselves, NCVER states it cannot verify them, the register is expressly not an endorsement or a guarantee of quality, and compliant systems exist that are not listed. So it is not a gate you have to pass — but it is a shortlist of products whose vendors have committed to the standard, and a system you build has no equivalent.

None of this makes building impossible for an RTO. It does mean the compliance surface should be priced into the decision at the start, and that a provider on per-registration pricing should weigh a genuinely expensive meter against a genuinely expensive obligation.

When the answer is genuinely to build

Cost savings are the weakest reason to commission software, and the one most often used to justify it. Here are the reasons that actually hold.

A capability the product cannot deliver at any price

A placement or matching process specific to how you teach. An assessment model nobody else runs. A reporting obligation to a funder that no vendor supports. A workflow your business genuinely competes on.

The test is concrete: name the specific thing, then put a rough hourly figure on the workaround, per month. If you cannot do both, you do not have a case yet. “It does not quite fit how we work” is not a case; “two staff spend six hours a week reconciling the timetable against the invoices because the product cannot model our group structure” is.

Three or more products held together by manual re-entry

The second legitimate trigger, and the more common one. When the same student record is typed into three systems, the cost is the re-entry and the errors, not the subscriptions. That cost is real, it is usually invisible in the budget, and it is the thing consolidation actually fixes.

Note that this argument does not depend on the software being expensive. It works even if every subscription involved is cheap — which is exactly why the subscription-based break-even sum gets it wrong.

What we have actually seen building in this sector

Two builds, and the surprising part was which bits were hard.

On PathMed, a GAMSAT preparation platform, the parts people assume are difficult turned out to be commodity: scheduling, dashboards, question banks. Those are solved problems and there is no advantage in rebuilding them. The expensive part was the separation between student, tutor and administrator views — three genuinely different pictures of the same underlying data, each of which must not leak into the others.

The part that justified building rather than buying was the study prioritisation: looking at time spent per question and patterns in repeated mistakes to decide what a student should work on next, section by section. That is not a feature you can buy, because it encodes a particular view of how people should prepare.

On Skyhigh Academy, tutoring management software, the same pattern held. The operations side — scheduling, students, tutors, invoicing — is well served by products. What was worth building was the fit between the business’s own process and the student and parent portal sitting on top of it.

The general lesson, which argues against us more often than for us: the commodity parts are commodity, and you should buy them if they are most of what you need. Building earns its place where the thing you do differently lives.

One more observation. Both platforms handle student data, in a sector where students are often under 18. That raises the standard for access control, for who can see what, and for how carefully you treat information about children — and it is a real cost of ownership that a vendor otherwise carries for you.

Where the Privacy Act sits in this decision

General information, not legal advice — the OAIC and the Act are linked so you can check it, and you should get your own advice.

Many small Australian tutoring businesses sit outside the Privacy Act entirely. Under section 6D(1), a business with annual turnover of $3,000,000 or less in the previous financial year is generally a small business operator, and is excluded from the definition of “organisation”.

Three things worth knowing, stated neutrally:

• It is a rolling test, applied each year against the previous year’s turnover — not a one-way door. A business that goes over and later comes back under can qualify again. Several Australian pages say otherwise; the Act does not.
• The carve-outs are easy to trip: providing a health service and holding health information, disclosing personal information for a benefit, providing a benefit to collect personal information, or being a Commonwealth contracted service provider.
• A Children’s Online Privacy Code is required under section 26GC to be registered by 10 December 2026. If your platform is used by people under 18, that is worth watching.

There is a structural point buried in this that is worth saying out loud. The revenue at which a percentage-of-revenue software bill gets big enough to argue about is roughly the same revenue at which the Privacy Act starts applying to you. Businesses tend to hit both at once, and tend to plan for neither.

Common questions