Dymocks Tutoring and Talent 100 shut five Sydney centres and told parents to spend their money on Gemini or ChatGPT instead. Chief executive Mark Buckland was blunt about the maths: human tutors are still better, but parents were paying $800 to $900 per subject per term against $30 a month for a chatbot. “We think for the vast majority of users, AI is, for the price, the best compromise solution,” he said. A company voluntarily declared its own service obsolete because the price gap got too wide to defend.

If your business bills for repeatable human expertise, that same calculation is already running in your customers’ heads. This is what AI service disruption looks like from the inside, and what to do about your pricing and scope before someone else decides for you.

A Company Just Told Its Own Customers to Stop Paying It

Read the closure notice carefully and it says something no distressed business ever says. This wasn’t insolvency. It wasn’t a pivot into software, or a quiet wind-down blamed on rents and enrolment numbers. It was a functioning operation, launched in 2018 to make tutoring accessible, telling its paying customers to go spend the money elsewhere.

“Most of our customers would be better off using their hard-earned cash on cheaper solutions rather than simply paying for overpriced tutoring.”

Buckland said that to the Financial Review on the record. The service still worked. The price just stopped making sense next to the alternative.

That arithmetic is already running inside your operation, whether or not anyone has written it down. Quality documentation, deviation reports, supplier audits, first-article inspections. Ask what each one costs per unit of output, and then ask what the alternative costs.

The $900-a-Term vs $30-a-Month Math Behind the Shutdown

Run the annual numbers. Three subjects across four terms, at the top of that pricing band, lands a parent somewhere near $10,800 a year. A consumer AI subscription runs about $360. That is a 25 to 30x gap, and the company holding the premium end decided it could no longer defend it.

Why ‘better’ stopped being the deciding factor

Buckland never argued the AI was better. He argued the opposite, and then closed anyway:

Sure AI won’t perform as well as our human tutors with all the support behind them, but it’s also $30 a month.

That is the whole disruption pattern in one sentence. Quality superiority does not protect a price point once the cheaper option crosses the threshold of “good enough for what I actually need.” Buyers do not purchase the best available output. They purchase the cheapest output that clears their internal bar.

Which means the question for your service lines is not whether AI matches your specialists. It is how much quality your customer, or your internal stakeholder, will trade for a 90 percent cost reduction. Most will trade more than you expect.

Running the price-to-quality-delta calculation on your own service lines

Pick one repeatable output. A supplier audit report. A CAPA write-up. A monthly quality dashboard. A batch record review. Then get three numbers on paper.

• Fully loaded cost per unit of output: salary, overhead, review cycles, rework, and the delay cost of waiting in a queue.
• Cost per unit with AI doing the first pass: tooling plus the human time still required to check and sign off.
• Quality delta: measured, not assumed. Run both on the same 20 historical cases and count the defects each approach misses.

If the cost ratio is 10x or wider and the quality delta is small enough that a reviewer catches the difference in minutes, your premium is already gone. You just have not repriced yet. That gap is where AI service disruption starts, and it usually closes from the outside.

Where AI Actually Wins in This Story, and Where It Doesn’t

The repeatable middle is what gets automated first

Look at what students were actually buying and then stopped buying. Exam preparation. Immediate feedback on a practice answer. Unlimited repetition of the same question type at 11pm on a Sunday. Those four things are the repeatable middle of tutoring, and a chatbot does them without a booking calendar or a drive to Chatswood.

The same middle exists in every operations function. Drafting a deviation report, cross-referencing batch data against spec, producing a first-pass root cause list, answering “has this failure mode appeared before” in under a minute. None of that requires judgement. All of it eats hours from people you hired for judgement.

Automate that layer and you have not replaced a quality engineer. You have removed the part of their week that was never worth their salary in the first place.

What survives: accountability, context and consequence

Buckland was careful about this. He said human tutors were still better, and pointed at the reason: “all the support behind them.” Structured progression, someone noticing a student had quietly given up, a person answerable to the parent when results slipped. A chatbot has none of that. It has no stake in the outcome.

Manufacturing has a harder version of the same boundary. Signing off a CAPA, deciding whether to release a borderline lot, negotiating a corrective action with a supplier who is also your only supplier, standing in front of an auditor and defending a decision. Those tasks carry consequence, and consequence needs a name attached to it.

So the honest framing of AI service disruption is narrower than the headlines suggest. AI absorbed the tutoring hours that were transactional and left the accountable ones exposed as the real product. Work out which of your billed or budgeted hours fall on each side of that line. The transactional ones are already re-pricing, whether you have looked at them or not.

The Uncomfortable Backdrop: Falling PISA Scores and Cheap Tools

There is a detail in this story that should make anyone cheering the shutdown pause. Australia’s 2025 PISA results put teenage literacy and mathematics at their lowest level in two decades. Scores in both areas have fallen consistently since 2009, the year smartphone adoption in Australia passed half the population.

So the closure arrives at the exact moment the evidence says cheap, always-available technology and better outcomes are not the same thing. A generation got unlimited access to the most capable information tools ever built, and measured performance went down. That is not an argument against the tools. It is an argument against assuming access alone does the work.

Tool access without process redesign degrades outcomes

Handing a student a chatbot does not create a study method. Handing a quality team a copilot licence does not create a deviation process. In both cases the tool absorbs the effort that used to build capability, and nobody notices for two or three quarters because the output still looks fine on paper.

I have watched this play out as approvals that nobody actually reads, root cause narratives that are fluent and wrong, and CAPA records that pass an internal audit and fail an external one. The cost came out of the budget line and reappeared in rework, escapes, and customer complaints. Cost parity is not capability parity, and a $30 subscription does not carry the judgment that sat behind the person you removed.

Redesign means deciding what the tool drafts, what a qualified human verifies, and what evidence proves the check happened. It means measuring first-pass approval rate and escape rate before and after, not just headcount hours saved. It means deciding in advance which decisions are never delegated to a model, and writing that down.

Dymocks made a defensible call on price. What it could not do was redesign the learning process for parents on the way out the door. You can. That is the whole difference between AI service disruption happening to you and you running it deliberately.

What Operations and Quality Leaders Should Do Before the Market Decides for Them

Start with your purchase ledger, not your product. Every external service you buy that is billed per hour and delivered from a template is a candidate: validation documentation, supplier audit write-ups, training material development, translation of work instructions, second-tier compliance review. Then run the same test on what you sell, internally or externally.

The three-question audit for any hourly-billed service

Ask these out loud, in a room, with the invoice on the table:

• What percentage of this deliverable is repeatable?: If a competent tool with your templates and historical data gets 70% of the way there, you now know what you are actually buying.
• What are we paying for the other 30%?: Name it. Judgement on ambiguous cases, regulatory accountability, sign-off that carries liability. If nobody can name it, it does not exist.
• Can we prove it to a CFO?: Not assert it. Prove it with rework rates, audit findings avoided, or decisions the tool would have got wrong.

Most teams fail question two. Mark Buckland passed it and still closed, because he could name the premium (human tutors “with all the support behind them”) and concluded the price gap had outgrown the value. That is a harder, more honest answer than most procurement reviews produce.

Re-scope upward or re-price down, there is no third option

Re-scoping upward means the human hours move to exception handling, root cause judgement, supplier negotiation and accountability for the call. The repeatable middle goes to the tool, with a reviewer attached. Your headcount stays; your output per head changes, and so does what you can credibly charge or justify internally.

Re-pricing down means accepting the service is now a commodity and billing like it. Painful, but survivable. Doing neither is how a functioning operation ends up explaining itself to customers who already did the maths.

The ROI sits with whoever runs the audit first. Capture the margin yourself, or have your pricing publicly reset by someone with nothing to lose.

Source: afr.com