Picture driving a bakkie down the N1 with R100 worth of bricks on the back, and the toll gate charges you 1%, so you hand over a rand and carry on. The next day you take the same bakkie down the same road, but this time you’re carrying R100,000 worth of TVs, and the toll gate wants R1,000. The road is the same, the wear on the tar is the same and the guy in the booth is doing exactly the same job, yet your bill has gone up a thousand times because of what happened to be in the back.
You’d lose your mind if a real toll road did that, but a big chunk of the software industry prices itself exactly this way and most of us have just accepted it. Let’s unpack this.
“But we sit in your revenue path!”
The pitch normally goes something like “our software sits between you and your money, so we’re entitled to a slice of it.” Payment providers are the obvious example, where you pay a percentage on a R100 transaction and the same percentage on a R100,000 one.
Look, in payments there are real reasons for some of that. Bigger transactions carry more fraud exposure and more chargeback risk, and the banks, card networks and acquirers charge the payment providers percentage-based fees of their own, so I can live with it there.
The problem is that the same thinking has crept into almost everything else. Subscription billing platforms charge according to how much you bill, marketplaces and app stores take a cut of everything that flows through them, and plenty of e-commerce platforms want a slice of what you sell. The Gringotts goblins would be proud. In most of these cases the software is doing pretty much the same work whether the sale is R100 or R100,000. The database insert didn’t get a thousand times harder, the API call didn’t suddenly eat a thousand times more CPU, and nobody wrote a thousand times more code to produce the invoice.
Shopify, the toll booth with two lanes
If I had to pick one company to explain this whole article, it would be Shopify, especially if you’re selling from South Africa.
First you pay a monthly subscription for your plan, and honestly that part is fair, because hosting your store and keeping a platform that size running costs real money every month. Then you need a way to take payments, and this is where it gets interesting. Shopify’s own in-house option, Shopify Payments, isn’t available in South Africa, so every local merchant has to plug in a third-party gateway like Payfast, Yoco, Peach Payments or Ozow, and that gateway takes its percentage on every sale.
And then Shopify takes its own percentage on top of that, 2% on the Basic plan, as a fee for using a payment provider that isn’t theirs, in a country where they don’t offer one. The fee does drop on the pricier plans, which is a neat way of telling you that you can pay less tax on your sales by paying more rent.
Let’s put that on the bakkie. Say your store does R100,000 in sales a month on the Basic plan. Shopify collects R2,000 of that, on top of your subscription and on top of whatever your gateway charges you. The next month you have a great run and do R200,000, and Shopify collects R4,000. Their servers did exactly the same work for both months. They also aren’t the ones processing your payment, because your gateway does that, and your gateway is the one carrying the fraud risk and dealing with the chargebacks, which is precisely what you’re already paying it for.
So you end up paying three times: rent for the store, the gateway for moving the money, and Shopify for the privilege of letting somebody else move it. That’s a toll booth charging you by the cargo on a road where somebody else has already collected the toll.
To be fair, Shopify is a brilliant product with a huge app ecosystem, great themes and a polished checkout, and plenty of merchants happily pay for that. My issue is much narrower. That 2% has nothing to do with the work Shopify does on your sale and everything to do with where Shopify happens to be standing. Sometimes there’s a proper reason for a bill that grows with your sales, like financial risk the company carries or infrastructure that genuinely scales with value. Here, if we’re honest, the answer is “because we can.”
That kind of pricing worked beautifully while software was hard and expensive to build, and I don’t think it’s going to hold up nearly as well now that AI is in the picture.
Adobe, you might want to sit down
Software engineer Brandon Thomas recently used Claude and Rust to build open-source alternatives to a large part of Adobe Creative Cloud, with apps aimed at Photoshop, Lightroom, Illustrator, Premiere, Acrobat, After Effects and InDesign. The project is called ArtCraft, and I took some of it for a spin.
Truth be told, it isn’t Adobe. It’s slower, rough around the edges, missing features and has some performance issues, and for serious work I’d still pick Adobe today. Other people who have tested it have said much the same thing.
What caught my attention is that somebody actually did it. A relatively small effort produced recognisable, usable versions of software that historically needed huge engineering teams and years of development. The interesting question isn’t whether PhotoCraft can beat Photoshop today, it’s what PhotoCraft looks like after another year of work, or after ten experienced engineers spend a year turning something like it into a proper commercial product. Or what happens when someone raises a few million rand, hires specialists for performance, security, UX and compatibility, and sells the finished thing for a once-off R499 instead of a subscription that never ends.
Before the engineers start sweating, you still need architecture, security, testing and performance work, and you need people who understand the thousand strange things real customers eventually do to software, plus someone to phone when it all falls over at 2am. What AI does is chew through a big pile of the repetitive grunt work that used to come before any of the interesting engineering could even start. The starting line has moved a long way forward, and that should worry a few established software companies quite a bit.
“But we have more features!”
For decades, software companies protected themselves by stacking features. Five years turned into ten, ten turned into twenty, and thousands of developer-years piled up inside a single product. If you were a startup looking at Photoshop or Microsoft Office, you first had to rebuild twenty years of functionality before you could even think about competing, and that alone kept the incumbents safe.
AI changes that maths. Twenty years of engineering isn’t worthless, because there’s a huge amount of hidden knowledge in all those small decisions and edge cases, but rebuilding the obvious 70 or 80 percent (maybe eventually 90) is getting a lot cheaper. And since most customers never touch the whole of a mature app anyway, a competitor who nails the 20% of features that 80% of people actually use leaves the incumbent with a PRICING problem.
So “we have more features” is a moat that’s slowly drying up. The ones that will still hold water are the things that are genuinely hard to copy:
- TRUST AND RELIABILITY: Customers stay with software that has never lost their work, and that kind of reputation takes years to build.
- PERFORMANCE AND UX: Something fast and pleasant to use will beat something feature-packed and clunky, because people feel that difference every single day.
- DATA, INTEGRATIONS AND ECOSYSTEM: The more your product plugs into everything else a business runs on, the harder it is to rip out, and that’s real value rather than a trap.
- SERVICE AND BRAND: A human who actually picks up the phone is worth more than a hundred features nobody uses.
- PRICING PEOPLE DON’T RESENT: This is probably the cheapest moat on the list, and the one most companies ignore.
Want this button? Upgrade
Subscriptions make complete sense when there are ongoing costs behind them. Cloud storage, servers, AI inference, support, updates and security work all cost real money every month, and charging for that is perfectly fair.
What’s much harder to defend is taking features that cost almost nothing to run and hiding them behind ever more complicated tiers because customers have nowhere else to go. You want this button, so you upgrade. You need another user or a different export format, so you upgrade again. You’d like to use an API that already exists, and suddenly you’re on the enterprise plan, buddy, and the feature the software was always capable of doing costs another R300 a month. At some point you half expect a pop-up saying “Error 402: Your current plan does not include happiness.”
That game becomes a lot more dangerous when somebody with AI tooling can rebuild your core product far faster than was possible a few years ago.
Wrap it up
I think we’re going to see a very interesting breed of company over the next ten years, one that looks at mature SaaS products and asks what would happen if they rebuilt 90% of it, made it excellent and dropped the pricing model everyone hates.
That doesn’t mean free. Good software should make money, because engineers need paying, servers cost money and profit is what keeps a product improving. But charging a fair price for software is a very different animal from taking a permanent cut of somebody else’s revenue just because you once had the market power to do it.
ArtCraft is a warning shot. Today it’s slow and buggy and Adobe is clearly better, and that’s fine, because what matters is how much shorter the distance has become between “I hate how this company prices its software” and “I’m going to build an alternative.”
The incumbents still hold big cards in their customers, expertise, infrastructure, distribution, brand and decades of know-how. They should use those to build products so good that people want to stay, rather than building fancier toll booths. If your business model still boils down to “I sit between you and your customer, so I get a percentage,” or “you used someone else’s gateway, so I get a percentage anyway,” or “you have nowhere else to go, so every useful feature costs extra,” then AI is quietly deleting the most important part of that sentence, which is “you have nowhere else to go.”
So if you run one of those toll booths, now is a good time to ask yourself why your customers keep driving through. And when someone finally combines AI-assisted development, serious engineers and pricing people actually like, I’ll be sitting in the front row with a large popcorn, paid for once, with no subscription and no 2% on the butter.
Emile · free to read, no 2% on the way out