I would think really hard about signing up for a £500 per month SaaS subscription. The number sits there on the pricing page, visible and final, demanding a decision that feels consequential. I'd compare it against alternatives, justify it to myself, maybe even ask someone else whether it made sense. That friction is deliberate. It's designed to make you pause.

But I can easily spend £2,000 a month on credits if I'm building something tangible.

That gap between £500 and £2,000 is doing a lot of work. It's not about the money. It's about the psychological architecture of how the spending happens. Credit-based pricing in AI tools like Lovable, Cursor, and others has engineered a model that pairs expenditure with achievement in a way that flat subscription fees never could. You build something, see the results, feel the hit of accomplishment, spend credits largely invisibly, and end up topping up instinctively. The abstraction severs the emotional connection to real money. It's brilliant product design. The question is whether buyers are sophisticated enough to recognise it.

The Endorphin Economy

Credit-based pricing has been in and out of SaaS forever, but it feels like it's coming back strong now because the consumption model finally aligns with how people actually use AI tools. You're not paying for access. You're paying for output. And output feels productive in a way that access never quite does.

Lovable does a fantastic job pairing credits with endorphins. The cycle is tight and satisfying: you prompt the tool, it generates a working prototype, you see tangible progress, you feel like you've accomplished something. The credit spend happens in the background. When you run low, topping up doesn't feel like a purchasing decision. It feels like refuelling mid-journey. You're not stopping to evaluate cost. You're maintaining momentum.

One developer described spending an entire month's Pro credits in a single afternoon debugging a Stripe integration. Heavy users report burning through 100 monthly credits in a single debugging session, pushing real monthly spend well past the sticker price. A CRUD app with authentication burns 30 to 60 credits just for the initial scaffold. Two revision rounds, a payment integration, a bug-fix session, and you're at 100 credits before the end of week one. Credits hit zero mid-build, mid-feature, mid-debug.

The friction that would stop you signing up for a £500 subscription doesn't exist here. You've already started building. The cost is invisible until it's not. And by the time it's visible, you're committed.

Casino Chips for Engineers

This isn't accidental. It's engineered very deliberately. Kudos to the vendors.

Research from MIT shows that subscription models reduce the "pain of paying" by separating the purchase decision from the payment experience. Credit cards create 12 to 18 per cent higher spending than cash payments due to reduced payment friction. A peer-reviewed meta-analysis across 71 studies and 392 effect sizes confirmed the pattern: people consistently spend more when removed from cash.

Credit-based pricing abstracts away the underlying mathematics. "You have 10,000 credits left" is easier to internalise than "you've consumed 4.3 million tokens at £0.0025 per thousand." But the abstraction runs deeper than most buyers realise. Credits and tokens are not always directly proportional. In some tools, one credit maps cleanly to a fixed token count. In others, the relationship shifts depending on the model being called, the complexity of the task, or the context window in use. A simple prompt and a complex multi-step reasoning task might cost the same number of credits on the surface, whilst consuming wildly different amounts of compute underneath. The credit is a unit of convenience for the vendor, not a unit of transparency for the buyer.

That inconsistency matters. It means you can't reverse-engineer your spend. You can't look at last month's credit consumption and build a reliable model for next month, because the same number of credits doesn't represent the same amount of work. It depends on what you built, which models you invoked, and decisions the platform made on your behalf that were never surfaced to you.

It's the same principle as casino chips. Nobody spends £2,000 at a casino table, but they'll burn through £2,000 in chips without blinking because the abstraction severs the emotional connection to real money. The chip is a proxy. The credit is a proxy. And proxies don't hurt the same way.

I used to be highly sceptical about credits as a whole. How could anyone buy something where they don't know how much it will cost at the end of the day? Many tools make the value of a credit very ambiguous or even hidden. But the market seems to be proving otherwise. Lovable is used by around 8 million vibe coders and churns out 1 million projects a week. The market seems very happy with it.

The Enterprise Fault Line

Perhaps there's a difference here between enterprise and single contributor appetite. Because a solo founder topping up credits instinctively is one thing. When that same behaviour pattern gets replicated across a hundred engineers in an enterprise, the numbers compound in ways nobody modelled at procurement stage.

KPMG's Q2 2026 Global AI Pulse survey of 2,145 senior executives across 20 countries found that 29 per cent had no idea where the growing costs associated with AI were coming from. A further third confessed that their own cluelessness about AI economics was a barrier to successfully deploying AI in the workplace. Only 26 per cent of large enterprises have full, real-time visibility into what their AI systems cost to operate. While 66 per cent have monitoring dashboards, only 36 per cent have implemented direct token or usage controls.

This isn't a usage problem. It's a governance problem.

The market is super immature here. Enterprises are generally capping usage, but everyone is having a hard time estimating what the real usage or cost is or should be. We've seen news of organisations blowing through their annual AI budget in a month. When organisations cap the credit usage, engineers end up using their session, daily, or weekly limits and then being less productive waiting for the limits to be replenished.

Someone said somewhere that you should say good morning to Claude before saying good morning to your spouse, so that the first five-hour usage session starts early and you get two sessions per working day. Obviously Anthropic does this slightly differently in some plans where usage is limited by the monthly plan, but it's a good example of some of the things people do to try and get the most out of the tools.

That line is genuinely one of the most revealing things about where we are with this. It shows that engineers have already internalised the credit economy as a constraint they have to work around, not a fair exchange they're participating in. There's something almost absurd about a senior engineer optimising their morning routine around a token window reset. That's cognitive overhead that nobody accounted for in the productivity calculation.

Two Ways This Could Go

When governance catches up, and enterprises do start modelling this properly, credit-based pricing will either evolve or collapse under scrutiny at scale. There are two paths forward.

Path one: organisations start cost-modelling based on the average user. Some users use less, some more, but the average lands at a predictable point. This is hard because the market is new, and very few organisations have even 12 months of actual usage data to base their models on. The dual billing structure is the most common source of "I expected £25, I got £60" complaints on Lovable, especially for founders who get past the prototype stage and suddenly have real traffic.

Path two: credit-based usage survives for some segments, but enterprise clients start gravitating towards flat fee offerings, which are easier to buy and budget for. There's real value in offerings that are easy to buy. SaaS and AI tools could offer both.

A Bain & Company analysis published in October 2025 examined 30-plus SaaS vendors introducing generative AI capabilities and found roughly 65 per cent had adopted hybrid pricing, layering a credit or usage meter on top of a base subscription. As one director of monetisation at an enterprise productivity company stated: "Credits gave us breathing room while we figured out the real value metric. But they're not intuitive to buyers."

That dual-track model is interesting. Credits for the builders who want flexibility, flat fee for the enterprises who want predictability. But it creates a different problem: vendors now have two levers to pull, and the incentive is always to nudge enterprise clients toward whichever model extracts more value.

The Question Procurement Isn't Asking

When a vendor comes in pitching a credit-based model, the question that most procurement and finance teams aren't asking but absolutely should be is this: what is the value of a credit to my business?

Not what a credit costs. What it generates. If you can get to the point where you know that a credit, at 50p each, generates £2 in revenue, the conversation changes entirely. It stops being about cost governance and becomes about investment optimisation. That's a completely different negotiation — and a far more useful one.

Getting to that correlation, credit spend to actual business value, is genuinely hard. Most organisations can't even connect their SaaS spend to revenue outcomes, let alone something as granular as token consumption. The responsibility lands between a CTO and a Chief AI Officer, both purely business-focused roles. One of the core competencies of a CTO is to correlate technology investment to business value, and be able to articulate that to the board and the rest of the executive team.

And yet most CTOs are still being handed a credit bill that looks nothing like a traditional software invoice. No line items they recognise, no benchmarks to compare against, no historical data to trend. They're being asked to articulate value to a board using a currency nobody fully understands yet.

KPMG's Rob Fisher framed the shift clearly: "AI is now as much a financial management priority as it is a technology one. The real risk isn't investing in AI but doing so without cost visibility and an understanding of the economics of AI."

What Credit-Based Pricing Should Look Like

I'd like to see credit-based pricing as a genuine evolution in how software value gets exchanged. But it is definitely very convenient to vendors, especially when buyers are still immature enough to not question the value of a credit.

The version of credit-based pricing that I'd actually be proud to recommend to a client without reservation would look like this: transparent credit value and pricing to help the client draw the correlation, cost that genuinely scales with usage, with enterprise-level volume discounts built in. Commit to spending a certain amount with us over a certain time period, and we'll give you a discount that reflects the relationship.

Transparent value, genuine scale economics, and commitment-based discounts that reward the relationship rather than obscure it. That's essentially asking vendors to behave like partners rather than extractors. The model you'd trust is the one that mirrors how the best consultancies operate: no hidden agendas, value you can actually measure, and terms that make sense for both sides of the table.

The market will mature. Governance will catch up. CTOs will start demanding that credit-to-revenue correlation. The question is whether vendors will meet that demand with transparency, or whether they'll keep optimising for the gap between what buyers expect to spend and what they actually do.

For now, the credit economy is working exactly as designed. You build something, feel productive, top up instinctively. And somewhere between the endorphin hit and the invoice, £500 becomes £2,000 without anyone noticing.