I AI'd Everything. Then the Bill Arrived.

I went all in on AI. Every task that could be automated, I automated. Every process that could be streamlined, I streamlined. Every role that could be supplemented, I supplemented.

Customer support? AI handles the first line. Content? AI drafts it. Data processing? AI crunches it. Code reviews? AI catches the obvious stuff before a human touches it.

It was brilliant. Genuinely. Problems I'd been throwing hours at were solved in seconds. Work that used to need a dedicated person was running on autopilot. I felt like I'd cracked the code.

I saved money on headcount. I moved faster. I delivered more with less. Every business podcast and LinkedIn guru told me this was the future and for once they were right.

Then the bill arrived.

The honeymoon phase

When you first start using AI, it feels free. Not literally free, but the cost per task is so low it barely registers. A few pence here, a fraction of a penny there. You're replacing work that used to cost £30 an hour with something that costs £0.003 per request.

The maths is intoxicating. You start finding new things to throw at it. Things you'd never have paid a person to do. Summarise every meeting. Draft every email. Analyse every document. Why not? It costs nothing.

Except it doesn't cost nothing. It costs almost nothing, thousands of times a day.

The creep

Nobody notices AI costs creeping up because they don't look like costs. They look like progress.

Your developer adds AI code review to the pipeline. Great idea. Catches bugs early. But every pull request now triggers three API calls to GPT-4o. Twenty developers, ten PRs a day, three calls each. That's 600 calls a day nobody budgeted for.

Your marketing team discovers that Claude writes better landing page copy than the junior copywriter. So they run everything through Claude. Headlines, descriptions, email subjects, social posts. Dozens of prompts a day, each one a few thousand tokens. The quality is great. The bill is growing.

Your customer support bot handles 80% of queries without a human. Brilliant. But each conversation averages six messages. Each message hits the API. At peak times you're burning through tokens faster than you ever expected because the bot is so good that more customers are using it.

None of this is waste. It's all productive work. That's what makes it invisible. You're not overspending on something useless. You're spending more than you realise on something useful.

The model nobody changed

Here's the bit that really stings. Six months ago your developer picked GPT-4o because it worked. Good choice at the time. It went into the config file and everybody moved on.

Since then, GPT-4o-mini launched. Claude Haiku got cheaper. Gemini Flash arrived at a fraction of the cost. For 70% of the tasks you're running, a model that costs 15 times less would produce identical output.

But nobody went back to check. Why would they? It works. Nobody's complaining. The tickets are getting answered, the code is getting reviewed, the emails are getting drafted.

Meanwhile you're paying premium prices for basic work because inertia is the most expensive AI strategy there is.

The subscription stack

It's not just API costs. It's the subscriptions nobody's tracking.

Your developer has ChatGPT Plus. Your designer has Midjourney. Your content person has Claude Pro. Someone in sales signed up for an AI transcription tool three months ago and used it twice. The marketing intern has a Jasper subscription that auto-renewed last month.

Each one is £15 to £30 a month. Individually harmless. Collectively they're hundreds of pounds a month and nobody in finance even knows they exist because they're scattered across personal cards, expense claims, and buried in software budgets.

Add the API costs on top and suddenly the money you saved by not hiring that extra person doesn't look so saved anymore.

The AI officer problem

And then someone mentions the EU AI Act.

August 2026. Businesses using AI need to catalogue their systems, classify them by risk, maintain audit trails, and ensure transparency. High-risk AI needs human oversight, conformity assessments, and documented governance.

Someone needs to manage this. Someone needs to know what AI tools your business is using, what data they're processing, what decisions they're influencing, and whether you're compliant.

So you need an AI officer. Or at least someone spending a significant chunk of their time on AI governance.

Let me just replay that sequence. You replaced staff with AI to save money. The AI costs are creeping towards what the staff cost. And now you need to hire someone to manage the AI.

The irony isn't lost on anyone who's been through it.

This isn't an argument against AI

Let me be clear. AI is genuinely transformative. The work it does is real. The value it creates is real. Going back to doing everything manually would be madness.

But the "AI is basically free" era is ending. Providers are raising prices. Token costs are going up. Usage limits are tightening. The subsidised pricing that made everything feel like a bargain was funded by venture capital money that now wants a return.

The businesses that thrive with AI won't be the ones that use it the most. They'll be the ones that use it the smartest. The ones that know exactly what they're spending, which models they're using, and whether a cheaper option would do the same job.

What to actually do about it

First, find out what you're spending. Not roughly. Exactly. Every subscription, every API cost, every embedded AI feature in tools you're already paying for. Add it all up. The number will be higher than you think.

Second, audit your models. List every API integration and check which model it's using. For each one, ask whether a cheaper model would produce the same quality output. For routine tasks like drafting emails, summarising documents, and answering FAQs, the answer is almost always yes.

Third, consolidate your subscriptions. If four people are each paying for ChatGPT Plus, switch to a Team plan. If someone's paying for a tool they haven't used in a month, cancel it. If two tools do the same thing, pick one.

Fourth, start tracking ongoing. A one-off audit helps but AI spend doesn't stay static. New tools get adopted, new models come out, pricing changes. You need continuous visibility, not a quarterly spreadsheet.

That's what SpendLil does. It sits between your application and your AI provider, tracking every API call in real time. Provider, model, tokens, cost, per key, per project, per team. When pricing changes, the cost calculations update automatically so you see the real impact immediately.

Your API keys are never stored. Requests are never blocked. One header added to your existing API calls and you have complete visibility over what AI is actually costing your business.

The bottom line

AI isn't the problem. Invisible AI spend is the problem.

The businesses that went all in on AI and never looked at the bill are the ones who'll be caught out. By rising costs, by stealth price increases, by compliance obligations, and by the slow realisation that the money they saved on headcount quietly reappeared on the cloud bill.

The fix isn't to use less AI. It's to know what you're using.

Start there. The rest gets easier.