Your AI Bills Keep Growing. Here's How to Find Out What You're Actually Paying For.
It never starts as a decision. It starts with one tool, then a colleague adds another, then a per-seat plan quietly renews, then an API bill arrives that nobody can explain. Eighteen months later the company is spending thousands per month on AI-related tools and exactly nobody can say what the total is or what it returns. This is AI spend sprawl, and it's the natural consequence of pricing models designed to grow invisibly: low entry points, per-seat expansion, usage-based billing. The fix isn't to spend less on AI — it's to spend knowingly. This page gives you the audit method we run with clients, the build-vs-subscribe math, and honest calibration ranges for what working AI actually costs per month.
The 4 places AI money leaks
Leak 1: overlapping subscriptions. Three tools that do substantially the same thing, because marketing bought one, operations another, and someone signed up for a third with a credit card. Nobody ever compared them because nobody owns the category. Leak 2: per-seat pricing on ghost seats — you're paying for 25 seats, the usage dashboard shows 7 active humans. Leak 3: unmonitored API usage. Usage-based pricing means a single inefficient automation — one that re-processes entire datasets instead of deltas, or loops without limits — can burn hundreds of euros monthly with no one watching the meter. Leak 4: enterprise tiers bought for a feature that was never switched on. The audit usually finds all four, and the surprising part is rarely the total — it's how little of it connects to outcomes anyone can name.
The AI spend audit: one focused afternoon
- pull every AI-related charge from accounting — tool subscriptions, API invoices, per-seat licenses. Don't forget the credit cards: shadow signups are normal, not a scandal.
- for each line item, answer four questions: who uses it, for which process, how often, and what happens if it disappears tomorrow. That last question is the honest one — 'nothing much' tells you everything.
- open the usage dashboards. Every serious tool shows active users; compare that number to what you're paying for.
- for usage-based spend, identify the top three consumption sources. In practice one workflow usually dominates the bill, and often it's an inefficiently built one — the fix is engineering, not budget.
- mark every line keep, consolidate, downgrade, or kill. In an illustrative audit of this kind, 30–50% of current spend is identified as recoverable — before anything changes about how the work actually gets done.
Build vs subscribe: the honest math
The question isn't 'which is better' but 'which fits this situation'. Subscriptions win for light, generic, uncertain uses. Custom-built automation wins at volume, when deep integration matters, and when you're paying for several overlapping tools that one purpose-built system replaces. Breakeven versus stacked subscriptions arrives at around 12–24 months.
| Situation | Subscribe | Build custom |
|---|---|---|
| Light, generic use; low volume | Best fit | Overkill |
| High volume with per-use pricing | Costs scale linearly forever | Usually wins from year 1–2 |
| Needs deep integration with your systems | Often impossible | Design strength |
| Unproven use case | Validate cheaply first | Build after validation |
| 3–5 overlapping subscriptions | Consolidate | Often fully replaced by one system |
What businesses actually pay — indicative ranges
Illustrative reference ranges for an SME: uncontrolled sprawl totals in the region of €1,000–5,000 per month across tools, seats, and usage. A consolidated stack — one or two subscriptions plus one or two purpose-built automations with managed API usage — runs in the region of €300–800 per month ongoing, with better outcomes attached. The purpose-built pieces carry a one-time build cost (€8,000–25,000 depending on scope), which is where the 12–24 month breakeven comes from. Two implications worth stating plainly. First, if someone proposes building custom AI while you're at €300/month in effective subscriptions, that's usually the wrong advice — custom earns its cost at volume. Second, if you're at €3,000/month of sprawl with no measurable outcome, the problem isn't the amount — it's that you can't name what any of it does. The audit fixes that in an afternoon.
Related pages
- AI automation cost guide coming soon
- Build vs buy AI coming soon
- AI ROI not visible
- 5 ways AI can reduce operational costs
Frequently asked questions
How do I find all the AI tools my company is paying for?+
Pull accounting records by vendor category, then run a quick team survey for credit-card signups — shadow subscriptions are normal in every company we audit.
Are API costs really that unpredictable?+
Yes. Usage-based billing means one inefficient automation or one heavy month can spike costs. Monitoring dashboards and budget alerts are essential, not optional.
When does building custom actually save money?+
When per-transaction pricing or several overlapping subscriptions exceed the amortized build cost — typically at 12–24 months of the subscription alternative.
Can you run this audit for us?+
Yes — a spend audit is a fixed-price engagement, scoped to pay for itself from identified savings alone, before any optimization work begins.