Two years ago your software bill was boring and fixed. You paid for seats, the number didn’t move, and you forgot about it. Then usage-based AI pricing arrived and the number started breathing. A model provider charges by the token. A transcription tool charges by the minute. Your “flat” writing app added credits that top up automatically when you run out. None of it is a big line item on its own — that’s the trap. It leaks in $9, $19, and “$0.002 per call” increments until you look up and your stack costs more than your rent.
This is a one-sitting audit. Give it 45 minutes and you will typically cut 30–50% of a personal or solo-business software bill without losing a single tool you actually use. The point isn’t to go cheap — it’s to stop paying for things you forgot about and to put a ceiling on the bills that can run away from you. Do it once properly, then keep it clean with a 20-minute monthly pass.
Step 1 — Pull the whole list (30 minutes of truth)
You cannot cut what you can’t see, and memory is a liar here. Do not list your subscriptions from recall. Pull the receipts:
- Export the last three months of your credit-card and bank statements to CSV. Three months catches quarterly and annual charges, not just monthlies.
- Search your email for “receipt,” “invoice,” “your subscription,” and “payment successful.” Annual renewals hide here.
- Check the App Store / Google Play subscription screens and your PayPal “automatic payments” — the classic blind spots.
Drop every recurring charge into a single table. Keep it brutally simple:
- Tool — what it is.
- Monthly cost — divide annual plans by 12 so everything compares.
- Pricing type — fixed seat, or variable/usage-based. Mark the variable ones with a star; those are the ones that bite.
- Last real use — when you genuinely opened it, not when it last charged you.
Total the column. That number — the real one, annualized in your head ×12 — is usually the motivation you needed.
Step 2 — Score every line with the KEEP rubric
Now judge each tool on four questions. Score each 0, 1, or 2 and add them up (max 8). This turns a vague “eh, I might need it” into a number you can act on.
- K — Kept busy? Did you use it in the last 30 days? (0 = no, 1 = once or twice, 2 = weekly+.)
- E — Edge? Does it do something no other tool you own does? (0 = fully duplicated, 2 = genuinely unique.)
- E — Earns? Does it make or save you real money/time? (0 = nice-to-have, 2 = tied to income or hours.)
- P — Pain to replace? How locked-in is your data/workflow? (0 = swap in 10 minutes, 2 = deeply embedded.)
Read the totals like this: 6–8 = keep, no debate. 3–5 = fix it (downgrade, consolidate, or cap — see below). 0–2 = cut today. The magic is in the low scorers: the tool you bought for one project, the “pro” tier you upgraded during a deadline and never used again, the second AI subscription that does what your first one already does.
Step 3 — Run the four-way decision
Every line resolves into one of four moves. Decide out loud:
- Keep. High score, fair price. Leave it — but if it’s a variable-priced tool, it still gets a cap in Step 4.
- Downgrade. You use it, but not at the tier you’re on. Most people are one plan too high “just in case.” Drop a tier; you can always move back up in 30 seconds if you feel the ceiling.
- Consolidate. Two tools, one job. A general-purpose AI assistant now does what your separate summarizer, your grammar add-on, and your slide-drafter each charged for. Pick the one that scored highest and kill the overlaps.
- Cut. Low score. Cancel it now, while you’re looking at it. “I’ll cancel later” is how you got here.
Step 4 — Put a ceiling on the variable bills
This is the step that didn’t exist three years ago and is now the whole ballgame. Every tool you starred as usage-based can, in a bad month, cost 5× what it did last month. You are not going to babysit them — you’re going to fence them.
- Set a hard budget, not a soft one. In the billing settings, find “spend limit” or “budget cap” and set an actual number that stops service, not just an email. If the tool only offers an alert, set the alert low — at ~60% of your ceiling — so you get warned with room to react.
- Turn off auto-top-up. The single most expensive default in AI billing is credits that silently refill. Switch it to manual. Running dry for a day is cheaper than a runaway invoice.
- Watch the double-charge. Some tools bill you per seat and per usage. If you’re paying a monthly platform fee and metered credits, ask whether the seat fee already includes an allowance you’re not using before the meter starts.
- Prefer a predictable cap when volume is steady. If your usage is consistent, a flat monthly plan often beats pay-as-you-go. If it’s spiky and rare, metered wins. Match the plan to the shape of your usage, not to the marketing.
For the model-level stuff specifically, the discipline is the same one behind the $2 test: know the per-run cost of the thing before you automate it a thousand times.
Step 5 — Cancel without the guilt spiral
Cancellation flows are designed to make you flinch. Don’t negotiate with the pop-up; use a script. Two you’ll reuse forever:
The clean cancel (when you’re done): “Please cancel my subscription effective at the end of the current billing period and confirm in writing that no further charges will occur. I don’t need a call — email confirmation is fine.”
The retention play (when you’d stay for less): “I’m planning to cancel because the cost has outrun my usage. If there’s a loyalty discount, a lower tier, or paused billing, I’ll stay — otherwise please cancel at period end.” Then wait. The discount, if it exists, arrives in the next reply.
Do the cancels in one sitting with your inbox open, and file every confirmation email in a “Cancelled” label. That folder is your proof if a “whoops, still charging you” renewal shows up.
A worked example: $740 → $310 a month
Take a freelance consultant — call her the composite of every stack we’ve seen. Her audit table came to $740/month. Here’s how the moves landed:
- Two AI writing assistants ($20 + $30). Both did drafting; her general assistant already covered it. Consolidate — kill both, −$50.
- A design tool at the top tier ($55). Used weekly, but never near the pro limits. Downgrade to mid — −$30.
- Usage-based transcription (~$90 last month, auto-top-up on). Real workhorse. Keep + cap — manual credits, hard limit at $40. −$50 in a normal month.
- A project tool from a client engagement that ended in March ($45). Score: 0. Cut — −$45.
- A “pro” analytics add-on ($60) she opened twice. Cut — −$60.
- Stock-media + a scheduler + a form tool she’d stacked over time ($75 combined), all overlapping her main platform. Consolidate/cut — −$60.
- Model API credits with auto-refill. Switched to manual and set a monthly ceiling. −$40 of “phantom” spend, and one nasty surprise-month avoided.
New total: about $310. Nothing she uses weekly got touched. The cuts came entirely from duplication, dead tools, over-high tiers, and uncapped meters. That’s $430 a month — north of $5,000 a year — recovered in under an hour. Where it goes next is a separate decision; if it’s heading to savings or investing, run it through your money operating system so it doesn’t just re-leak somewhere else.
Keep it clean: the 20-minute monthly pass
Audits decay. New tools sneak in during busy weeks and never leave. So attach a tiny recurring review to something you already do — the day your main card statement posts is the natural trigger.
- Open the statement. Scan only for new or bigger charges since last month.
- For anything new, ask one question: did I mean to keep paying for this? If not, cancel now with the script.
- Glance at your two or three variable-priced tools. Did any blow past their cap? Adjust the ceiling or the behavior.
Twenty minutes, twelve times a year, and the stack never balloons again. This pairs naturally with a broader annual look at your personal AI stack — the audit controls cost; the stack review controls whether you’ve got the right tools at all.
Failure modes to avoid
- Cutting the workhorse to feel virtuous. The goal is less waste, not less capability. If a $90 tool saves you ten hours, it stays — cap it, don’t kill it.
- “I’ll cancel at renewal.” You won’t. Cancel now; most services run to period end anyway, so you lose nothing.
- Downgrading into pain. If a lower tier throttles something you rely on daily, that’s a false economy. Move back up — the point is to match the tier to reality, in both directions.
- Ignoring the annual traps. The subscription you don’t see monthly is the one that renews for $240 in a single hit. That’s exactly why Step 1 pulls three months, not one.
- Auditing once and calling it done. Without the monthly pass, you’ll be back here in a year with the same leak and a bigger number.
Our take: Usage-based pricing quietly moved the risk from the vendor to you — your bill now depends on behavior no one is watching. The fix isn’t frugality, it’s visibility plus a ceiling. Pull the real list, score it honestly, cap the meters, and put a 20-minute pass on the calendar. Do that and your software spend stops being a thing that happens to you and goes back to being a thing you decide.
