How to prevent surprise AI software bills

The short answer

Surprise AI software bills come from six places: usage meters such as credits and tokens that run faster than expected, charges for every person who opens an app, overages that bill automatically, AI features inside apps that draw on the same balance, security features that only come with a higher tier, and add-ons. Prevent them by choosing pricing your finance lead can forecast in one sentence, setting a hard cap, turning on alerts at 80 and 100 percent, confirming nothing bills by itself, and reviewing usage monthly.

A surprise software bill is rarely about the headline price. The plan said $25 or $50 a month, and that number was true. The surprise comes from everything around it: a usage meter that ran faster than anyone expected, a charge for every coworker who opened an app, an overage that billed itself, or a security feature that turned out to live on a plan costing several times more.

AI software makes this worse, because AI work is measured in units most people cannot picture, and the amount used depends on how hard the AI has to work, including the times it gets things wrong. This guide maps the six sources of surprise AI software bills, with real examples from vendor pricing pages as listed in October 2026, and gives finance leads and owners a complete set of controls to keep the bill predictable.

The six sources of surprise

Source	How it surprises you	Control
1. Usage meters	Credits, tokens or workload units run out faster than planned	Prefer pricing you can count; set a cap
2. App user charges	Every person who opens an app adds to the bill	Price app users before rollout; prefer free app users
3. Automatic overages	Usage past the allowance bills without anyone approving it	Turn off automatic overages; require approval
4. AI inside the apps	Finished apps draw on the same balance as building	Ask what draws on the balance besides building
5. Tier jumps	Security features you need sit on a much higher plan	Price the plan your IT lead actually needs
6. Add-ons and extras	Storage, extra editors, hosting traffic	List every add-on before you sign

1. Usage meters

Many AI tools charge by usage. The unit has different names, but the idea is the same: a balance of AI work that each request draws down.

Tokens. On May 13, 2025, Vercel moved v0 to usage metered on input and output tokens, converted into credits, instead of a fixed number of messages. v0 Business, as listed in October 2026, is $100 per user per month with $30 of included monthly credits per user.
Credits. Lovable prices its plans by credits, and its pricing page says one balance covers building, hosting with its Cloud, and AI features inside your apps.
Workload units. Bubble introduced Workload Unit pricing in April 2023, which drew heavy pushback on its forum. Bubble's manual says usage over the included amount is billed per extra 1,000 workload units.

The surprise with usage meters is that nobody knows in advance how much the AI will need. A request that works on the first try uses a little. A stubborn problem the AI fixes, breaks and fixes again uses much more. Review sites are one rough signal of how customers feel about a vendor overall; as of October 17, 2026, for example, Bolt.new's Trustpilot TrustScore was 1.6 out of 5 across 204 reviews.

Some meters are also per person. Bolt's pricing page says Teams is $30 per member per month and that tokens are not shared between team members, so one busy builder can run out while another has plenty left.

2. Charges for people who only use the apps

This is the surprise that hits at rollout. A tool is built by one person, then shared with the whole crew, and the bill grows with every coworker who opens it.

Power Apps pay-as-you-go is $10 per active user per app per month, according to Microsoft Learn. An active user is someone who opens the app in the month.
AppSheet is $5, $10 or $20 per user per month, and every app user, including guests, needs a license.
Retool Business charges $15 per internal user per month billed annually, and external users are free up to 50, then $4 to $8 per user per month.
Softr caps the people who can log in to your apps on each plan, and bills extra app users monthly on Pro and above.
Glide Classic Business includes 30 users, then charges $5 to $6 per additional user, plus 2 cents per update over 5,000.

For Lovable, Bolt, v0, Replit and Base44, no per-user charge for app users is listed on their pricing pages. Our guide to AI app builders that charge app users compares this in detail.

3. Automatic overages

When you hit your allowance, one of three things happens: work stops, you are asked to buy more, or the extra usage is billed automatically. The third is where surprise bills come from. Find out which applies, and whether you can switch automatic overages off.

4. AI features inside the apps

Some platforms let the apps your team builds use AI themselves, such as summarizing notes or drafting messages. If those features draw from the same balance as building, a popular app can consume the allowance your builders were counting on. Ask what draws on the balance besides building.

5. Tier jumps for security and control

The plan that fits your budget may not be the plan your IT lead needs. As listed in October 2026, Lovable puts SCIM and audit logs on Enterprise only; Bolt lists SSO, audit logs and data retention policies as Enterprise features; Retool lists SAML/OIDC SSO and SCIM as Enterprise only. If you need those controls, the real price is the price of that plan.

6. Add-ons and extras

Storage, extra editors, hosting traffic and outside services can each add a line to the bill. List every add-on that applies to your use before you sign.

An example: how a reasonable plan becomes a surprise

Here is a worked example using one vendor's listed price, to show how per-user charges multiply. It is an illustration, not a quote, and it leaves out any other charges.

A 25-person company builds three internal tools: a job board, a purchase request form and an equipment log. It uses pay-as-you-go pricing at $10 per active user per app per month, as listed by Microsoft for Power Apps.

Month one: only the 5 office staff use the job board. 5 people x 1 app x $10 = $50.
Month three: all 25 people use the job board and the purchase form, and 8 use the equipment log. (25 + 25 + 8) x $10 = $580.

Nothing about the price changed. The bill grew more than tenfold because the tools worked and people used them. That is the pattern to plan for: success should not be the thing that surprises your budget.

A forecasting formula for finance leads

Use this to compare vendors on the same footing. Fill it in for a realistic month, then for a busy one.

Builder cost: number of builders x price per builder seat on the plan you need.
App user cost: number of people who only use the apps x price per app user (or per user per app).
Usage cost: expected usage beyond the included allowance x the overage or top-up price. Add a buffer, because AI usage varies.
Tier cost: the difference between the plan you priced and the plan that includes the security features you need.
Add-ons: storage, extra editors, hosting and outside services.

Total = 1 + 2 + 3 + 4 + 5. If you cannot fill in line 3 with any confidence, the pricing is usage-based and you should rely on caps and alerts.

Controls that keep the bill predictable

Before you buy

Write a one-sentence forecast. "Next month this costs X because Y." If your finance lead cannot write it, the pricing is too hard to predict.
Count builders and app users separately. Price each against the vendor's model.
Price the right tier. Include the plan your IT lead needs for roles, sign-in and records.
Ask what happens at the limit. Stop, ask or bill automatically.
Ask whether failed attempts count. If the AI's mistakes use your allowance, budget for them.
List add-ons. Storage, extra editors, hosting, outside services.
Start monthly, then switch. Measure real usage for one or two months before committing to annual billing.

After you buy

Name one owner. A single person owns caps, alerts and the monthly review.
Set a hard cap. The most the company can spend in a month, enforced by the platform.
Turn on alerts. At 80 percent and 100 percent of the allowance, sent to the owner and the finance lead.
Turn off automatic overages where you can. Make extra spending a decision.
Put it on one company card. Spending scattered across personal cards hides the total.
Review monthly. Ten minutes: total spend, top users, unused seats, any overage.
Remove unused seats. People who stopped building should not hold a paid seat.
Hunt for shadow subscriptions. Check expense reports for AI tools bought outside the approved platform. Our guide to shadow AI explains how.

Questions to ask every vendor about billing

What exactly is the unit we pay for, and what uses it up?
Do the AI's failed attempts and fixes count?
Do people who only use the apps cost anything?
What happens when we reach the limit?
Can we set a cap, and who can raise it?
What alerts do we get, and at what levels?
Which plan includes the security and admin features we need?
What add-ons or outside charges might apply to us?

A simple monthly review

Put this on the calendar for the first business day of each month.

Check	What to look for	Action
Total spend vs budget	More than 10 percent over	Find the cause before the next month
Usage by person	One person using far more than others	Ask whether something is stuck in a loop
Overage charges	Any at all	Decide whether to raise the plan or cap usage
Unused seats	Seats with no building last month	Remove or reassign
New subscriptions	AI tools on expense reports	Consolidate onto the approved platform
App user growth	New people added to apps	Confirm the cost, if any

How FORGE handles it

FORGE was built so a finance lead never has to guess. The pricing is designed around things you can count.

$49 a seat a month for each person who builds, with 3 builds a month included.
Changes are a small fraction of a build and only count when something actually changes.
Extra builds come in packs of 10 for $150. Your team can share one company pool or give each seat its own allowance.
App users are free. Rolling a tool out to the whole crew is never a budget decision.
A build FORGE fails is free. You never pay for the platform's mistakes.
Caps and alerts. Admins can cap how many pack builds the team uses each month, and owners get alerts at 80 and 100 percent.
Nothing ever bills by itself. When builds run out, new builds wait until someone decides.
Apps that rest when idle. Apps nobody has opened for 10 days sleep and wake the moment someone opens them.

See the full details on our pricing page, or talk to us about what your team would pay.

Questions people ask

Why do AI software bills vary so much month to month?

Many AI tools charge by usage, measured in credits, tokens or workload units, and the amount used depends on how much the AI has to do, including retries and fixes. Others charge per active user, so the bill rises as more people open the apps. Both are hard to predict without caps and alerts.

What is a token or a credit?

Both are units of AI work. A token is a small piece of text the AI reads or writes; a credit is a vendor's own unit that bundles AI work, and sometimes hosting, into one balance. The more the AI reads and writes, the more you use.

What is an overage?

An overage is a charge for usage above what your plan includes. Some vendors bill overages automatically, some let you turn them off, and some stop work until you buy more. Find out which before you start.

How do I forecast AI software costs?

Count three things: the people who build, the people who only use the apps, and how much building you expect each month. Then price each against the vendor's model, including the plan tier you need for security features, and add a buffer for usage-based charges.

Should we pay annually or monthly?

Annual billing is usually cheaper per month, but it commits you before you know your real usage. Many teams start monthly for one or two months, measure, then switch to annual once usage is stable.

Who should own AI software spending?

One named person, usually the admin or finance lead, should own caps, alerts and the monthly review. Spending spread across personal cards and expense reports is how surprise bills hide.

How does FORGE keep bills predictable?

FORGE is $49 a seat a month for each person who builds, with 3 builds included. Changes are a small fraction of a build and only count when something changes. Packs of 10 builds are $150, admins can cap pack use, owners get alerts at 80 and 100 percent, app users are free, a build FORGE fails is free, and nothing ever bills by itself.

See what FORGE builds for your team

We onboard every team ourselves, starting with the work that costs you the most time.

Talk to us