Bubble workload units (WU) measure the server work your app does: page loads, database searches and writes, backend workflows and API calls. Plans include 50K (Free), 175K (Starter), 250K (Growth) or 500K (Team) a month, then overages cost $0.30 per 1,000 WU. Find the spike in Logs, App metrics, fix it, and buy a tier rather than upgrading.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowThe upgrade is the expensive way to buy units. On monthly billing, Bubble's manual lists the Starter web plan at $39 and Growth at $134, so moving up costs $95 more a month for 75,000 extra units. Workload tier 1 adds 200,000 units for $29 a month ($26 billed yearly) and halves the overage rate to $0.15 per 1,000. Upgrade for Growth's features (two editors, 14 days of logs, advanced security checks), not for its workload.
How many workload units does each Bubble action cost?
Bubble publishes the base charge for each kind of server work on its activity types page: a page load costs 0.15 WU, a database search 0.3, and writing one record 0.5. You never see an invoice line per search, but once you know the rates, the causes of a spike stop being mysterious. These are the rows that matter most for a small business app:
| Activity | Workload units |
|---|---|
| Loading a page's HTML | 0.15 |
| Performing a database search | 0.3, plus 0.015 per record returned |
| Each character of data returned | 0.000003 |
| Writing or changing one record | 0.5 |
| Deleting one record | 0.1 |
| Aggregate query (count, sum and similar) | 0.2 |
| Running a server-side workflow action | 0.6 |
| Adding an item to the API workflow scheduler | 0.1 |
| Each database trigger event evaluated | 0.05 |
| Each outbound call to an external API | 0.1 |
| Each inbound call to your Data or Workflow API | 0.01 |
| Each call to watch a search for live updates | 0.005 |
A quick sum shows how fast small numbers stack. A page that loads and runs one search returning 40 records costs about 0.15 + 0.3 + (40 × 0.015) = 1.05 WU, before the tiny per-character charge. Load it 3,000 times a month and that is 3,150 WU: harmless. Put that same search on a timer that reruns it every five seconds and the arithmetic changes completely, as the next section shows.
Three rules from the same page catch owners out. Work done on the visitor's device (showing and hiding elements, most navigation) costs nothing, unless a condition on it asks the server something. Your development version counts towards the monthly total alongside live. And while browsing records in the editor's App Data tab is free, uploading, exporting or bulk-editing through the CSV tools uses workload even in the editor. Bubble also notes that a search's real cost varies with its complexity and the amount of data involved, so treat the table as base rates, not an invoice.
A veterinary practice's 408,000-unit month, traced to three causes
Consider a three-vet practice on the Starter plan, billed monthly, running a Bubble app for online bookings, a reception dashboard and appointment reminders. A normal month uses about 120,000 WU, comfortably inside the 175,000 allowance. Then, in the month a new reception screen goes live, usage reaches roughly 408,000 (all figures illustrative, worked from Bubble's published rates).
Cause 1: a reception screen that re-searched every five seconds
Whoever built the new dashboard added a Do every 5 seconds event to refresh today's appointment list "so it is always current". Each refresh searches for about 40 appointments: 0.3 + (40 × 0.015) = 0.9 WU. That is 720 searches an hour per screen. Two reception screens, open nine hours a day, 22 working days: 2 × 9 × 720 × 22 = 285,120 searches, or about 256,600 WU. One checkbox-sized decision used more workload than the whole practice normally does in two months.
The fix costs nothing. Bubble's optimisation checklist points out that search results on a page update automatically over a live connection, so there is no need to re-query; watching a search for updates is charged at 0.005 per call rather than 0.9 per re-search. Delete the timer and the list still changes the moment a booking is made.
Cause 2: a reminder workflow that rescheduled itself every 15 minutes
The reminder logic was a backend workflow that searched all 500 upcoming appointments, sent any reminders due, then scheduled itself to run again in 15 minutes, around the clock. Each run's search alone is 0.3 + (500 × 0.015) = 7.8 WU; 96 runs a day for 30 days is 2,880 runs, or about 22,500 WU a month spent mostly finding nothing to do.
The fix: one daily scheduled run that searches only tomorrow's appointments (say 45 of them, about 1 WU) and uses Schedule API workflow on a list to send them, which Bubble's checklist recommends over recursive workflows because it schedules everything in one operation. Checking costs fall from about 22,500 WU a month to around 30.
Cause 3: an import that set off a welcome email for every old client
Someone imported 8,000 historical client records through the CSV uploader. The writes alone are 8,000 × 0.5 = 4,000 WU. But the Client data type had a database trigger that sends a welcome email whenever a client is created, so it fired 8,000 times: 8,000 × 0.05 for evaluating the trigger plus 8,000 × 0.6 for the email action, another 5,200 WU. Total about 9,200 WU, which in workload terms is the smallest of the three problems. The bigger cost was 8,000 former clients receiving a welcome email. The fix is a condition on the trigger (for example, only when the record's Source is not "Import"), which the checklist also recommends for keeping triggers from firing more often than they need to.
The bill, before and after
| Item | Spike month (WU) | After fixes (WU) |
|---|---|---|
| Normal usage | 120,000 | 120,000 |
| Reception polling | 256,600 | under 100 for live updates |
| Recursive reminders | 22,500 | about 30 |
| Import with trigger | 9,200 | 4,000 (writes only, one-off) |
| Total | about 408,300 | about 124,100 |
| Overage at $0.30 per 1,000 above 175,000 | about $70 | $0 |
Seventy dollars is not a disaster, but the polling would have repeated every month, and a second reception screen or longer opening hours would have pushed it past $100. The tempting response, upgrading to Growth, would have cost $95 more a month and still left the practice over its 250,000 allowance.
How do you turn off Bubble overages, and what happens?
Turn off Bubble overages in Settings, App plan, by unticking Enable overages, and the app goes offline if it exhausts its included units and any add-ons you have bought. Bubble's pricing FAQ says you can disable overages at any time so the app never uses more than the plan's units plus any workload tier you have bought. Bubble's Workload Explained article is clear about the consequence.
So the cap converts a money risk into a downtime risk. Which one you prefer depends on what the app does:
| Kind of app | Overages | Reasoning |
|---|---|---|
| Customer-facing bookings or payments | Leave on, add alerts | An offline booking page costs more than most overage bills |
| Internal staff tool used in office hours | Consider turning off | A few hours offline is survivable; a runaway bill is not |
| Prototype or app still being tested | Turn off | Nobody depends on it yet, and test workflows are where loops start |
| App with recursive bulk jobs | Leave on, set recursion limits | Going offline mid-job can leave data half-processed |
Apps on the Free and Agency plans cannot enable overages or buy tiers at all, according to the same FAQ; those plans are meant for building, so the Free plan's 50,000 units is a hard ceiling.
What alerts does Bubble send before the bill arrives?
Bubble sends two emails automatically: one when the app has used 75% of its monthly allowance and one at 100%, when overages begin. The workload notifications page also lists automatic spike alerts, sent when usage jumps against your recent pattern. Applied to the practice's Starter allowance of 175,000, the thresholds work out like this:
| Spike rule (from Bubble's manual) | What it means on Starter |
|---|---|
| Last hour above 3% of the monthly allowance | More than 5,250 WU in an hour |
| Last 6 hours above 6% | More than 10,500 WU in six hours |
| Last 12 hours above 10% | More than 17,500 WU in twelve hours |
| Last 24 hours 3x the 7-day average (when that average is over 300 a day) | About 12,000 WU in a day for an app averaging 4,000 |
| Last 24 hours 2x the 7-day average (when that average is over 30,000 a day) | Applies only to much busier apps |
The polling screen added about 11,700 WU a day on top of a 4,000-a-day baseline, nearly four times the weekly average, so the 3x rule would have fired on its first full day. Emails only help if the right person reads them. Bubble's manual says custom alerts and recursion warnings go to the app's admins; if the only admin is a freelancer who built the app last year, nobody at the practice sees them. Make sure the owner, or whoever pays the bill, is an admin on the app.
Bubble's manual also describes custom notifications, set from the Notifications tab in Settings or from the Logs page, where you choose an amount of workload and a time period. Two worth creating: one at half the monthly allowance (87,500 on Starter), and one at roughly three normal days' usage within 24 hours (12,000 for the practice). The manual does not state a plan restriction for these; if the option does not appear in your Settings, the automatic emails above are your safety net.
What is Bubble's infinite recursion protection?
Infinite recursion protection is an app-wide depth limit on chains of workflows that schedule other workflows. The most expensive Bubble accidents are workflows that schedule themselves forever. Once a chain passes the limit, the next one is terminated, logged under Workflow errors in the server logs, and an email goes to app admins (at most once a day).
- Where: the API sub-tab under Settings.
- Default: apps created on or after 1 July 2024 start with a limit of 10. Apps created before that have no limit unless someone switched it on, so check any app built earlier.
- Value: leave it at 10 if you have no recursive workflows. If you do, Bubble suggests a buffer above your longest intended chain; its own example is a list of around 500 items, for which it recommends 1,000 or more.
- Two settings: development and live have separate limits, so a lower one in development stops test loops early.
The practice's reminder loop would not have been stopped by this, because it was supposed to run forever. That is the difference between a backstop and a fix: the limit catches accidents, while the daily schedule removes the design problem.
How do you find the culprit in Bubble's Logs?
Open Logs, App metrics to find the culprit, because that is where a spike gives itself away. The Logs tab has three sub-tabs: App metrics, Server logs and Scheduler.
- Open Logs, App metrics. The bar graph shows one column per day for the last 30 days.
- Click the tallest day to zoom into it.
- Read the Workload by activity pie chart. Hovering a slice shows its total WU, its share and the number of runs. In the practice's case, the database search slice would dwarf everything else.
- Click the slice. A list of individual operations appears, ranked by consumption.
- Click the top entry to open the workflow responsible in the workflow editor.
- For a live check after a fix, use the real-time chart, which shows the last 24 hours at one-minute granularity, and Server logs, which show the workload charged for each action and each workflow.
Two habits make this faster. Note the date any new page, workflow or integration goes live, so a step change in the graph can be matched to it. And remember the dashboard uses the time zone of the device viewing it, which matters if a contractor in another time zone is reading the same chart.
What usually causes a Bubble workload spike?
The usual causes of a Bubble workload spike are timers that re-run searches, nested searches, recursive workflows, unconditioned triggers, recalculated totals, open APIs and bulk jobs.
| Cause | How it shows in App metrics | Fix from Bubble's checklist |
|---|---|---|
| Timers: Do every X seconds, Do when condition is true | Steady high search volume during opening hours | Remove the timer; rely on live updates |
| Search inside each repeating group row | Search runs far above page loads | Avoid nested searches; load the data once |
| :filtered and :advanced on big lists | A few searches with outsized cost | Avoid them where a plain search constraint will do |
| Recursive backend workflows | Scheduled workflows slice, runs at all hours | Schedule API workflow on a list instead |
| Triggers with no conditions | Trigger evaluations on every write | Add conditions; combine triggers |
| Totals recalculated on every load | Aggregate queries matching page loads | Store counts and sums; show them on request |
| Inbound API open to anyone | Inbound API calls with no matching activity | Require authentication; limit returned fields |
| Bulk changes and CSV jobs | A one-day spike in writes | Run rarely; test on a small sample first |
Public pages deserve their own mention. A services page that loads a repeating group of 200 records on page load costs about 0.15 + 0.3 + (200 × 0.015) = 3.45 WU per view, and every load counts, whoever or whatever requested it. At 20,000 loads a month that is 69,000 WU for a page whose content changes twice a year. Bubble's checklist suggests loading data after a user interaction and returning less of it; for content that rarely changes, typing it into the page itself removes the search entirely. Limiting what the Data API can return is also a privacy job, covered in setting Bubble privacy rules so clients see only their own data.
AI features are a smaller line than people expect. A workflow that sends each consultation note to an outside AI model for a summary is one outbound call (0.1), a server-side action (0.6) and a few thousand bytes of data, so roughly 0.72 WU per note; 1,500 notes a month is about 1,100 WU. The AI provider's own per-call bill will usually be the larger cost, so watch both.
Can an AI assistant diagnose your Bubble workload spike?
Yes, an AI assistant can help you read what App metrics is telling you, if nobody in the business built the app. Copy the top five operations from the list by hand and ask for a diagnosis against Bubble's published rates:
My Bubble app used 408,000 workload units this month on a 175,000 plan.
Top five operations from Logs > App metrics (WU, runs):
1. Search: Appointments (today) on page reception - 256,600 WU, 285,120 runs
2. Backend workflow send_reminders (scheduled) - 22,500 WU, 2,880 runs
3. Database trigger client_created - 5,200 WU, 8,000 runs
4. CSV upload Clients - 4,000 WU
5. Page load: booking - 9,800 WU, 9,400 runs
For each line: what is probably causing it, the cheapest fix, and the
likely saving. Use Bubble's rates: search 0.3 + 0.015 per record,
write 0.5, server action 0.6, trigger 0.05. Flag anything you are unsure of.
A useful reply blames a timer for line 1 and the 15-minute schedule for line 2. Check every suggestion before acting, and watch for two kinds of bad advice in particular: upgrading to a bigger plan "for headroom", which the sums in the next section show is the dearest option, and adding a Do every X seconds refresh to keep data current, which is the cause of line 1, not a cure. Treat the reply as a list of places to look in the editor, then confirm each one in Server logs.
If you want help working out where AI belongs in your business app, I offer one-to-one AI consultations for small businesses.
Is a workload tier, overages or a plan upgrade cheaper?
A workload tier is usually cheaper than a plan upgrade, and overages alone are cheapest only while usage sits just above the allowance. Once the waste is gone, some apps simply need more units. Here is the monthly extra cost for a Starter web app on monthly billing, using the prices on Bubble's pricing page and manual in November 2026 (tier 1: 200,000 units for $29 with overages at $0.15 per 1,000; tier 2: 750,000 for $99 at $0.14; Growth: $95 more than Starter for 250,000 units with $0.30 overages):
| Monthly usage | Overages only | Add tier 1 | Add tier 2 | Upgrade to Growth |
|---|---|---|---|---|
| 220,000 WU | $13.50 | $29.00 | $99.00 | $95.00 |
| 400,000 WU | $67.50 | $32.75 | $99.00 | $140.00 |
| 700,000 WU | $157.50 | $77.75 | $99.00 | $230.00 |
The break-even points are worth remembering. Tier 1 beats paying overages once you regularly use more than about 272,000 WU a month on Starter, because $29 buys what $0.30 per 1,000 would charge for the first 96,700 units over the allowance. Tier 2 beats tier 1 only above roughly 842,000 a month. Bubble's own workload article makes the same point: if your needs fall between two tiers, the lower tier plus overages is often cheaper. Yearly billing lowers tier 1 to $26 and tier 2 to $89, which shifts both break-evens a little lower.
A plan change still makes sense for reasons other than workload: a second editor, longer server logs, two-factor authentication or the advanced security checks that start at Growth. The overview of Bubble's Starter, Growth and Team plans sets those features side by side, and if you are still on the Free plan, when Starter pays off covers the first step. For the wider question of seat-based against usage-based pricing across tools, see which pricing model costs less for you.
If you want to build the tier-versus-upgrade sums in a spreadsheet, try The Complete Excel Course with AI Integration on Coding Liquids.
What should a monthly Bubble workload check include?
A monthly Bubble workload check takes 15 minutes on the first working day of each month, so put it in the diary. The filled-in column shows what the practice's check looked like after its fixes (illustrative):
| Check | Where | Practice, after fixes |
|---|---|---|
| Last month's total against allowance | Logs, App metrics | 124,100 of 175,000 (71%) |
| Any day above twice the normal daily use | App metrics bar graph | None |
| Top activity type and its share | Workload by activity chart | Database searches, 48% |
| Top single operation | Slice drill-down | Booking page search, 9,800 WU |
| Recursion limit set for live and development | Settings, API | 10 and 10 |
| Owner receives workload emails | App admins list | Yes, plus the bookkeeper |
| Overage setting matches the app's role | Settings, App plan | On (customer bookings) |
| Anything new going live this month | Team diary | Online forms page, week 2 |
If the month shows waste you cannot trace, or the fix needs workflow changes nobody in-house is comfortable making, a few hours of a Bubble developer's time usually costs less than a year of overages; what a Bubble developer costs and what to put in the brief explains how to scope that work so it ends with lower numbers in App metrics, not just changed workflows. Bubble's current rates are on its pricing page, and the per-activity charges on the manual's activity types page.
Bubble workload questions owners ask after a spike
Do AI features in my Bubble app use workload units or AI credits?
Both, for different things. Bubble's pricing page says AI credits pay for Bubble's own AI features, such as chatting with its AI agent while you build. A workflow in your live app that calls an outside AI model is an ordinary outbound API call, so it uses workload units, and the AI provider bills you separately for the model itself.
Does testing in the development version count towards my workload?
Yes. Bubble's activity types page says both development and live usage count towards the monthly total. Browsing records in the App Data tab or deleting them in the editor does not use workload, but uploading, exporting or bulk-editing through the CSV tools does, even inside the editor.
Can I add workload on the Free plan instead of upgrading?
No. Bubble's pricing FAQ says apps on the Free and Agency plans cannot buy workload tiers or switch on overages, because those plans are meant for building. Once you need more than the Free allowance of 50,000 units a month, the next step is a paid plan, from Starter upwards.
Will Bubble refund an accidental workload spike?
Not as a promise: Bubble's public pages I checked do not promise refunds for accidental usage, so plan as if the overage will be billed. Contact Bubble support straight away with the date, the workflow responsible and the fix you made; the infinite recursion protection setting and spike alerts exist precisely so that accidents are caught early.
How many workload units does each Bubble plan include?
Bubble's plans include 50K workload units a month on Free, 175K on Starter, 250K on Growth and 500K on Team, and overages beyond that cost $0.30 per 1,000 units. The Free and Agency plans cannot enable overages or buy tiers, so the Free plan's 50,000 units is a hard ceiling.
Should I turn off Bubble overages?
It depends on what the app does. Turning overages off swaps a money risk for a downtime risk: the app goes offline if it exhausts its units. That suits prototypes and internal staff tools. For customer-facing bookings or payments, leave overages on and set alerts, because an offline booking page usually costs more than an overage bill.
Further reads
- Bubble vs Lovable: Which Should a Small Business Use? — How Bubble's usage billing compares with Lovable's credits over a year.
- How Much Does Lovable Cost? Credits for a Small Business App — What running costs look like on a credit-based AI app builder instead.
- Bubble AI App Builder: Can It Make a Customer Portal? — Build a portal with Bubble's AI and keep its searches lean from day one.
- What Does ManyChat Charge for Extra Contacts After a Viral Reel? — Another tool where a busy month changes the bill, and how to plan for it.
Browse all AI-for-business tutorials
Sources: Bubble pricing page and full plan comparison; Bubble manual pages on pricing and plans, workload activity types, tracking workload, workload notifications, infinite recursion protection, the WU optimisation checklist, using app metrics and the Logs tab; Bubble's Workload Explained article. All read 17 November 2026.
