Working out the profitability of an AI assistant you resell to clients
Credit costs, time spent, resale price: a simple method to check whether the AI assistant you resell to your clients actually makes money for your agency.
Costs and AI models5 min read
Reselling an AI assistant to your clients can become a steady source of revenue, provided you know what it costs you. Many agencies set a price by gut feeling, then find out a few months later that the time spent tuning the agent weighs more than expected, or that one busy client consumes far more than the others.
This article offers a simple method to calculate your margin, with a worked example based on the current prices in the blabla.bot catalogue. The figures in the example are estimates: real consumption depends on the conversations.
The three parts of the cost
Your cost for a client site breaks down into three parts.
- The assistant’s consumption. Each answer consumes tokens (small pieces of text) going into and coming out of the AI model. blabla.bot debits them from your credits at the provider’s cost multiplied by two.
- Knowledge indexing. Turning the client’s pages and PDFs into usable knowledge also has a cost, billed the same way. It applies when a source is added and each time it is re-indexed.
- Your time. Setup, testing, tuning instructions, updating sources, talking with the client. It is often the largest part, and the one most often forgotten.
The basic formula is therefore:
Monthly margin = monthly resale price − consumption − indexing − value of time spent
Estimating consumption
Consumption depends on three things: the number of conversations, their length and the AI option chosen.
To support estimates, the blabla.bot catalogue uses assumptions per option: a reference conversation has 4 exchanges, each with a given number of input tokens (instructions, knowledge passages, history, question) and output tokens (the answer). With the catalogue prices checked on 13 November 2026 and the ×2 multiplier applied, the estimates are as follows.
| AI option | Estimated billed cost per conversation | Per 1,000 conversations |
|---|---|---|
| GPT-6 Luna · Simple | €0.0042 | €4.20 |
| Mistral Small · Simple | €0.0060 | €6.00 |
| GPT-6 Luna · In-depth | €0.0080 | €8.00 |
| Mistral Small · In-depth | €0.0113 | €11.28 |
These amounts are orders of magnitude, not measurements. Longer conversations, very detailed instructions or a large knowledge base push them up. Once the agent is live, the console shows you the real cost: the “AI model” screen displays a monthly estimate based on the last 30 days of conversations, and the portfolio shows each site’s billed cost over 30 days.
Estimating indexing
Indexing is usually modest. The embeddings service (the computation that finds the right passages) currently costs €0.10 per million tokens before the multiplier, so €0.20 billed per million tokens.
For example, a 50-page site with around 4,000 characters per page comes to roughly 50,000 tokens. Indexing it costs about one cent. The console shows an estimate before each indexing run and the actual cost afterwards. A page that has not changed since the last indexing is not billed again.
Valuing your time
Only you can put a figure on this part. Track the time you actually spend on your first clients:
- initial setup (sources, instructions, appearance, testing, installation);
- monthly follow-up (reviewing answers, adding a source, adjusting an instruction);
- client communication (reports, change requests).
Multiply by your hourly rate. Keep setup time, which is often billed once, separate from follow-up time, which your monthly price has to cover.
A worked example
Here is a fictional example to illustrate the method. Resale prices and time spent are assumptions: replace them with your own.
An agency equips 5 client sites. Each site handles about 400 conversations a month. The chosen option is GPT-6 Luna · Simple, the default recommended option.
| Item | Calculation | Monthly amount |
|---|---|---|
| Consumption | 5 sites × 400 conversations × €0.0042 | €8.40 |
| Indexing | a few re-indexing runs, about a cent per site | under €1 |
| Follow-up time | 5 sites × 30 min × €60 hourly rate | €150.00 |
| Total cost | about €159 |
If the agency charges a package of €50 per site per month, it takes in €250 a month, for a margin of about €91. The lesson is immediate: follow-up time weighs far more than consumption. That is what you need to control first.
Protecting your margin
Two risks threaten the profitability of a fixed package.
The client who consumes a lot. A site jumping from 400 to 4,000 conversations multiplies its consumption by ten. To protect yourself, set a monthly budget per site in the console. An alert warns you at 80% of the budget by default. Once the budget is reached, that site’s assistant alone pauses until the next month or until you raise the cap. Spell out in your contract what happens then: a higher tier, billed overage, or an accepted pause.
Time creep. A client who keeps asking for changes eats into your margin. Define what monthly follow-up includes (for example, one review and one source update) and what counts as extra work.
Choosing the AI option with margin in mind
The AI option has a direct effect on consumption. In-depth modes produce longer answers and draw on more passages, which costs roughly twice as much. Keep them for sites where questions are complex. For a shop or a service business with simple, repetitive questions, Simple mode is often enough.
Redo the maths after a month
Estimates get you started. After a month of real use, rerun the calculation with the actual figures from the console: conversations, billed cost per site, time spent. Adjust your price or budget if needed.
To run this calculation in seconds with your own assumptions, use the calculator on the Pricing page: enter the number of sites, conversations per month, the AI option and your resale price, and it estimates your cost and margin.
Tags
- profitability
- resale
- costs
- agencies
On the site
Read next
Costs and AI models
How much does a conversation with an AI assistant cost? Tokens explained
Input tokens, output tokens, context and history: what really makes up the cost of a conversation with an AI assistant, and how to bring that cost down.
· 5 min read
Costs and AI models
Monthly budget per site and alerts: avoiding unpleasant surprises
Monthly cap per site, alert threshold, balance alert: how to set the blabla.bot safeguards so that a sudden traffic spike does not eat into your margin.
· 5 min read
Agencies and white label
Web agencies: offering an AI assistant to your clients and reselling it
Why an AI assistant fits naturally into a web agency offer, how to set one up for a client step by step, and how to set your own resale price with confidence.
· 5 min read