BYOK Cost Allocation: How to Budget, Track and Split API Spend Across Your Team

BOTrento Team27 October 20268 min read

What you actually pay for with BYOK

Your own key costs nothing in itself. You pay for the text the model reads and writes when it drafts replies, quotes and follow-ups. If you are new to the concept, our explainer on what BYOK is for business owners covers the basics.

Two things drive the bill:

• Volume: how many cases the model touches in a month.
• Length: how much text goes in (long email threads, pasted briefs) and how much comes out (full quote drafts versus a two-line reply).

Nothing else is hidden, but also nothing is capped unless you cap it. That is why allocation and limits matter more with BYOK than with a flat fee.

Estimate monthly spend from enquiry volume

Do not start from users. Start from cases, because one busy salesperson on a quiet month costs less than a quiet salesperson on a busy one.

The formula

Monthly spend = cases × AI actions per case × average cost per action

• Cases: distinct enquiries you work on actively in a month. Count them from your inbox, Pipedrive or HubSpot, or your quote spreadsheet.
• AI actions per case: typically a first reply, a quote draft, and one or two follow-ups. Count what you really do, not what you hope to do.
• Average cost per action: take your provider's current price for input and output text and apply it to a typical thread length. Check the live price list yourself; prices change.

Estimate low, typical and heavy cases separately

A repeat customer asking for a re-quote is cheap. A multi-supplier event brief with a long thread is not. Sort a sample of last month's cases into those three groups, estimate each, and weight by how often they occur. Our guide to estimating AI costs in advance goes through the method and a monthly review template in more depth.

Allocate spend across the team

Per-person billing looks fair but usually creates the wrong incentives: people avoid using the tool on complex, valuable cases because those cost most. Better options:

• By pipeline or service line: for example conference production versus equipment hire. This shows which part of the business the AI cost belongs to.
• By case type: new enquiries, re-quotes, follow-ups. This shows where the text volume really sits.
• By person, only for visibility: useful to spot a habit such as pasting entire threads repeatedly, not to charge anyone back.

Whichever you choose, tag the cost against a revenue line in your accounting tool (Xero, QuickBooks or similar) as a single software or AI usage line. Splitting it any finer than your management accounts need is wasted effort.

Track and cap spend: a monthly checklist

Providers differ in what they show, so check what yours offers. In general, look for separate keys or projects, a usage page broken down by day, and spending limits or alerts. Where you can, use these steps:

1. Use one key per purpose. A key for your sales process should not also run side experiments, or you cannot tell the costs apart.
2. Set a monthly limit or alert at the provider, at a level you have calculated, not guessed.
3. Check usage weekly for the first month, then monthly. Compare against your case count.
4. Calculate cost per case (spend divided by cases). This single number tells you more than the total.
5. Investigate jumps. A rise in cost per case usually means longer threads or repeated regeneration, not more business.
6. Rotate or remove keys when someone leaves the team.

Keep key handling tight: store keys in a shared password manager, never in email or chat, and give access to as few people as possible.

BYOK versus AI included in the tool

The choice is mainly about predictability versus transparency.

BYOK tends to make sense when you already have a provider account and policies around it, when you want usage visible on your own account, or when your volumes are high enough that you want to see exactly what the model costs. Included AI makes sense when you want one predictable line and no extra account to manage. Our article on what your own API key costs and when it pays off compares the two in more detail, and the answer page on AI sales assistant costs for small businesses gives the wider picture.

How BOTrento handles AI cost and BYOK

BOTrento Chief of Staff is a managed process for enquiries, quotes and follow-ups. It watches your inbox and uploaded data, shows a Today overview each working morning, and prepares draft replies, quotes and follow-ups. Nothing is sent without a person approving it.

AI is included in the plan with a monthly usage limit. You get a warning at 80%, and at 100% new work pauses until the next month. There is no automatic extra charge; a higher limit or a bigger plan is agreed with us individually.

BYOK is optional. Today it works with Anthropic (Claude) only; other providers are in preparation. If you use your own key, you pay the provider directly with no markup from us. Keys are stored encrypted, only the last four characters are visible, and you can remove a key at any time. Details are on the API keys help page, and plan details are on the pricing page.

You can test the process before deciding anything. The 14-day trial is free, needs no card and no API key of your own. The 14 days start with your first successful data upload, so you see real usage on your own enquiries before you choose between included AI and your own key.

When BYOK is not worth the effort

Be honest about the overhead. BYOK adds a provider account, a key to protect and a bill to read. If your team is small and your volume is modest, the saving in clarity may not outweigh the admin. Also remember that a key only covers model usage. It does not replace the process around it: who reviews drafts, who owns follow-ups, and who notices a promise made to a customer. Time lost to manual admin is usually a bigger cost than the model itself; see the cost of manual sales admin for a model you can fill in with your own numbers.