How much does it cost to implement AI in a business? The six costs of an honest proposal

The cost of implementing AI in a business has six parts: setup, a monthly operating fee, AI model usage, licenses for other tools, future changes and support. The total depends on the workflow you pick, the volume and the integrations. An honest proposal shows each part on its own line and puts a cap on usage. If it shows a single number, information is missing.

You get two proposals. One costs twice as much as the other. Both promise “smart customer service.” Neither says how much you’ll pay in the third month, when volume goes up.

That’s where regret lives. The project works, but the monthly bill becomes a surprise.

The six costs

1. Setup. Mapping, designing, integrating and testing. Paid once, by scope.
2. Monthly operating fee. Hosting, monitoring and small adjustments to keep the system running.
3. AI usage. AI models are billed by the volume of text processed, measured in tokens. More conversations, bigger bill. It’s the cost that varies the most and the one that most often disappears from proposals.
4. Licenses. The official WhatsApp API, the CRM, the automation tool. Each has its own price and its own increases.
5. Changes. The process changes, the rule changes and the system needs to keep up. Agree up front on how that gets priced.
6. Support. Who answers when something stops, how fast and through which channel.

Why usage needs a cap

AI usage varies by nature. A campaign that takes off, a peak season month, a customer who sends a five-minute voice message. The bill follows.

A monthly cap solves it. You define how much the system can spend and what happens near the limit: it sends an alert, saves AI for priority tasks or hands the rest to the team. Without a cap, you’re signing a blank check.

You can cut usage right in the design. Use rules where rules work, call AI only for messages that need interpretation and pick a model sized to the task. To tell a rescheduling request from a quote request, the most expensive model on the market is a waste.

Hours freed up and money saved are different things

Be careful with the math that goes “the system does ten hours of work a week, so it saves this many dollars.” Freed-up hours only turn into money when someone uses that time on something else that brings in revenue.

Often the real gain is something else: the front desk serves people better, the partner stops getting interrupted, the client gets an answer the same day. That has a lot of value. It’s just a different kind of value, and mixing the two inflates everyone’s expectations.

That’s why the baseline comes first: time spent, volume and errors, measured in your operation. After the test, compare the same period. The number you get is yours, not a market average.

The cost nobody puts in the spreadsheet

It’s the cost of the project left half done. The automation was bought, nobody was put in charge and the team went back to the spreadsheet. The monthly fee keeps coming.

A small pilot, on a single workflow and with an agreed stopping criterion, limits that risk. If it doesn’t work, you find out while spending little.

Questions for any proposal

• How much does AI usage cost per month at the current volume? And at double?
• Is there a cap? What happens when it’s reached?
• Which licenses do I pay for separately?
• How are changes billed after delivery?
• If I want to leave, do I get documentation and access to everything that was built?

If the vendor can’t answer, the proposal isn’t ready yet.

At Korai Labs, each of these costs gets its own line in the proposal, with a defined usage cap. Before any proposal, you can start with the AI-ization Diagnostic: it shows which workflows need AI and which can be solved with something cheaper.

How much did your business spend last year on tools nobody uses?