Skip to content
Matas Beniušis

What AI implementation costs: what drives the price and how to compare proposals

Five things that determine the price of an AI and automation project, three hidden costs that proposals leave out, and the questions that let you compare two proposals.

Author: Matas BeniušisPublished 3 min read

In short

The price of AI implementation is driven not by technology but by the scope of the process, the number of integrations, the state of the data, the cost of an error, and how much work your team does. Two proposals with the same figure can differ several times over in what stays with you after the project. Compare not the figure but what is included: a pilot on real data, documentation, handover, and the cost of changes a year later.

“How much does it cost?” is the first question and the hardest to answer honestly before an assessment. Not because the price is hidden, but because it is driven by things that cannot be seen from a conversation. This article explains what drives it, which costs proposals usually leave out, and how to compare two proposals when their figures are similar.

What drives the price?

The five factors that change the price most. None of them is technology.
FactorWhy it changes the priceWhat you can do yourself
Scope of the processOne step with a clear input and output is several times cheaper than a process that crosses three departments.Break the process into steps and start with one.
Number of integrationsEvery system to connect to is separate work. Systems without an API cost the most.Find out whether your programs have APIs before asking for a proposal.
State of the dataData in tables is cheap. Data in photos, scanned PDFs and free text is expensive.Collect data samples and show them as they are.
Cost of an errorThe more expensive one mistake is, the more checks, approvals and testing are needed.Decide where a person must approve a decision and where not.
Your team's involvementIf your people do the assessment and the pilot, there is less outside work.Assign one person who will give time to the pilot.

Which costs do proposals leave out?

  • Maintenance. A system that connects to other programs stops working when those programs change. Ask who will fix that and at what price.
  • Model and tool fees. AI models charge per request. A proposal should include a rough monthly figure at your volume.
  • Changes. Six months in, you will want the system to do one more thing. If the code and documentation are not with you, every change is a new proposal.

How do you compare two proposals with a similar figure?

The figures are similar; the contents rarely are. Give both vendors the same six questions and compare the answers, not the numbers.

  • Will there be a pilot on our real data before the system is built, and can we stop after it?
  • Where will the system run: in our accounts or yours?
  • Who will own the code and documentation after the project?
  • Who on our team will be trained to maintain the system, and is that included?
  • What will one change cost a year from now?
  • What will models and tools cost per month at our volume?

A proposal that answers all six in writing is almost always cheaper over two years, even if the first figure is higher.

When does the price not matter at all?

When the benefit has not been counted. If you do not know how many hours a month the process takes now and what one error costs, you cannot say whether a proposal is cheap or expensive. That arithmetic takes one afternoon and is the first thing I do in the assessment stage. Without it, any figure in a proposal is just a figure.

My own prices and pricing model are stated on the service pages.

Let's talk about where the most manual work is in your business.

A 30-minute call. You show me how the process runs today. I tell you what I would automate first and what I would leave alone.

No pitch. A diagnostic conversation. Oopsify: a chat with an AI, no signup, a report by email.