What an AI Agent Costs, and What Drives the Price
What actually drives the price
Most buyers searching for what an AI agent costs find two useless answers: a price list that looks precise but was written without seeing their data, or “it depends” with nothing after it. Both are avoidable, because the things that move the price can be named exactly.
Data readiness. This is the largest factor and almost always the surprise. A year or more of history that reflects the business is enough to start, even if it lives in an ERP and a set of spreadsheets. Missing, duplicated or inconsistently recorded data adds work before a single model is built. You can size this one yourself before we speak: is your data AI-ready sets out the check and what drives the cost of fixing a gap.
How many decisions the system serves. A system that answers one question, how much to order of each item, is one project. A system that also detects customer churn and schedules purchasing is three. Staged pricing exists precisely so you do not pay for the third before seeing the first work.
Integration depth. A report emailed out is cheap. A recommendation that lands inside Priority or SAP and appears in front of the buyer at the moment of ordering costs more, and is worth more, because that is the one that changes decisions.
Closed networks. When everything must run in an isolated environment with no internet connection, the infrastructure is built differently: self-hosted models, controlled file import, and nothing leaving the network. It is a legitimate and common requirement in defense and critical infrastructure, and it affects cost.
How the pricing is structured
Three shapes, matched to the need:
Hourly consulting. For a specific question, a second opinion on a proposal you received, or help with a technology decision. It is also the cheapest way to find out whether there is a project here at all.
Fixed price per stage. For projects. You commit to one stage at a time and know its full price before it begins. The estimation risk sits with us, not with you.
Monthly retainer. After delivery, when a live system needs monthly grading against actuals, calibration and maintenance.
Why scoping is priced separately
Scoping is a bounded, fixed-price work package, sized on purpose so that answering “is this worth it” is itself within reach. It ends with a scope document carrying defined success metrics, for example reducing purchasing error by at least 10% from the current state, and a binding quote for the next stage.
Every stage has an honest exit point. If scoping concludes that the data is not ready, or that AI is not the right tool yet, it stops there and the findings are yours. That is a legitimate outcome, and far cheaper than discovering the same thing after a year of building.
What we do not do
We do not quote before seeing data, we do not charge a platform meter, and we do not lock you in: the models, the code and the data pipelines transfer to you on delivery. If a vendor names a final project price in the first call, it is worth asking what that number is based on.
Where to start
With one decision currently made on gut feel. A scoping conversation of 20 to 30 minutes is enough to tell whether there is a project here and roughly what size it is. If you are still comparing routes, three ways to buy AI maps them, and the questions worth asking before choosing a consultant help filter proposals.
What does an AI agent cost for a business?
Honestly: it depends on scope, and a number quoted before scoping is a guess. What can be said up front is how the pricing is built. Scoping is a bounded, fixed-price work package whose cost you know in advance, and it ends with a written scope and a binding quote for the next stage. You are not committing to a project, you are committing to one stage.
Why don't you publish a price list?
Because a price list would require us to guess your scope before seeing it, and invented numbers are exactly what we teach clients to refuse. Instead we publish what determines the price, so you can size it yourself before we have even spoken.
What affects the cost most?
Four things, in order: data readiness, how many decisions the system has to serve, how deeply it integrates with your existing systems, and whether everything must run inside a closed network. The first two set how much work there is; the last two set how hard it is to deliver.
What is the difference between hourly, fixed-price and retainer?
Hourly suits a specific question or a second opinion. Fixed price per stage suits a project, because it moves the estimation risk to us. A monthly retainer suits life after delivery, when a running system needs monthly grading against actuals and maintenance.
How long before I know whether it is worth it?
Scoping usually runs a few weeks, and it is deliberately sized so that finding out is itself affordable. At the end you know whether your data can carry the decision, what it would take, and what it costs. If the answer is 'not yet', you heard it at the cheap stage.