Thinking · Guide

What AI projects actually cost.

There's no honest price tag we can put on a webpage — but there's a very real structure to where the money goes. Here's what drives the number, what to nail down before you budget, and how to read an estimate, so you walk into the conversation informed.

Where the money goes

It's almost never the model.

The cost of an AI project is dominated by the work around the model — the data, the integrations, and keeping it running. The model itself is a rounding error by comparison.

30–60%
Goes to the data

Cleaning, labeling, and wiring up your data is the single biggest line item on most AI builds.

Under 10%
Is the AI model

Model and compute fees are a rounding error next to all the work that surrounds the model.

15–30%
Iterative costs

Monitoring, retraining, and upkeep costs to maintain the solution ROI.

3–5×
Demo to production

Hardening a working demo into a system real people can depend on multiplies the cost this much.

Ranges reflect 2026 industry benchmarks (TechTarget, Gartner, and others). If a partner leads with model pricing, ask what they're leaving out.

What drives cost

Six things that move the number.

In rough order of impact. Almost nothing about scoping AI is about the AI — it's about the environment it has to live in.

01

Data readiness

Clean, governed, accessible data means you're ready to build. Scattered and owned by nobody is the project before the project — good data habits cut build cost 20–35%.

02

Integration surface

How many systems must the AI read, write, or sit beside? One is a prototype, three a project, seven a program — and integration alone adds 20–50% to the budget.

03

Scope certainty

"Use AI for support" is a research project. "Sort tickets into seven types at 90% accuracy" is a build. Vague scope isn't cheaper — it's just longer to pin down with you.

04

Who owns it after

An AI system needs a human owner to monitor it, retrain it, and rule on edge cases. With nobody named, the handoff that keeps it working never happens, and run cost climbs.

05

Decision velocity

Every build runs on dozens of small calls — which field, which edge case, good enough or not. Teams that decide in days ship; teams that debate for weeks pay for the wait.

06

Model choice

Almost never the biggest cost, and model pricing keeps falling fast. Unless you're doing something unusually heavy, it's a rounding error next to the five above.

Before you budget

Get clear on these first.

Most AI estimates are wrong because the inputs were fuzzy — a 2025 survey found a quarter of organizations underestimate AI cost by 50% or more. Nail these down and any number gets sharper.

What different sizes look like

From a few weeks to a phased program.

A rough sense of scale — no dollar figures, because the real number comes from the six factors above, not a webpage.

Light · a few weeks

A focused assistant or automation

One job, clean inputs, a clear outcome — like drafting recap emails or summarizing documents. Often lives inside a tool you already have.

Moderate · 1–3 months

A workflow across a few systems

AI inside a defined path, connected to two or three systems, with a human in the loop — like drafting AR collection emails from your ERP.

Substantial · 3–6 months

An agentic system on live data

Multiple steps, live integrations, and real data work underneath — like an operations command center that reads across your tools.

Major · 6+ months

An org-wide, governed program

Several use cases, governance, change management, and a roadmap — best broken into phases so value lands before it's all done.

How we land a real number

The number comes from the work — in three moves.

We don't quote from a webpage. Cost gets pinned down as we move through Dream, Plan, and Build, and each step makes the next one's number sharper.

Dream

Surface the ideas

Find what's worth building.

Workshops demystify AI and surface real use cases, scored on value and data readiness. You leave knowing what's worth building first — and roughly how heavy each idea is.

Fixed-fee workshop · the cheapest way to get clarity
Plan

Pin the scope

Turn "it depends" into a figure.

A short, fixed-fee engagement turns the priority idea into a concrete scope built on your real data and systems — not a guess. This is where an estimate becomes a real number.

Fixed fee · where the number gets real
Build

Build the proven scope

No surprises by design.

We build what Plan proved, priced from a scope you've already seen and signed off on. The number was set before a line of production code — so there's nothing to flinch at later.

Priced from the proven scope, not a hunch
Read any estimate

Questions to ask any AI partner.

If an estimate can't answer these, it isn't a real estimate yet.

Want a real number for your project?

Tell us what you're trying to build, and we'll turn it into a scope and a number you can take to a budget owner — in about 30 minutes, no slides.