G.28Guides · Decision brief
What drives the cost of AI implementation
The AI line item everyone debates, model and tooling subscriptions, is usually the smallest number in a real implementation. The costs that decide the budget are older and less fashionable: data readiness, integration depth, process redesign, governance, and the human work of adoption. Vendors quote the small number; the business pays the whole column.
The distinction
What is actually being compared?
No responsible page publishes a dollar figure for what AI implementation costs: the honest range across use cases spans several orders of magnitude, and any specific number would be an invented statistic. What can be said rigorously is which variables move the total, in which direction, and which ones buyers systematically underestimate. On that last point the pattern is stable: technology is over-estimated, data and people are under-estimated, and the gap is where AI budgets go to die.
The first variable is use-case scope, and it dominates. Assistive tools for an existing workflow, drafting, summarizing, searching, sit at the cheap end: low integration, human judgment intact, failure tolerable. Automated decisions inside core operations sit at the expensive end: deep integration, high reliability requirements, real governance, and failure that costs money or reputation. A one-sentence description of the use case, with the words assist or decide in it, predicts the budget's magnitude better than any vendor quote.
The six drivers, in the order they surprise people
Data readiness: if the data the use case needs is scattered, inconsistent, or undocumented, its preparation becomes the largest single line, and no model choice changes that. Integration depth: connecting to real systems of record, with permissions, logging, and failure handling, costs multiples of any pilot that ran in a sandbox. Build versus buy: buying a product is cheaper until your requirements diverge from its roadmap; building is dearer until scale and specificity pay it back. Reliability and governance requirements: review workflows, audit trails, and compliance sign-off scale with the decision's stakes, and in regulated settings they can exceed the engineering. Adoption and process redesign: training, role changes, and the redesign of the workflow the tool lives in, the line vendors omit because it is not theirs to sell. And ongoing operation: monitoring, evaluation, and model change management, a permanent cost that pilots never surface.
How to keep the budget honest
Budget the column, not the line: for every technology dollar, force estimates for the other 5 lines, data, integration, governance, adoption, and operation, before approving anything. Pilot for evidence of the expensive parts: a good pilot is designed to reveal integration and data costs, not to hide them in a sandbox where everything works. And stage the spending against value gates: each phase releases budget only when the previous one produced measured operating value, which is the discipline that separates AI adoption from AI theater. The pattern to refuse is the fixed all-in quote over admitted unknowns; in AI work the unknowns are the budget.
Side by side
The drivers and what moves each.
| Pushes cost up | Keeps cost honest | |
|---|---|---|
| Use-case scope | Automated decisions in core operations | Assistive scope first; decide-scope only with evidence |
| Data readiness | Scattered, inconsistent, undocumented data | A data audit before any tool commitment |
| Integration depth | Live systems of record, permissions, logging | Piloting against real systems early, not last |
| Governance load | Regulated decisions, audit and review needs | Sizing governance to stakes from the start |
| Adoption work | Role changes and process redesign, unbudgeted | A named adoption owner and budget line |
The call
Before you approve an AI budget
- 01
Write the assist-or-decide sentence.
State the use case in one sentence containing assist or decide. It sets the reliability bar, the governance load, and the budget's order of magnitude before any vendor conversation starts.
- 02
Demand the full-column estimate.
Reject any budget that shows technology without data, integration, governance, and adoption lines. The missing lines are not zero; they are merely unowned, and they will be yours.
- 03
Stage spend behind value gates.
Release budget phase by phase against measured operating value. AI work rewards this discipline unusually well, because each phase generates exactly the evidence the next phase's estimate needs.
A note on interest. Praxis sells consulting, so treat this page as an informed party’s brief, not a referee’s ruling. The discipline we hold ourselves to is written down: category-level comparisons only, no named competitors, and a public page on when we are not the right fit.
Questions
Asked before scoping.
- Why do AI pilots so often succeed while rollouts fail?
- Because pilots are usually built to avoid the expensive parts: sandbox data, no integration, volunteer users, no governance. The rollout then meets all four at once. A pilot designed honestly, against real systems and real users, costs more and predicts the rollout, which is the entire point of piloting.
- Does Praxis implement AI or advise on it?
- Praxis advises: use-case selection, build-versus-buy reasoning, budget structure, governance design, and the adoption work, with engineering delivered by your team or your vendors. The practice's AI implementation and adoption services cover exactly the lines this page says vendors omit, and that is the declared interest in writing it.
Where this leads on the site
Other decision guides
- G.01Strategy vs management
- G.02Consultant vs contractor vs fractional
- G.03Boutique vs Big Four
- G.04Consultant vs in-house
- G.05Change vs transformation
- G.06Fractional vs retainer
- G.07AI consultant vs implementation partner
- G.08Data strategy vs engineering
- G.09Consulting vs coaching
- G.10Interim vs consultant
- G.11SEO consultant vs agency
- G.12Transformation vs modernization
- G.13What drives cost
- G.14Fee structures
- G.15Fixed vs T&M
- G.16First-engagement budget
- G.17Questions to ask
- G.18Red flags
- G.19Writing an RFP
- G.20Evaluating proposals
- G.21Do you need one?
- G.22Getting the value
- G.23When to hire strategy
- G.24SEO for construction
- G.25Board vs advisory board
- G.26Marketing consultant vs agency
- G.27Piloting an advisor
- G.29Strategy vs ESG reporting
Decided what kind of help you need?
Then the next conversation is about fit and scope. Tell us what you are deciding, and we will tell you honestly whether we are the right resource for it.
No obligation · a scoping conversation first