AI consulting cost depends mainly on what the consultant is responsible for delivering. An assessment, a focused pilot, and a production implementation are different products. Scope, systems, data, risk, testing, adoption, and ongoing ownership all affect the price.
I know “it depends” is not the answer anybody wants when they ask how much AI consulting costs.
But a single number would be misleading.
One company may need an assessment of a clearly defined workflow. Another may need an AI system connected to five applications, using permission-sensitive information, with human review and monitoring built around it.
Calling both of those “AI consulting” does not make them comparable.
Before comparing prices, first identify what you are buying.
The main types of AI consulting engagement
| Engagement | Primary output | Main cost drivers |
|---|---|---|
| AI readiness assessment | Current-state findings and recommended next steps | Stakeholders, business units, processes, data, risk, and systems reviewed |
| Opportunity assessment | Prioritized use cases and business case | Number of opportunities, depth of analysis, and evidence available |
| Focused pilot | A tested solution for one defined use case | Data access, integration, evaluation, security, and user testing |
| Production implementation | A working system inside the real workflow | Systems, users, controls, exceptions, deployment, adoption, and handover |
| Ongoing improvement | Monitoring, support, and controlled changes | Usage, service requirements, model changes, vendor changes, and workflow changes |
None of these engagement types is automatically better than another.
The problem is when a company believes it is buying implementation and later discovers the proposal only covered strategy—or when it pays for a large strategy exercise even though the workflow and required outcome were already clear.
What changes the cost?
1. The size of the process
One workflow with one owner is different from a process crossing several departments, locations, and approval paths.
The more people, handoffs, rules, and systems involved, the more work is required to understand and change it safely.
2. The systems involved
If the solution must connect with CRM, finance, document management, case-management, or operational systems, the work includes field mapping, access, validation, testing, monitoring, and failure handling.
The model may be the easy part. The integration around it is often where the real implementation work lives.
3. The condition of the data
Clean, accessible, well-owned information reduces preparation work.
Scattered documents, duplicated records, inconsistent definitions, missing permissions, and unclear ownership increase it.
AI cannot quietly fix disagreements about what the data means. Somebody still has to resolve them.
4. Risk and controls
A tool drafting an internal summary needs fewer controls than a system affecting payments, employees, customers, claims, regulated records, or other consequential decisions.
Permissions, audit evidence, human review, security testing, retention, and incident handling all affect the scope.
5. Exceptions and testing
The normal path is rarely the expensive part.
Missing information, contradictory records, unusual inputs, failed connections, low-confidence outputs, and human escalation rules create much of the real complexity.
If a proposal only covers the happy path, the lower price may not mean a lower final cost.
6. Adoption and ownership
Training, documentation, workflow changes, support, and named ownership determine whether the system is used after launch.
A technically correct tool that the team avoids has not created value.
What should an AI consulting proposal include?
Use this checklist before comparing quotes:
| Proposal item | What should be clear |
|---|---|
| Business problem | The process, users, current issue, and intended outcome |
| Scope boundary | What is included, excluded, and assumed |
| Deliverables | The specific documents, designs, integrations, software, tests, and training provided |
| Client responsibilities | Required people, access, information, decisions, and approvals |
| Systems and data | Applications, sources, permissions, integrations, and preparation work |
| Controls and testing | Human review, security, evaluation cases, exceptions, and acceptance criteria |
| Operating cost | Licenses, model usage, infrastructure, monitoring, support, and maintenance |
| Ownership | Who owns the code, accounts, configuration, prompts, documentation, and future changes |
If one proposal includes implementation, testing, monitoring, documentation, and handover while another only includes recommendations, comparing the totals tells you very little.
Compare the deliverables and responsibilities first.
How to avoid paying for the wrong thing
Start with one measurable operational problem and the smallest useful process boundary.
Confirm who owns the process and information. Decide what must remain under human review. Establish the baseline. Then compare a process change, existing software, integration, normal automation, AI, and focused custom development.
Sometimes a short assessment is the right first step. Sometimes the business already understands the problem well enough to begin implementation.
The point is not to make every engagement smaller. It is to make sure the scope is tied to a decision or result the company actually needs.
At Losung Consult, we scope AI work around the process, systems, controls, and intended outcome—not around a predetermined tool.
Bring us the initiative and the constraint holding it back. We will help define what needs to be understood before it can be estimated responsibly.