What should an AI project cost?
A day rate tells you very little without a scope. Here is how to compare AI project proposals, understand the ongoing costs and decide whether a first experiment is worth funding.
AI consultancy costs depend on the decision or system you are commissioning. Separate discovery, development, integration, evaluation and team handover from recurring model, hosting, licence and support charges. Dragon AI quotes against an agreed scope; the example below is a planning exercise, not a price list or market average.
Compare the deliverable before the fee
An assessment of one workflow is a different purchase from an internal knowledge assistant connected to several permission-controlled sources. A prototype may demonstrate an idea without including production authentication, monitoring or support. Two proposals with the same headline can therefore cover very different work.
Ask each supplier to identify what will exist at handover, what evidence will show it works and what your team must supply. Dependencies on data cleanup, licences or third-party access should be visible before you compare prices.
| Component | What to clarify |
|---|---|
| Discovery | Workflows, interviews, source review and the decision it will support. |
| Implementation | Features, systems, supported inputs and explicit exclusions. |
| Evaluation | Test cases, quality criteria, permissions and failure recovery. |
| Handover | Documentation, training, account ownership and support arrangements. |
| Recurring costs | Model usage, hosting, storage, software licences and maintenance. |
The scope choices that change the cost
Integration work depends on the access a system offers. One supported API with a clear data structure is a different challenge from reconciling several inconsistent records. Permissions can also change the design: an assistant with one approved public collection has different requirements from a system that must respect each employee’s document access.
Output quality needs evidence. Allow time for representative test cases, difficult examples and corrections. Adding more sources or actions increases the set of behaviours that need checking. A smaller, well-evaluated first release can give you a better investment decision than a broad demonstration.
- Number of systems and whether supported access exists.
- Source quality, update frequency and access boundaries.
- Consequences of a wrong answer or action.
- Usage volume, response-time requirements and review effort.
Build a business case using your own assumptions
Consider a hypothetical process with 600 tasks per month. If a pilot reduces net handling time by four minutes per task, that releases 40 hours. At an assumed internal cost of £25 per hour, the capacity value is £1,000 per month. Subtract an assumed £200 in monthly operating costs and the planning figure becomes £800.
With an illustrative £6,000 setup cost, simple payback would be 7.5 months. These are invented inputs to explain the calculation, not Dragon AI prices or a promised result. Released capacity only becomes a cash saving if expenditure actually falls; otherwise its value depends on what your team can do with that time.
Now stress-test the assumptions. At two minutes saved instead of four, the same operating costs leave £300 per month of capacity value and a 20-month simple payback. Measure review time, exceptions and support effort in the pilot before approving a larger investment.
| Assumption | Four minutes saved | Two minutes saved |
|---|---|---|
| Tasks per month | 600 | 600 |
| Hours released | 40 | 20 |
| Capacity at £25/hour | £1,000 | £500 |
| Less £200 running costs | £800 | £300 |
| £6,000 setup / net value | 7.5 months | 20 months |
Choose a pricing structure that matches the uncertainty
A fixed scope is useful when the deliverable and dependencies are understood. A discovery phase can resolve uncertainty before an implementation quote. Time-based work needs a budget boundary, visible priorities and a clear review point. Ongoing support should state what is included and how additional changes are handled.
Ask about account ownership, access to source code, export options and the work required to move to another supplier. Those details affect long-term cost just as much as the initial build.
What to bring to a scoping conversation
Describe the task, its monthly volume, the tools involved and the outcome you need. Bring representative examples and explain who will approve the result. If the budget is fixed, say so; it helps define a useful first scope.
We can then separate what needs investigation from what can be implemented, explain ongoing charges and agree a decision point before expanding the project.