An AI system’s cost includes defining the process, preparing information, implementation, integrations, usage, review and maintenance. A subscription or model usage fee covers only part of it. An estimate becomes useful when it relates to a workload, an acceptable outcome and clear responsibilities.
A demonstration may answer a few questions correctly. Daily use brings updated products, incomplete documents, exceptions, changing offers and situations that were not prepared in advance. The budget needs to account for that work too.
What do you pay for at the start?
First, understand the need: where the process gets stuck, who will use the system and what a good result would look like. Some businesses already have a strategy and organized information. Others need significant work on their catalog or internal responsibilities.
The project then includes user experience, connections to sources and systems, permissions and testing. A public information assistant and a system that records appointments have different requirements, even when both appear in a chat window.
The estimate should specify deliverables, who provides the information and how the work will be accepted. A phased implementation also needs to explain what will be usable at the end of each phase.
Which costs continue after launch?
| Component | What needs to be clear |
|---|---|
| Platforms and usage | What is billed and how it changes with volume |
| Integrations and maintenance | Who responds when a connection or rule changes |
| Information | Who updates products, services and policies |
| Review | How much time evaluation and correction require |
| Support | Who handles cases the system cannot resolve |
| Improvement | How changes are tested before use |
Some components are direct expenses. Others consume team time. Both matter to the business decision, but they should be separated so that cash spending is not confused with the value of resources used.
A simple calculation with explicit assumptions
The following example is entirely hypothetical. It is not a NOT MORE. price, a vendor quote or a market benchmark.
Assume implementation costs €4,000, followed by €300 per month for platforms and usage and €250 for maintenance. Also assume internal review time valued at €150 per month. These figures remain constant for 12 months and exclude additional work.
First-year direct spending would be 4,000 + 12 × 550 = €10,600. Including the value of internal time gives an economic cost of 4,000 + 12 × 700 = €12,400. Tax treatment needs to be specified in a real proposal; every amount here uses the same basis, excluding VAT.
If one month brings 1,000 requests and 600 outcomes validated as resolved, the monthly operating economic cost per valid outcome is 700 ÷ 600, approximately €1.17. This excludes initial implementation. Allocating implementation across 12 months solely for comparison gives (700 + 4,000 ÷ 12) ÷ 600, approximately €1.72.
Define “resolved” in advance, for example as correct information provided or a confirmed operation. A closed conversation is not enough. The calculated cost does not, by itself, demonstrate the investment’s value.
Measure the benefit separately
Compare the system with the existing process: total time, quality, errors, customer experience and relevant commercial outcomes. Released time becomes useful when the team can put it to work. It should not automatically be counted as cash savings or additional revenue.
Double the volume does not necessarily mean half the unit cost. Platform usage, review requirements and difficult cases may grow at different rates. A low cost per conversation can also conceal significant correction work.
How do we keep the investment proportionate?
Start with a bounded use case and a validation budget. Use existing data and systems where they are suitable. After the pilot, decide whether expansion is worthwhile, whether the process needs simplifying or whether conventional automation would be sufficient.
At NOT MORE., we can help with analysis and estimation, implementation or the agreed ongoing operation. The collaboration depends on what is already in place and which part we need to build together.
What would the system need to achieve for the investment to make sense after you include the work around it?