Operations2 min read
What happens after you launch an AI solution?
Maintenance, costs and evolution: what to agree on before putting a system into operation.
By the team at Daleki Lab
Building and operating are different responsibilities
Building delivers a solution with an agreed scope. Operating makes sure that solution stays usable as data, tools or needs change. Mixing the two stages leaves open expectations from the start.
Our model separates a fee per solution from a monthly fee for ongoing support. What each includes depends on the project. It does not mean unlimited development, permanent support or automatic inclusion of every external service.
Make recurring costs visible
Beyond ongoing support there may be infrastructure costs, AI providers, messaging or other tools. The proposal should identify who contracts each service, who manages it and how its usage is controlled.
A system needs usage limits and a periodic cost review. An estimate should stay labeled as an estimate until it is compared with real operations. It is not wise to promise a fixed third party cost without verifiable conditions.
Agree on what happens during an incident
Define a support channel, owners, hours and criteria to prioritize incidents. Also a way to suspend actions and restore service. A contact button is not an availability agreement.
Before expanding permissions or autonomy, confirm that the team can intervene and that there is enough documentation to understand the expected behavior. Human approval has to be practical, not just a sentence in the proposal.
Prioritize evolution with evidence
Improvements should respond to observed needs. Review what gets used, what fails and what is still solved manually. From there, agree on priorities and decide which fit within the monthly capacity and which require a new project.
A long term partnership is sustained by visibility and shared decisions. The goal is for your company to gain capacity, without depending on a black box or on commitments nobody defined.
