AI Implementation Services
AI implementation is the work between deciding to automate and having something running that people trust. OperStack sequences it with an exit test per step, so each release can be judged on its own, and stops the programme early when a step fails its test. Most engagements take 6 to 12 weeks for a full sequence, 2 to 4 weeks per step.
| Entry point | Setup · Custom quote Custom quote after the audit. Priced per step, so you can stop between steps. |
|---|---|
| Typical timeline | 6 to 12 weeks for a full sequence, 2 to 4 weeks per step Depends on how much of your data is already clean. |
| Updated | Reviewed by Maksim Shchegolev |
Implementation is mostly sequencing
The technical work in a typical inbound automation is not hard. What decides the outcome is the order, and whether anyone agreed what “done” means before starting.
We sequence so each step has an exit test and a rollback. That constrains us usefully: we cannot claim a step worked because the demo looked good, and you can stop between any two steps with something running.
Data readiness is the invisible half
Plans assume the capture layer emits consistent payloads. It usually does not. Three channels, three field names for the same thing, one of them optional, and the automation downstream now has to guess.
Fixing that is unglamorous and it is frequently the entire difference between an implementation that lands and one that produces confident nonsense. The checks that surface it are in the inbound lead audit.
The exit test, in practice
Before building the routing step, we agree the number: records with no owner after twenty-four hours drops below a stated threshold. After release we measure it. If it did not move, the step failed, and we say so rather than adding a dashboard that shows something else improving.
This is the discipline that most distinguishes an implementation from a project. The economics behind choosing which tests matter are in inbound automation ROI.
Why we price per step
A fixed-price programme creates an incentive to finish the plan rather than to stop when the value runs out. Per-step pricing means we have to justify the next step on the result of the last one.
It also means you can take the roadmap and implement the rest internally, which some clients do after two or three steps, and that is a reasonable outcome rather than a lost sale.
Where implementation is the wrong purchase
If nobody will own the rules after handover, the implementation degrades within two quarters. If the process is still being argued about internally, build a specification first, which is sold separately as AI automation consulting.
What the engagement covers
- Sequenced roadmap with an exit test per step, agreed before any build starts
- Data readiness work: consistent payloads, identity matching, owner fields populated
- Implementation of each step in your environment, on your credentials
- Evaluation against a set built from your real cases, run before and after each release
- Rollback path per step, with a named owner permitted to use it
- Handover documentation written for whoever inherits this after us
What you keep
- The roadmap, with exit tests, usable even if you implement it yourself
- Working steps in production, one at a time, each independently reversible
- Evaluation sets and scores, so the next change has a baseline to beat
What this does not cover
- Enterprise programme management across departments. We implement one process
- Procurement, vendor negotiation, or licence management
- Guaranteed outcomes. We guarantee the exit tests are honest, not that they all pass
If one of these is the actual problem, say so on the audit call and we will point you somewhere better rather than sell you the wrong module.
Start with the written assessment
Free and in writing: send the site and how inbound works today, get the gap we see, the module that fixes it and a fixed price within two working days. No call at this stage.
Frequently asked questions
- Why do most AI implementations stall?
- Two reasons in our experience. The data was less consistent than the plan assumed, and nobody was named as the owner of the rules after launch. Both are visible before the build starts if anyone looks.
- What is an exit test?
- A number agreed before the step is built that decides whether it worked. Without one, every release is judged by whether it feels better, and programmes continue past the point where they stopped adding value.
- Can we stop halfway?
- Yes, and the pricing is per step for that reason. Each step is independently reversible and leaves you with something working. A programme you cannot exit is a programme that keeps going for the wrong reasons.
- Do you need access to production?
- Eventually yes, on credentials you issue and can revoke. Early phases run on exports and a sandbox. We do not ask for broad admin access, and you should be sceptical of anyone who does.
- What if a step fails its exit test?
- We stop, report why, and either fix the underlying cause or recommend abandoning that step. Continuing past a failed test is how programmes end up delivering a lot of work and no measurable change.