What actually happens and when
One of the most common questions about AI automation is how long it takes. The honest answer depends on scope, but the pattern is consistent enough to describe a typical timeline. Understanding this timeline helps set realistic expectations and plan the resources you will need at each stage.
Stage one: scoping (one to two weeks)
Scoping is the work that determines whether the project is worth doing and what success looks like. It involves:
- Mapping the current process from trigger to completion
- Quantifying the time, cost and error rate of the current approach
- Defining what success looks like in observable terms
- Confirming that the process is a good candidate for automation
- Identifying the systems, data and people the automation will need to interact with
The output of scoping is a clear brief: this is what we are automating, this is how we will know it worked and this is what we need to build it. A well-scoped project rarely fails. A poorly scoped project rarely succeeds, regardless of how well the technology is implemented.
Stage two: design (one to two weeks)
Design is where the automation takes shape. It includes:
- Designing the workflow structure: what happens at each step, what the assistant does independently, where human review points sit
- Defining the business context the assistant needs
- Specifying the inputs the assistant receives and the outputs it produces
- Designing the handoff points between assistant and team
- Planning how exceptions and edge cases will be handled
Design should involve the people who do the work. Their input on where the friction actually is and how the assistant should fit into their day is more valuable than any process diagram.
Stage three: build and configuration (two to four weeks)
This is the stage most people think of as the whole project. It includes:
- Configuring the AI assistant with the business context and workflow rules
- Connecting the assistant to the systems it needs to access
- Building the integration points where the assistant receives inputs and delivers outputs
- Setting up the review interfaces where the team will interact with the assistant's work
- Creating the monitoring and feedback mechanisms
The duration depends on the complexity of the process and the number of systems involved. A straightforward single-system workflow can be built in two weeks. A multi-system workflow with complex decision logic takes longer.
Stage four: testing (one to two weeks)
Testing against real scenarios, not idealised ones, is where most projects discover what they missed in design. It involves:
- Running the assistant against historical examples to measure accuracy
- Testing with the team using realistic inputs
- Identifying edge cases that were not anticipated
- Adjusting context, constraints and thresholds based on what is learned
- Confirming that the assistant's output meets the success criteria defined in scoping
Testing should include a period of parallel running where the assistant operates alongside the existing manual process. This allows comparison and builds confidence before the switchover.
Stage five: deployment and adoption (two to four weeks)
Deployment is not a single event. It is the period during which the team transitions from the old way of working to the new way. During this period:
- The assistant goes live for real work
- The team begins using the review interfaces
- Early issues are identified and resolved quickly
- The assistant's context is refined based on real-world performance
- The team develops confidence in the automation
The first week of deployment is critical. The team needs visible support, quick responses to issues and regular communication about what is happening and why. A deployment that feels abandoned after launch will struggle to achieve adoption regardless of the technology's quality.
Stage six: steady state (ongoing)
Once the assistant is running reliably and the team has adopted the new workflow, the project enters steady state. The focus shifts from building to maintaining:
- Monitoring performance against the success criteria
- Updating context as the business changes
- Handling new scenarios as they arise
- Expanding scope where appropriate — adding new capabilities to an assistant that has proven reliable
- Reviewing thresholds and review points based on accumulated data
Total timeline
A typical AI automation project for a single, well-defined process takes eight to fourteen weeks from scoping to steady state. The range depends on complexity, not on the size of the organisation or the ambition of the project. Starting small and doing it well is faster than starting large and spending months fixing what went wrong.
The AI automation roadmap provides a broader view of how individual projects fit into an organisation-wide automation programme.
Moonshot Monkeys manages AI automation projects from scoping through to steady state, with a focus on practical timelines and honest communication about what to expect at each stage. If you are considering a project and want to understand what the timeline would look like in your context, we are here to talk.