When a client asks me how long an AI project takes, my answer is always: it depends entirely on what happens in the first four weeks.
The first 90 days of an AI implementation set the trajectory for everything that follows. Projects that spend those weeks properly – validating the use case, auditing the data, getting the environment right – routinely reach a working pilot by day ninety. Projects that skip or rush that foundation phase routinely stall at month five, having spent significant budget without a production system.
According to McKinsey’s 2025 State of AI research, the average enterprise AI project takes 17 months from initiation to production deployment. That number reflects what happens when foundation work is not done in the first phase. With proper upfront discipline, a focused implementation can reach production significantly faster.
Phase 1: Foundation (Weeks 1–4)
The foundation phase is not glamorous and it does not produce anything you can demo. It is also where most projects are won or lost.
Data audit
Before any development begins, audit the data the AI will use. Gartner’s February 2025 research found that 85% of AI projects that fail do so because of poor data quality or missing relevant data. A data audit surfaces these problems before they become mid-project emergencies. It should assess data availability, completeness, consistency, and any access or privacy issues that need to be resolved before development starts.
Use case validation
Confirm that the use case defined in the brief is technically feasible with the data available. This is a two-to-three day exercise with a technical lead and the business owner. It produces either a confirmed go-ahead with a clear technical approach, or a recommended scope adjustment that is far cheaper to make now than after six weeks of development.
Environment and access setup
Development environments, data pipeline access, security approvals, API credentials, and integration documentation – all of these take time to provision in real organisations. Starting this process in week one means it is ready when development begins. Starting it in week four means development stalls while IT processes catch up.
Phase 2: Development (Weeks 5–10)
With the foundation in place, development can move at pace. The development phase has three parallel tracks that need to stay synchronised.
Model development
Building and iterating the AI model against the defined accuracy benchmarks. This phase should have clear exit criteria – the model needs to reach a specified performance threshold before the project moves to integration. Without a performance threshold, model development can extend indefinitely.
Integration development
Building the connections between the AI and the existing systems it needs to read from and write to. This is the work that almost always takes longer than estimated – plan for it. The integration layer is where legacy system complexity, API limitations, and data format mismatches surface in real conditions.
User experience and workflow design
Defining how the intended users will interact with the AI in their actual workflow – not a theoretical interface, but the specific screens, triggers, and outputs that fit the way those users actually work. Involving a sample of end users in this design in week six or seven is one of the highest-leverage investments a project can make in its adoption outcome.
Phase 3: Pilot Deployment (Weeks 11–13)
A controlled pilot with a defined user group, a specific use case, and active measurement against the success metrics agreed in the brief. The pilot is not a test of whether the AI works technically – that was established in development. It is a test of whether the AI works operationally: whether real users use it, whether the output quality holds in real conditions, and whether the integration performs at production load.
The pilot should run for at least two weeks to accumulate enough data to evaluate performance. It should have a named reviewer – the business owner from the brief – who is responsible for assessing pilot outcomes and making a recommendation on full rollout.
Where We Come In
DoSystems structures every AI implementation engagement around this 90-day framework – with the understanding that the foundation phase determines whether the rest is possible. For businesses that have stalled on a previous AI project, the first conversation is almost always about what happened in weeks one through four. DoSystemsInc.com
Frequently Asked Questions
How long does an AI implementation take?
A focused implementation following a structured 90-day framework can reach a working pilot by day 90. McKinsey’s 2025 research shows the average enterprise AI project takes 17 months – a figure that reflects what happens when foundation work is inadequate and projects stall mid-development.
What should happen in the first 30 days of an AI project?
The first 30 days should be the foundation phase: data audit, use case validation, and environment and access setup. This phase does not produce a demo but determines whether the project is feasible and sets up everything that follows.
Why do AI projects stall after the pilot?
Pilots stall most often because the pilot was not designed as a production test – it was designed as a technical demonstration. A proper pilot involves real users, real workflows, active measurement, and a named business owner accountable for the outcome assessment.
What is the most important phase of an AI implementation?
The foundation phase (weeks 1–4). Gartner’s research found 85% of AI project failures trace back to data quality problems that a proper data audit would have surfaced. Projects that rush or skip this phase almost always encounter the same problems later, at significantly higher cost.
#AIImplementation #AIStrategy #AIProjects #DoSystems #AIConsulting #AIRoadmap #BusinessAI
Comments are closed