• AI implementation
  • Process

The 4-phase AI implementation framework: from audit to live

A step-by-step AI implementation framework for growing businesses: audit, blueprint, build and scale, with the decisions and outputs at each phase.

ForceMX Team5 min read

TLDR:

  • An AI implementation framework breaks a project into phases, each with outputs and a go/no-go decision before more money is spent.
  • The ForceMX framework has four phases: audit, blueprint, build and scale.
  • The audit takes one to two weeks, the blueprint about a week, and most builds two to four weeks per system.
  • Each phase can end the project early and cheaply if the case for it does not hold up.
  • Adoption and measurement are part of the plan from the start, not an afterthought.

An AI implementation framework is a phased method for taking an AI idea from first assessment to a system people use every day, with a decision point at the end of each phase. The point of the phases is to spend small amounts to learn whether a project is worth doing before spending large amounts to build it. Research from RAND on why AI projects fail points to the same causes a phased approach is designed to catch early: misunderstood problems, inadequate data and missing infrastructure [1].

This article describes the four phases ForceMX uses, what each produces, and the decision at the end of each.

What happens in the audit phase?

The audit maps how work actually flows today and scores each process for AI readiness. It typically takes one to two weeks.

Activities:

  • Interviews with the people who do the work and the people who manage it
  • A map of the main processes: steps, tools, hand-offs, time spent and common errors
  • A readiness score for each process, covering problem clarity, data, stability and team readiness (see the AI readiness checklist)
  • A rough estimate of the benefit for each promising process

Outputs:

  • A ranked list of AI opportunities
  • A list of processes where AI is not the right answer, with the reason
  • A recommendation for the first project

Decision: is there at least one process where the expected benefit clearly outweighs the effort? If not, the project stops here, and the business keeps the process map.

What happens in the blueprint phase?

The blueprint turns the chosen opportunity into a costed, measurable plan. It usually takes about a week.

Outputs:

  • The exact scope: what the system will do, and what it will not
  • How it fits the current tools, such as the CRM, email, accounting system or shared drives
  • Where a person must review or approve the output
  • The success metrics, with today's baseline measured
  • The cost, timeline and expected return

Decision: does the business case still hold with real numbers? A blueprint that shows a weak return is a good result, because it costs far less than a failed build.

What happens in the build phase?

The build phase creates the system, tests it with real users and rolls it out in stages. Most builds take two to four weeks per system.

Activities:

  • Build in short cycles, with a working version to review at the end of each
  • Test on real past cases and compare results with how the team handled them
  • Launch to a small group of users first and fix what they report
  • Train each team on their own tasks
  • Roll out to everyone once the small group uses it reliably

Outputs:

  • The live system, connected to your existing tools
  • Documentation and a named owner on your side
  • A first measurement against the baseline

Decision: is the pilot group using it, and are the numbers moving? If usage is low, the cause is found and fixed before the wider rollout.

What happens in the scale phase?

The scale phase makes the result stick and extends what works. It is ongoing.

Activities:

  • Monthly reporting against the baseline
  • Refinement based on errors and user feedback
  • Extending the system to similar processes or other teams
  • Refresher training as the team and the tools change

Decision: which process comes next? The audit's ranked list becomes the roadmap.

What does each side do in each phase?

The business supplies the knowledge of the work; ForceMX supplies the method and the build.

Phase Your team ForceMX
Audit Explains how the work is really done Interviews, maps, scores and estimates
Blueprint Confirms scope, metrics and priorities Designs the solution and prices it
Build Tests with real cases and gives feedback Builds, integrates and trains
Scale Owns and uses the system Measures, refines and extends

What are the most common pitfalls?

Most pitfalls come from skipping a phase to save time.

  • Skipping the audit leads to building for a process that was never the real bottleneck.
  • Skipping the baseline means nobody can prove the project worked.
  • Launching to everyone at once turns small problems into a failed rollout.
  • No owner after launch leaves the system to decay as the business changes.

For more on these failure patterns, see why most AI projects fail.

Frequently asked questions

How long does AI implementation take?

For most growing businesses, the time from the first audit interview to a system in daily use is around a month: one to two weeks for the audit, about a week for the blueprint and two to four weeks for the first build. Larger or more complex systems take longer.

Can we skip the audit if we already know what we want?

You can shorten it, but not skip it. Even when the process is chosen, the audit checks the data, the baseline and who will use the system, which are the points where projects most often fail.

What happens if the blueprint shows the project is not worth it?

The project stops before the expensive part. You keep the process map and the analysis, and can move to the next opportunity on the list or fix the groundwork first.

Who owns the system after launch?

You do. ForceMX hands over documentation and trains your team, and can stay on to maintain and extend the system if you want.

Sources

  1. RAND Corporation, The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed: Avoiding the Anti-Patterns of AI (Ryseff, De Bruhl and Newberry, 2024). rand.org