From AI Ambition to Implementation: The First 90 Days

Explore the series.

By the end of an initial 90 days, an executive team should have evidence for its next AI investment decision. That may include an implementation operating within an agreed scope. It may also include a decision to narrow an ambitious idea because the information, economics or controls don’t support it.

The first period of an engagement needs to create that evidence while implementing the agreed priorities. An assessment that stops short of implementation leaves the business waiting. A rushed launch can create ongoing work that the original business case didn’t account for.

The sequence below shows how an AI leadership and implementation engagement could work. It is a planning example, not a fixed service package or a promise that every business can reach production within 90 days. Procurement, access and the consequences of failure can change the pace. Several activities should run together once their dependencies are understood.

Begin with the business decision

During the opening weeks, the AI leader needs to understand what the executive team wants to change. Improving operating margin is different from responding faster to customer enquiries. Each calls for different evidence and different choices.

Define the problem in terms you can investigate. If proposals are slow, find out where time goes. Staff may spend hours assembling material, but the larger delay could be waiting for commercial approval. Automating the first activity might make the work easier without materially changing the date a customer receives the proposal.

Walk through the current process with the people who perform it. Include the exceptions, the unofficial spreadsheet and the person everyone calls when a record is wrong. These details help explain why the documented process may differ from what the business relies on.

At the same time, establish the engagement’s mandate. Agree which decisions the fractional AI leader can make, which need executive approval and how disagreements will be resolved. Allocate internal capacity. If the process owner and technology team cannot participate, the delivery plan needs to reflect that constraint.

Record the business problem, its current performance and the decisions needed to proceed. Include the measurements and assumptions so executives and delivery teams can question them.

Examine the systems and information early

A promising use of AI can fail a basic readiness test. The information may be inaccessible, poorly maintained or subject to restrictions that rule out the proposed use.

Take an illustrative professional services business that wants help preparing proposals. It has previous proposals, a service catalogue and commercial terms. Before connecting an assistant, the team needs to know which material is current, who can approve changes and which previous proposals contain information that shouldn’t be reused for another customer.

The technical assessment should establish where that information lives and how the proposed tool would access it. A demonstration based on a few uploaded files doesn’t resolve integration with the working document system. Nor does it show that a departing employee’s access will be removed or that restricted material will stay restricted.

Some preparation can happen alongside a bounded trial. The service owner might approve a small collection of current material while the broader catalogue is being cleaned up. That provides a way to test that part of the process without pretending the entire information estate is ready.

Record the dependencies with someone responsible for each. If access approval is pending, show which work can continue and which cannot. The AI leader should bring the recommended next action to the executive team when a decision exceeds their authority.

Select an initial implementation on both value and feasibility

By the end of the first month, aim to have enough evidence to choose an initial use or explain why more investigation is necessary. Avoid selecting it solely because it is easy to demonstrate.

Compare the likely commercial benefit with the full effort required. Include integration, process redesign and ongoing supervision. Consider how many people perform the process, how often it occurs and whether the expected improvement addresses the cause of delay or cost.

Then test the technical and operating assumptions. Can the necessary information be used? Can the output be checked at a reasonable cost? Would a failure be detectable before it causes significant harm? Can the team keep working if the tool is unavailable?

For the illustrative proposal process, the initial scope could be preparing a draft from approved service material, with an employee checking every proposal before it goes further. Negotiating terms or sending a proposal would remain outside that scope. The trial would then measure drafting and review, without confusing those results with negotiation or sending.

Compare the idea with alternatives. Existing templates, better search or a simpler approval process might address part of the problem at lower cost. The investment decision should survive that comparison.

Agree what success and failure will look like

Define the baseline before the new process starts. Measure the work from beginning to end, including checking and corrections. If you measure only the time needed to generate a draft, you can miss extra work created elsewhere.

For a proposal assistant, the measures might cover preparation time, material corrections and the proportion of eligible proposals completed through the new process. Track approval delays separately so the report doesn’t attribute every change to AI.

Set the quality conditions that time savings cannot override. An unauthorised commitment in a proposal needs a different response from awkward wording. Decide how defects will be classified, who reviews them and which findings would stop or narrow the trial.

The business owner and finance team should agree how any released capacity could create value. Will it reduce overtime, increase throughput or allow staff to spend more time on complex work? Treat those as assumptions until the operating evidence supports them.

Write down the decision the trial is meant to inform. For example, the business might need to decide whether a limited proposal assistant reduces preparation effort enough to justify extending it to another team. That is a more manageable question than whether AI can improve the entire sales function.

Build and coach together

During the middle part of the period, the delivery team can configure or build the agreed workflow, test representative cases and introduce it to a bounded user group. The AI leader remains responsible for coordinating the work and dealing with the issues that cross departmental lines.

Training should begin before the final release. Users need to understand what information the tool can use, how to check its output and when to revert to the existing process. Working with early drafts also gives them a chance to identify assumptions that the implementation team missed.

A user might find that the assistant assembles a sound service description but confuses two commercial options. The response could involve improving the source material, changing the way options are selected or requiring staff to check the selected option. Repeating general prompting advice would leave the underlying problem unresolved.

Keep a record of changes and retest the affected cases. Ask managers whether the new steps fit normal work. A technically sound process that requires more attention than the team can provide needs further design work.

Use the final weeks to make the next decision

Near the end of the initial period, compare the evidence with the agreed baseline and conditions. Include unsuccessful uses, rework and support effort. Explain which results are measured and which remain forecasts.

The decision may be to expand, continue within the current boundary, repair a weakness or stop. An implementation needs an operational owner, a support arrangement and a way to detect problems after the initial team moves on. A stopped initiative should record an account of what was learned and why further spending isn’t justified.

The executive review should answer four questions. What changed in the business process? What did it cost? What evidence supports the next investment? Who will carry the work forward?

For an AI implementation partnership, that review is part of the delivery responsibility. Use the findings to decide the next scope of work. If you’re considering an engagement, bring one priority and the process behind it. We can use that to discuss what the first 90 days would need to establish in your organisation.

Questions

Written by Todd Johnson

Founder of InnovateX Solutions. Todd provides fractional CIO, CTO and Chief AI Officer leadership for Australian businesses and government, and wrote the XAF Connected Architecture framework.

Todd on LinkedIn