How to Develop an AI Strategy: A Practical Guide

This is the fourth article in a series on what it takes to adopt AI in a way that is practical, secure and genuinely valuable. Over the coming posts, we will explore the strategic, technical and organisational foundations that help organisations move from isolated experimentation to lasting business value. If you missed the previous articles, you can find them here: 

https://skillfield.com.au/blog/ai-adoption-treat-it-like-transformation-not-tool-deployment/

https://skillfield.com.au/blog/the-three-objectives-of-a-practical-ai-strategy/

https://skillfield.com.au/blog/ai-adoption-turning-strategy-into-execution/

If I look at the organisations I speak with every week, most of them are already “doing AI”. They have Copilot licences, a few ChatGPT power users, maybe a pilot chatbot, and several enthusiastic teams building agents and workflows on their own. What most of them lack is adoption at scale, measurable return on investment, or a clear answer to the question: why are we doing any of this?

This is the pattern I keep seeing. Experimentation everywhere, value nowhere. Different parts of the business doing different things with different tools, driven more by pressure and fear of missing out than by intent. There is no shortage of research showing that many AI projects never make it past the pilot stage, and my own conversations reflect the same pattern.

The way out of this is not more experiments. It is to get back to basics and approach AI strategically. In my previous posts I argued that AI adoption should be treated as a transformation program, not a tool deployment, and that a practical AI strategy has three objectives: enabling AI across the organisation, using AI to support the business strategy, and using AI to redefine the business model. I also covered the foundations needed to turn strategy into execution.

This post answers the question that naturally follows: how do you actually develop the AI strategy itself?

The good news is that strategy development is not new. We have been doing it for cyber security, for digital transformation, and for business generally, for decades. The process below applies proven strategy development practice to AI, augmented with AI-specific frameworks such as the MITRE AI Maturity Model, the Google Cloud AI Adoption Framework, and the Australian Government’s Guidance for AI Adoption.

The process has two phases: Discovery and Analysis, then Strategy Development. Do not skip the first one. The most common reason AI strategies fail is that they were written before the organisation was understood.

Phase 1: Discovery and Analysis

The goal of this phase is to develop a deep understanding of the business, so you can articulate how AI will help the business meet its goals. AI is an enabler of the business strategy, not a strategy on its own. You cannot enable something you do not understand.

1. Understand the Purpose,Vision and Mission

Start at the very top. Every organisation exists for a reason (purpose), aspires to something (vision), and has a way of getting there (mission). Your AI strategy must serve all three, and you should be able to show the connection explicitly.

A simple exercise I recommend is to take the organisation’s existing statements and articulate how AI extends them. 

This exercise looks simple, but it forces an important discussion: does AI change what we are, or only how we work? The answer shapes everything that follows.

2. Review the Corporate and Business Unit Strategies

Next, get your hands on the corporate strategy and, where they exist, the strategies of individual business units. Read them carefully. Your job in this step is to understand what the organisation has already committed to, because that is where AI investment should be anchored.

For every strategic objective, ask: how can AI help achieve this faster, cheaper, better, or with less risk?

If the strategy says, “expand into two new markets”, the AI question is not “where can we use a chatbot?” It is “what would market expansion need — demand modelling, localised customer service, pricing analysis — and which of these can AI meaningfully accelerate?”

A useful tool here is a strategy map: a simple cascade that connects your future AI capabilities through internal processes and customer outcomes to financial results. If an AI initiative cannot be traced upward to a strategic objective on this map, it is a hobby, not a strategy.

3. Align Core Values with the AI Objectives

Your organisation’s core values describe how work gets done and act as a decision-making framework. AI will test those values in new ways, so make the alignment explicit. If one of your values is “respect everyone”, what does that mean for how AI handles customer and employee data? If it is “take informed decisions”, what does that mean for relying on AI outputs without verification?

Work through each value and write one or two sentences on what it demands of your AI adoption. This becomes the seed of your responsible AI principles, and later your AI policy. It also answers the ethics question I raised previously: just because AI can do something, should it? Your values are where that answer comes from.

4. Historical Analysis: What Has the Organisation Done with AI So Far?

Before planning the future, understand the past. Build a simple timeline of everything the organisation has done around AI and its enabling capabilities: tools procured, pilots run (successful or not), policies written, data platforms built, training delivered, and the informal or “shadow AI” usage that grew organically.

This analysis serves three purposes. It reminds management of the work already done and gives credit to the teams behind it. It surfaces lessons from pilots that stalled, so the strategy addresses the real blockers rather than assumed ones. And it gives you an honest inventory of the ungoverned AI already in use, which is almost always larger than leadership expects. Remember: if you have not enabled people to use AI properly, they are very likely using it anyway.

5. Assess AI Maturity Using the MITRE AI Maturity Model

Now measure where the organisation actually stands. My recommended instrument is the MITRE AI Maturity Model and Organizational Assessment Tool. It assesses the organisation across six pillars — Ethical, Equitable and Responsible Use; Strategy and Resources; Organisation; Technology Enablers; Data; and Performance and Application — broken into 20 dimensions, each rated on a five-level scale from Initial to Optimized.

The assessment does two things for your strategy. First, it gives you an evidence-based baseline: a picture of your strengths and gaps across ethics, governance, culture, skills, platforms, data and adoption, rather than a gut feel. Second, and more importantly, it gives you a roadmap. The model describes what the next level of maturity looks like for each dimension, so your strategic initiatives can target specific, measurable maturity uplifts. Note that the highest level is not the goal for everyone; the right target maturity depends on your mission and business, and part of the strategy work is deciding what “mature enough” means for you.

Run the assessment with a cross-functional team, not just IT, and plan to repeat it periodically to measure progress. If useful, complement it with the Google Cloud AI Adoption Framework (which looks at maturity through Lead, Learn, Access, Secure, Scale and Automate themes) and, for Australian organisations, the National AI Centre’s Guidance for AI Adoption and AI Impact Navigator.

6. Build a Stakeholder Register

An AI strategy lives or dies on its stakeholders. A stakeholder is anyone with a key interest in the strategy: the executive team, board, business unit leaders, IT, security, risk and legal, data owners, HR, and increasingly employee representatives, because AI touches how people work in a way few technologies do.

Meet each of them and record their power, their level of interest, what they want from the AI strategy, and what concerns them. A simple register looks like this:

Treating these as a shared layer rather than the property of one stream has a practical benefit: you build each capability once and every initiative draws on it, instead of every project reinventing its own governance, its own data fixes, and its own security review. When you document the strategy, show the three initiative streams sitting on top of these six foundations — that one picture explains the whole operating model. 

For every initiative and sub-initiative, document the objective it supports, the value it creates (improved willingness to pay, reduced cost, or increased volume), the owner, the dependencies, and the maturity dimensions it uplifts. 

3. Identify Key Success Metrics and Budget 

A strategy without measurement is a wish. Define metrics at three levels: 

• Business value metrics: the ultimate test. Cost reduction achieved, revenue influenced, customer satisfaction, cycle time. Tie each strategic initiative to at least one. 

• Adoption metrics: usage rates, active users, use cases in production versus in pilot, the ratio between them, and the health of the use case pipeline itself: ideas captured, assessed, and time from idea to production. If pilots pile up and production stays flat, the strategy is not working. 

• Maturity and risk metrics: progress against the MITRE maturity baseline on your next assessment, incidents and near misses, percentage of AI use that is governed versus shadow. 

Then allocate the budget deliberately across the three objectives rather than letting it accumulate accidentally across disconnected projects. Fund the cross-cutting foundations first as a shared investment (they de-risk every initiative in every stream), weight the majority of the remaining investment toward strategy-supporting initiatives, and ring-fence a small amount for the business model bets. Include the ongoing costs that organisations habitually underestimate: consumption, monitoring, retraining, and the people needed to operate AI, not just build it. 

4. Bring It Together and Keep It Alive 

Package the strategy so it can be communicated: the vision, the strategy map, the initiatives with owners and timelines, the metrics, and the budget, ideally with a one-page version for the board. Then treat it as a living document. AI is moving faster than any technology wave before it, so review the strategy at least every six months, re-run the maturity assessment annually, and be willing to stop initiatives that the metrics say are not creating value. 

Conclusion 

None of these processes are exotic. Understanding the business, analysing the environment, assessing maturity, engaging stakeholders, and building initiatives with metrics and budget. This is how good strategies have always been developed. What is new is the subject matter, and the frameworks now available to support it: the MITRE AI Maturity Model for assessment and roadmapping, adoption frameworks from Google and the Australian National AI Centre for structure, and screening and policy tools for governance. 

The organisations that will win with AI are not the ones running the most experiments. They are the ones that stop, do the discovery and analysis properly, and build a strategy that connects AI to the business it is meant to serve. If your organisation has been experimenting for a year with little to show for it, that is not a technology problem. It is a strategy gap, and it is entirely fixable. 

Author:Mouaz Alnouri