Most AI initiatives fail not because the technology doesn’t work, but because organizations don’t know what problem they are actually trying to solve.
According to the MIT study State of AI in Business 20251 most organizations have implemented at least some AI pilots, but many still mention low impact and no measurable return on AI initiatives.
With all the promises of the AI revolution, we still need a good old value-based approach to make sure that what we do drives results for stakeholders.
A Look of Strategists at AI Initiatives
As it typically happens when a new shiny technology appears, experts are born simply by changing the tagline on their LinkedIn profile. AI is not an exception. With so many experts and so much generic AI-generated content, why should you consider my perspective?
I started playing with AI back in 2000, during my pre-university years, when trying to solve the handwritten recognition problem using neural networks. Winning some junior competition was really nice, but the development of neural networks at that time was limited by compute. My area of interest later switched to IT science, applied mathematics, physics, and then to more business-oriented disciplines, such as performance measurement and strategic planning. AI was never my core expertise (still, I like making some parallels between chain functions, gradients, AI backpropagation, and the ability of organizations to learn).
Today, I look at AI from the viewpoint of a strategist, with a certain background in applied mathematics. LLMs are amazing as a technology, but from the viewpoint of strategic planning, I consider them more like a complex digital transformation project than a miracle-making initiative.
I do help organizations with their AI initiatives, but I do it with a strategy-first approach, not a technology-first one. We typically talk a lot about stakeholders, their needs, their strategic intent, how the vision of the organization is translated into long-term objectives, how those objectives are decomposed into specific goals, and how we make them even more specific and unambiguous with KPIs.
AI is just a part of the puzzle that may or may not fit into how this strategy will be executed.
Somehow, what I do resonates with the overall demand for bringing clarity into the AI domain, so I also share my perspective via conference talks. In 2026, it will be “AI implementation strategy” in Munich, and later in May “Measuring trust in AI” in Vienna.
Before Implementing AI – Do Your Strategy Homework First
The MIT report mentioned above confirms a simple truth:
Implementing AI is easy – creating value with AI is hard.
The perspective of strategists on this is straightforward: before considering any change initiative, make sure you have your strategy properly cascaded and monitored. Without these fundamentals, I don’t think it’s viable to go ahead with any transformation initiative.
Implementation of AI is a good reason to get back to the basics (the needs of stakeholders) and to reflect on the possibilities AI could offer.
Think in terms of limit conditions: how would your organization look if all possible barriers to AI implementation (technology, architecture, compliance, legal, people, etc.) were solved?!
One definition of strategy execution is validating hypotheses in practice. As a part of AI homework, it’s a good idea to formulate those hypotheses. Play with the technology a little bit, do some prototypes to get an idea where the pitfalls might be in terms of implementation, capability gaps, and user expectations.
Let’s discuss certain principles to make AI implementation more successful in terms of creating tangible value for the stakeholders.
Principle 1. Address the Real Challenge – Know the Needs of Your Stakeholders
What’s the difference between technology-first and strategy-first implementation? With strategy-first implementation, you always start with the business context. You know your stakeholders, their needs, your high-level objectives and specific goals, and you try to understand how the new technology will help you execute those goals more effectively – specifically, how it will impact the metrics you are tracking.
This creates a focus on what matters, rather than simply playing with technology.
Good target candidates for AI implementation are:
- Cost metrics
- Time metrics
- Complexity metrics as a derivative of cost, time, and cyclomatic complexity
- Quality metrics (error rate, percentage of returning problems)
- Talent metrics (areas with high turnover rate)
To reiterate the importance of initial strategy decomposition: it should not be “we’ll transform into an AI-first organization.” There should be specific challenges you want to address, with respective stakeholders behind them and clear ownership in terms of execution. This approach resonates strongly with the agile principles we use in software development.
If you insist on reinventing your organization and being AI-first, make sure to start with strategy, the stakeholders, and their needs!
Principle 2. Prepare for Long Run – Think About Architecture Early
I mentioned that I see AI as another digital transformation, a change initiative. But this change initiative is obviously more complex than, let’s say, implementing a CRM.
In this sense, planning architecture for AI is crucial. Take into account:
- How the context and prompts will be maintained;
- How you’ll connect AI to the existing business environment;
- Prepare to orchestrate multiple AI tools;
- Prepare to redesign certain workflows from scratch.
Imagine, for example, that you use AI for responding to users’ questions via a chatbot. The architecture you choose will be defined by questions like:
- What will be the learning loop?
- Will there be human oversight? How will it be implemented?
- How will corrective actions be introduced?
- Will AI have access to previous dialogs with the same user?
- Will AI be able to fetch data directly from the CRM?
- What security mechanisms will be implemented?
The ability of AI to learn, remember context, and improve will be a leading factor of the adoption rate over time. Make sure the architecture you choose for AI implementation supports these learning needs.
Principle 3. Making Quality and Compliance Meaningful for Stakeholders
AI touches too many sensitive points of an organization – access to customer data, working with third-party tools, supporting decision making, communicating with users, and keeping data for possible audits.
At a certain scale of AI implementation, quality and compliance controls are a must.
We are moving into the GRC domain, but again, it’s not about AI. It’s about your strategy, what risks AI imposes on it, and how we can prevent and mitigate them.
We hear a lot that for AI implementation there should be:
- Human in the loop,
- Audit trail,
- Explainability,
- …
What’s missing in practice is the connection between these ideas and what stakeholders actually care about.
I find bowtie risk analysis method to be well suited for this role. Do it for a central risk event, defining threats with respective preventive controls, as well as the consequences of the risk event with respective mitigation controls.
We discussed an example of such analysis in the “AI implementation in a medical quality control” case2 presented at OOP. In that case, the central risk event was formulated as “AI-validated results are approved without proper human review.”
Using risk prevention and risk mitigation controls, we aligned the AI implementation with the quality and compliance concerns of the stakeholders. If scaling this idea, the same controls will help to build a comprehensive AI governance framework. When scaling this approach across the organization, the same controls can be used to establish a comprehensive AI governance framework.
Executive Summary: Shifting AI from Technology to Strategy
Follow these principles for the strategy-first AI implementation:
- Focus implementation on specific stakeholder needs; ideally, quantification of actual vs. expected outcomes should be defined.
- AI implementation is a complex learning system, not a one-time connection to the API of an LLM – plan the architecture accordingly.
- Establish quality and compliance controls and communicate them to stakeholders; this will define future acceptance of AI implementation. The bowtie method has proven to be a great tool for this purpose.
- The GenAI Divide State of AI in Business 2025, MIT NANDA, July 2025 ↩
- “Strategy and Oversight of AI Implementation in Medical Quality Control,” BSC Designer, Alexis Savkín, 2025. ↩
Alexis Savkin is a Senior Strategy Consultant and the CEO of BSC Designer, a Balanced Scorecard platform. He has more than 20 years of experience in the field, with a background in applied mathematics and information technology. Alexis is the author of the “Strategy Implementation System”. He has published over 100 articles on strategy and performance measurement, regularly speaks at industry events, and his work is frequently cited in academic research.