Stop shopping for AI tools. Start building your AI operating muscle.
Move beyond permanent pilots.
Until you build your AI operating muscle, your AI strategy is just an expensive experiment. Tools don’t create value; operating change does. We lay out five concrete moves to build your AI operating muscle.
How many of you recall the MIT report, The GenAI Divide: State of AI in Business 2025? Media headlines turned research findings into clickbait: 95% of generative AI (GenAI) projects fail. Some of us, like me, thought it was an absurd claim, and others agreed that the promise of AI is too lofty. Whether some GenAI projects fail or returns are zero, the most important insight is that AI is being adopted quickly but integrated slowly. And that gap between experimentation and operations is where the friction is.
Most coverage has focused on individual tools like ChatGPT, Microsoft Copilot, Gemini, and Claude, which are widely used in organizations to improve individual productivity. And, they are needed. Yet, individual tools don’t always turn into overall business outcomes. Productivity isn’t necessarily the same as outcomes. Outcomes require workflow integration, process (re)design, and learning loops at scale. Most GenAI projects are scattered, some as shadow AI and others at the individual or unit level as one-offs or pilots. As a result, organizations end up with brittle workflows, and the value of GenAI may not be captured or reported.
When it comes to ROI, the MIT report shows it is mostly in back-office deployments, especially in operations, finance, and procurement. These deployments are also where day-to-day work sits within repeatable, measurable workflows. A related report from Google shows a similar pattern of early adopters. Most (88%) agentic AI early adopters are seeing positive ROI from GenAI. With 2025 being the year of AI agents, “co-Intelligence” is becoming real (an Ethan Mollick term), where agents support behind-the-scenes work for people. So, how do we ensure greater success? Here are our five recommendations for building your AI operating muscle.
1. Pick the right use case.
We have no shortage of ideas, and the possibilities for applying AI are virtually limitless. We recommend a simple decision lens to evaluate use cases on three dimensions:
Business Value: where the friction lives, and value is created
Feasibility: cost, time, complexity, and people’s readiness
Data Readiness: reliable end-to-end or fragmented across the organization
A quick clarification on data readiness, because it is easy to oversimplify. The question is, can we reliably use the data (assuming data quality is an inherent practice here) end-to-end in workflows without manual steps? It is not simply about whether the data is available, especially in a time when everything is data-driven by AI. In practice, ‘data-ready’ means you can move from idea to pilot quickly, and data not ready means your pilot becomes a permanent pilot.
2. Start with people when you are mapping out processes.
As many of you have experienced over the years, we have all lived through zombie projects, where IT initiatives were adopted for the sake of technology, or a “we need an app” mentality has not improved organizations. Tech rarely fixes broken workflows; the process (re)design has. And the process always starts with people.
If you can’t answer the question, "What is in it for me?" you don’t have a case for workflow improvement. And, if you can’t answer “Who owns the outcome?” you don’t have governance. You just have hope for transformation, usually an expensive one.
3. Invest in an AI learning culture.
Our view of AI is more than efficiency; it is about building capacity for people, processes, and organizations. Think cognitive skills and decision-making at scale. Yet, GenAI behaves differently from traditional automation. The latter is deterministic, same input, same output. GenAI is probabilistic unless constrained with data, rules, and controls. That means people need a clear understanding of how it works and how it is utilized. Especially when to trust it and when not to.
As Microsoft’s Satya Nadella puts it into perspective: move from a know-it-all culture to a learn-it-all culture. The days of “set it and forget it” are over; it is time for a culture of creativity, challenge, and community. Engage your teams in the processes and reward them for their efforts.
4. Buy versus build.
The MIT report identifies success around organizational design and practicality. Clearly, there has to be internal accountability. But time-to-market matters as consumer behavior shifts in how they interact with machines. In that manner, procuring external tools and co-developing with suppliers can be more effective, less costly, and less time-constrained than internal development. The latter is effective when differentiation is critical and constraints are present.
The good news is that the current infrastructure is moving toward more modular, interoperable approaches. Emerging protocols and integration standards are reducing the need to build everything from scratch.
Our recommendation for resource-limited organizations is to buy rather than build. But don’t turn it into a shopping spree. Your decision should come down to speed, risk, differentiation, and ownership. For example, you don’t want to be the entity that is trying to build its own GenAI-powered CRM platform with no progress over the last three years (based on a real case).
5. Ethics is a product feature.
AI is advancing faster than most policies can keep up with. Responsible development starts when strategy, values, and regulatory standards are aligned from day one. In this realm, ethics is a long-term commitment. It is a product feature, not an afterthought or contractual language. Non-negotiable.
Monitor and implement steps to reduce risks by creating red teams to address potential misuse, leveraging diverse-skilled teams and life experiences, and defining risk scenarios. Build guardrails around where AI is autonomous, what it can access, when people are in the loop, and who owns the outcome. Think systems across talent, teams, and activities.
. . .
AI projects don’t fail. It is how we approach AI that leads to a stall, especially when the speed of change outpaces adoption.
So, don’t just buy tools. Build Your AI Operating Muscle. Pick the right workflow, map out processes around people, nurture and reward a learning culture, and scale responsibly from day one to achieve business results without permanent pilots.
