The Unstuck Flywheel: 3 Friction Points That Stall AI Momentum (And How to Break Through)

Severin Sorensen
4d
4 min read

Many leaders have seen the electric vision for AI: a powerful, self-reinforcing cycle that personalizes customer experiences and streamlines workflows. Yet, this initial excitement often fades within months, replaced by stalled projects and quiet skepticism. The issue is rarely a catastrophic error but a slow death from a thousand small cuts; the vision of a rocket launch meets the frustrating reality of pushing a car through mud. Promising initiatives become stuck not because the vision was wrong, but because of pervasive, unaddressed friction that grinds progress to a halt.

To overcome this, leaders must adopt Jim Collins’ flywheel concept. An AI-powered flywheel creates a virtuous cycle where better data feeds smarter AI, which improves the product, attracts more users, and in turn generates more data. While AI is the ultimate accelerator for this wheel, it can only build compounding momentum if it can turn freely.

Therefore, the leader's role must evolve from being the visionary who provides the initial push to becoming the chief engineer—an obsessive friction detective, constantly seeking and eliminating the drag that holds the organization's flywheel back. This work begins by targeting the three most common and powerful sources of friction holding AI initiatives back.

Friction Point 1: The Data Quality Quagmire

There’s a classic saying in computing: "Garbage in, garbage out." In the age of AI, this has a more dangerous corollary: "Garbage in, gospel out." An AI can take messy, incomplete, or biased data and present it with a veneer of authoritative, machine-generated certainty, leading to deeply flawed strategic decisions.

Excited by AI’s potential, many organizations try to point the technology at their entire data ecosystem at once—a chaotic mix of CRMs, spreadsheets, and legacy systems. The project immediately gets bogged down in a multi-year "data cleaning" initiative that drains momentum and produces little value. The flywheel never even makes its first turn.

The Solution: The "Narrow and Deep" Approach

Instead of trying to boil the data ocean, dramatically shrink the scope. Identify one specific, critical business outcome and focus on the single, highest-quality dataset that can influence it.

For example, a software company wants to build an AI model to predict customer churn. The "boil the ocean" approach would be to try and connect the AI to every conceivable data source: CRM records, marketing emails, support tickets, and financial history. The project would stall for months.

The "narrow and deep" approach is to focus only on the clean, reliable product usage data for their top 20% of enterprise clients. By narrowing the scope, they can build a functioning, valuable predictive model in weeks, not years. This first turn of the flywheel—delivering a tangible win—builds the credibility and momentum needed to tackle the next dataset.

Friction Point 2: The Broken Feedback Loop

Many leaders treat AI as a static asset. They commission a model, deploy it, and consider the project finished. But an AI that isn’t learning is just a fancy algorithm, a depreciating asset whose intelligence is frozen in time. The true power of machine learning is its ability to learn. If you are not actively feeding it performance data, you are leaving 90% of its value on the table.

This happens when there is no mechanism for the AI to understand the outcome of its own predictions or suggestions. It makes a recommendation, a human takes action, and the AI learns nothing from the result. The flywheel spins once but never gains speed.

The Solution: Design Explicit Feedback Mechanisms

From the very beginning, design a system for the AI to learn from its own performance. Your goal is to create a closed loop where every action and outcome becomes training data for the next cycle.

For example, a marketing team uses AI to generate five potential subject lines for an email campaign.

With a broken loop, the marketing manager simply picks the one they like best. The AI learns nothing.

With an explicit feedback loop, the team A/B tests all five variants. The open rates and click-through rates for each subject line are then fed directly back into the AI model. The system now knows that, for this specific audience, variants #2 and #5 were highly effective, while #3 was a failure. The next time the team asks for suggestions, the AI’s output is sharper, smarter, and more effective. That is momentum.

Friction Point 3: The "Last Mile" Adoption Problem

This is the most human—and most underestimated—source of friction. You can have perfect data and a brilliant self-learning algorithm, but if the tool is clunky, disrupts a trusted workflow, or is perceived as a threat, your team will find a way to ignore it. A technically perfect solution that nobody uses is a complete failure.

This often happens when AI tools are developed in a silo by an IT or data science team and then "handed down" to the frontline users. The tool may be powerful, but it requires the user to open a new window, learn a new interface, and change the way they've worked for years. The friction is simply too high.

The Solution: Co-design with a "Super-User" Group

Instead of a top-down deployment, embed the end-users in the design process from day one. Identify a small group of respected team members—your "super-users"—and empower them to co-design the tool. Their mission is to ensure it solves one of their biggest headaches and fits seamlessly into their existing workflow.

For example, a company is building an AI tool to help its sales team score and prioritize leads.

The wrong approach is to build it in isolation and unveil it in a mandatory training session. The team will see it as another administrative task and quietly revert to their trusted spreadsheets.

The right approach is to have three trusted salespeople on the project team. They ensure the AI lead score doesn't live in a separate app but appears directly within the Salesforce or HubSpot contact record they already use all day. Because they helped build it, these super-users become the tool's biggest champions, organically driving adoption across the team.

The Main Takeaway

Building sustainable momentum with AI isn't about the heroic effort of the initial push. It's about the relentless, disciplined work of finding and sanding down the rough edges that create drag. As a leader, your most important role is to shift your focus from the grand vision to the granular reality. Become the chief friction remover for your organization, and you will find your AI flywheel beginning to spin with unstoppable force.