We had little time to perform a strategic review of our client’s innovation lab. The analysis was part of a broader enterprise strategy program, and I needed to work with the lab’s leadership team to provide an update that would feed into the broader assessment.

The lab originally had been established with a clear business model: It would disrupt the insurance value chain, capturing value that increasingly flowed through to intermediaries, and in doing so generate meaningful revenues for their parent organization (in the region of hundreds of millions of dollars annually) after an initial ramp-up period.

Early results signaled that it would take longer than expected to yield meaningful returns. Internal feedback shared with the lab suggested a shift toward building and partnering to create platform-based InsurTech services. When that didn’t prove fruitful either, the lab was steered toward incrementally broader topics—data brokerage, blockchain—before settling into a venture capital-style approach of placing many small bets in the hope that one would strike big and realize the returns that had been promised at the outset.

Now it was year five and the lab was nowhere close to achieving their target. It wasn’t clear how they had missed expectations so significantly. The answer lay in the sequence of decisions that gradually separated the business reality from the assumptions that supported its original model and revenue ambitions.

A Slow Drift

We approached this challenge by breaking down the key strategic decisions, both to formalize a storyline of how the lab had developed and to isolate the key moves so they could be viewed independently of each other. We found that the revenue expectations set at launch were plausible based on a clear set of assumptions. If all these assumptions held, and the business followed the path mapped at the outset, it was plausible that the lab could have realized the predicted results.

That meant that a lot needed to go right and assumed limited contextual variation. As it happened, changes crept in early. With each year, the annual review process provided the lab’s leadership team with feedback that was not strikingly different from year to year. Stand up a platform. Explore the data opportunity. Spread the bets more widely. Once adopted, however, the feedback incrementally steered the lab from its original course.

As each year brought new adjustments, the magnitude of the shift compounded, until after five years the lab’s mindset had shifted—unintentionally and largely unnoticed—from industry disruptor to quasi-VC, while remaining accountable for the original revenue expectations. This realization provided the basis for a complete re-evaluation of the lab’s strategy and business model.

The pattern of adopting incremental changes that steer a business from its original strategy is clear in hindsight. But as anyone who has run a business understands, it’s difficult to establish this perspective when you’re in the middle of it. Before exploring how leaders and their teams can cultivate this view, first we’ll explore the underlying mechanics of the challenge.

The Experience Paradox

Exceptional leaders set a vision and rally their team to execute. They make effective use of their position to identify the limits of their strategy and infer what may be coming next. In a different context, analyzing the history of scientific revolutions in his 1962 book “The Structure of Scientific Revolutions,” Thomas Kuhn observed that major breakthroughs leading to paradigm shifts—entirely new ways of interpreting a field—begin with the detection of anomalies, or instances where the prevailing interpretation no longer explains the data. Detecting an anomaly is itself an act of expertise. It requires understanding the prevailing interpretation deeply enough to recognize where it’s breaking down. The same holds true for leaders embedded within their organizations. Deep experience can help identify where a new interpretation is required.

Paradoxically, the same depth of experience can result in blind spots that lead us to miss such anomalies. In order to understand why this is so, we can think about our brains as resource-management systems. Given that the brain is responsible for managing many different responsibilities that require energy to perform, it follows that energy conservation is a priority—what cognitive scientists Falk Lieder and Thomas Griffiths describe as the optimal use of limited computational resources. (“Resource-rational analysis: Understanding human cognition as the optimal use of limited computational resources,” Cambridge University Press: February 4, 2019, https://doi.org/10.1017/S0140525X1900061X)

Perception is one place in which this economizing is most evident. When interpreting our environment, our brain operates like a prediction engine, drawing on models informed by memory, expectations and assumptions to anticipate what’s likely to be perceived next, and then validates these predictions against incoming sensory data. Expectation and habit do much of the heavy lifting, and unless our interpretation identifies data that are markedly different from our prediction, we may overlook important details. If we’re not attuned to anomalies that suggest the prevailing interpretation is breaking down, we may unconsciously overlook these signals.

Consider driving a familiar route. How often have you arrived only to realize you’d been daydreaming for the past five minutes and been only vaguely aware that you were driving? Imagine in contrast if you were driving while rigorously attending to your car’s distance from the curb, your hands on the wheel, your indicator timing and your following distance, all at once. It would be clunky and mentally taxing, not to mention risky. Prediction is what allows the experienced driver—or the experienced claims handler, underwriter or operations leader—to act fluently and at speed.

Our brains are well attuned to abrupt surprises. Think about how quickly you snap to attention when a car pulls out in front of you unexpectedly. But when change is gradual, discrepancies can slip by unnoticed. This is particularly true when new information appears to align with our predictions but is in fact suggesting something new.

It’s precisely this situation that arises in complex operating environments—repair and replacement costs that outpace filed rates a little more each year, or a commercial book that drifts from its intended appetite one underwriting exception at a time—where expert teams rely on practices that have worked to date and are unexpectedly caught out.

Returning to the example of the innovation lab, it explains how it’s possible to take all the right actions in the moment and still miss the underlying shift.

Reframing Like a Fox

Philip Tetlock’s research on expert judgment, set out in his 2005 book “Expert Political Judgment: How Good Is It? How Can We Know?” is helpful when exploring techniques to loosen the grip of a prevailing business narrative. He uses the analogy of the fox and the hedgehog, originally developed by Isaiah Berlin, built on the idea that the hedgehog knows “one big thing”—a holistic narrative that contextualizes and informs data—whereas the fox “knows many things.” The fox essentially gathers fragmentary points and evaluates them as a constellation—exploring the relations between the data points rather than trying to formalize them along a single narrative thread.

Both approaches have value in the right context, but when we aim to predict outcomes, or conversely when we seek to review how we’ve arrived at a certain point, adopting a constellatory approach—reframing like a fox—can help us to see patterns in the data with less bias.

Reframing like a fox means staying open to weak signals without premature judgment and assigning to each a rough probability of its being true or widely accepted. These signals could be offhand remarks from agents, a slow rise in complaints about claims turnaround or subtle hesitations in renewal conversations. While we may be inclined to dismiss these fragments, they could prove to be a loose thread that, once pulled, reveals an underlying pattern.

By cataloging and reviewing such signals periodically, paying special attention to those that contradict our assumptions or do not align with our prevailing hypothesis, we can build a more balanced and adaptive perspective.

This is the practice the innovation lab needed. The decomposition we performed was itself an exercise in reframing like a fox, applied after the fact. The same discipline, applied proactively, catches the drift before it compounds. Had the team built it into their strategic reviews, the drift may have surfaced earlier. It would have meant treating the original mandate as a documented starting position—a prior estimate, with its underlying assumptions made explicit—and treating each annual review input as a discrete signal to be weighed and considered. Every recommendation to shift focus, stand up a platform or follow an emerging trend would have been evaluated as a weighted adjustment from a starting position, and the cumulative size of those adjustments could have been quantified.

Put simply, it would have made it easier to critically evaluate the perceived value of the feedback and the impact of adopting it upon the business. The weighted feedback tool described in the accompanying sidebar below this article formalizes this discipline.

(Editor’s Note: For readers interested in a tear sheet, the sidebar is also published separately in the linked article, “How to Reframe Like a Fox.”)

I designed the weighted feedback tool for teams to evaluate prototypes ahead of a business sponsor presentation, but the underlying steps—document the prior, weigh incoming signals by the strength of their evidence and cap how far any single round of feedback can move the estimate—apply equally well to an annual review. Notably, the tool caps the total adjustment at 20%. Had the lab observed a similar cap, the moment their accumulated shifts exceeded it would have been the moment to pause and re-evaluate the strategy itself rather than adopt another course correction. To make this concrete, the sidebar closes with a worked example drawn from the lab’s situation—how the tool would have handled the year-two recommendation to pivot toward platforms.

This is the fox’s discipline applied to one specific arena: stakeholder feedback. Where the hedgehog hears a single story (“leadership wants a pivot”), the fox sees a constellation of individual signals, each weighted on its own evidence, and reads the pattern in their relationships.

The disciplines of attending to perceptual bias and reframing like a fox are relevant to insurance leaders today. As carriers begin delegating operational judgment to agentic AI systems, it’s worth remembering that these systems are very much prediction engines trained on the past. As such, they inherit a version of the very filtering challenges described here.

The human capacity to notice what doesn’t fit—anomalies in deeply understood data sets, or patterns in fragmentary data when compared with holistic narratives—becomes increasingly valuable as more routine work is automated.

***

Weighted Feedback Tool

A structured way to break feedback into discrete signals, weigh each on the strength of its evidence and track how far, in total, they move you from your starting position.

This was designed for teams testing a prototype with stakeholders but is equally applicable to strategic reviews.

Before feedback collection:

• Write down your estimated likelihood that the solution you’ve developed will solve the challenge at hand and assign a confidence level to that number (i.e., + or – x%). Document your reasoning. This becomes your prior estimate.
• Calibrate the prior against what you’ve learned so far, adjusting by no more than + or – 2% and documenting your rationale. This becomes your calibrated prior estimate.

During feedback collection:

• Treat each stakeholder as starting from a neutral belief of 50% in your solution’s success. This becomes the reference point for their perspective.
• Score each feedback signal by the strength of its evidence, applying larger adjustments for strong evidence and smaller adjustments for weak evidence. Treat subjective reads such as tone and body language more conservatively than what is said outright.
• As scores approach the extremes, progressively reduce each adjustment to reflect diminishing returns, preventing scores from reaching 0% or 100%.

After each session:

• Revisit your calibrated prior alongside the stakeholder scores and adjust it according to the strength of the evidence: + or – 12% where evidence strength was high, + or – 5% where it was moderate, and + or – 1% where it was low. These adjustments are conservative to prevent a single session from producing disproportionate shifts. This becomes your posterior estimate.
• Cap the total adjustment at 20%. If the evidence keeps pushing beyond the cap, that’s not a scoring problem. It’s a signal that the solution, or the strategy behind it, needs to be re-examined.
• Based on the outcome, review your solution to determine what, if anything, ought to change.

A Practical Example:

Consider the innovation lab preparing for its annual review. The team documents its prior estimate as 70% confident that the original model will reach its revenue milestones, with a confidence level of + or – 10%, along with the reasoning to support that view. First-year traction has been slower than planned, so the team calibrates downward by the maximum 2%, resulting in a calibrated prior of 68%.

Feedback is gathered from senior stakeholders as part of the review process and documented before being shared with the lab team. One senior executive points out in their feedback that the market is shifting toward platform-business model development (partnering to create platform-based InsurTech services) and queries whether the lab has developed anything here. They raise the point that the lab risks missing the opportunity entirely. If the disruptor model isn’t showing signs of progress, they suggest, it may warrant a shift in direction and investing more time in exploring building or partnering to realize a platform opportunity.

The team independently scores this point of feedback starting with the neutral 50% reference point. The executive’s statement is direct and unambiguous, making it high-strength evidence, and its direction is negative for the current model.

Following the session, the team adjusts their calibrated prior accordingly—by the corresponding -12%—giving a posterior estimate of 56%.

Critically, the team then works through every other point of feedback in the same manner. Each point of stakeholder feedback is treated as a distinct data point to be scored on its own evidence, rather than folded into a single holistic story (i.e., “the senior leadership team recommends a shift in direction”). Some points will move the estimate down, others up and many barely at all.

The conservative adjustment bands, together with the 20% cap, ensure that no single voice swings the estimate on its own.

This is the fox’s discipline applied to one specific arena: stakeholder feedback. Where the hedgehog hears a single story (“leadership wants a pivot”), the fox sees a constellation of individual signals, each weighted on its own evidence, and reads the pattern in their relationships.

Viewed this way, two things become visible in this example:

First, the original model still holds majority confidence. The platform suggestion is a signal to catalog and monitor, not a mandate to pivot.

Second, the cumulative adjustment already stands at 14% against the 20% cap. The moment accumulated feedback breaches that cap is the moment to deliberately re-examine the strategy itself, rather than trigger a course-correction.

Featured image: AI-generated (ChatGPT)