The Confidence Illusion

The number that stopped the room

We asked the executive team of a mid-size Australian insurer a simple question: on a scale of 0 to 100, how confident are you that your top five strategic initiatives are on track?

The average answer was 74.

Then we showed them the evidence-based assessment. We had spent four weeks tracing each initiative’s actual progress against its stated thesis, measuring evidence accumulation, checking whether underlying assumptions still held, and mapping interference patterns that were distorting execution.

The evidence-based confidence score for the same five initiatives: 28.

Forty-six points. The gap between what the leadership team believed and what the evidence supported was 46 percentage points. The room went quiet.

This wasn’t an outlier. We ran the same assessment across four enterprises, two in financial services, one in insurance, one in healthcare technology. The average gap was 41 points. The narrowest was 34. The widest was 46. Every executive team was substantially more confident than their data warranted.

Why this matters

Executive confidence drives resource allocation. When leaders believe an initiative is working, they continue investing. They defend its budget during reallocation conversations. They cite it as evidence that the strategy is on track. The confidence itself becomes a form of organisational momentum that’s very hard to interrupt.

If that confidence is well-calibrated, this is fine. Leaders should invest in things that are working. But if the confidence is systematically inflated, if executives consistently believe things are going better than they are, the organisation keeps pouring resources into failing work while starving the adjustments that could save it.

We found four mechanisms that drive the inflation. They’re structural, not personal. Good leaders in well-run organisations still fall prey to them, because the mechanisms are embedded in how organisations report, summarise, and communicate.

Mechanism 1: Narrative smoothing

Raw operational data is messy. Customer acquisition numbers spike and dip. Conversion rates fluctuate. Team velocity varies week to week. When this data gets packaged for senior leadership, someone has to make sense of it. That process of making sense, of constructing a narrative, systematically smooths out the rough edges.

We traced reporting chains for 23 metrics across two of our four enterprises. At each management layer, the narrative became more coherent and more optimistic. A product team would report: “Conversion dropped 12% this week, likely due to the payment flow bug identified Thursday. Fix deployed Friday, monitoring next week.” Their director would report: “Conversion experienced a temporary dip due to a technical issue, now resolved.” The VP’s summary to the executive team: “Conversion metrics stable with normal weekly variation.”

The product team’s report was accurate and specific. The VP’s summary was technically not false. But the signal, that a significant technical issue had caused a meaningful business impact, had been smoothed into background noise.

This happened consistently. In 17 of the 23 metrics we traced, negative signals were reframed as temporary, contextualised as normal, or qualified into insignificance by the time they reached the executive layer. Nobody was lying. Each individual summary was defensible. But the cumulative effect was an executive view of reality that was persistently, systematically rosier than the operational reality underneath it.

Mechanism 2: Aggregation bias

Executives deal in aggregates. Average customer satisfaction. Total pipeline value. Overall delivery velocity. Portfolio health scores. These aggregates serve a purpose: they make a complex organisation legible at the leadership level.

But aggregates destroy variance. And variance is where the important signals live.

One of our financial services partners had a strategic bet focused on cross-selling into their small business segment. The aggregate metric looked healthy: cross-sell revenue was growing at 8% quarter over quarter, ahead of the 6% target. The executive team rated this initiative at 81% confidence.

We disaggregated the data. The growth was being driven almost entirely by two product lines in one geographic region. Four of the six target product lines showed flat or declining cross-sell rates. Two regions were in negative territory. The “8% growth” was real, but it was masking a pattern that contradicted the bet’s thesis, which assumed broad-based cross-sell improvement driven by a new platform capability.

The platform wasn’t driving the growth. Regional relationship managers in one high-performing office were. When we showed the disaggregated data to the executive team, the confidence score dropped from 81% to 39%. The aggregate had been telling a story of success. The components were telling a story of a bet that wasn’t working the way it was supposed to.

Mechanism 3: Temporal displacement

This is the most psychologically comfortable mechanism: acknowledging a problem while placing its resolution in the future.

“We’re behind on the data migration, but the new vendor starts next month and we expect to catch up by Q3.”

“Customer complaints have increased, but the service redesign launching in April should address the root cause.”

“The partnership hasn’t delivered the volumes we expected, but the revised terms being negotiated will change the economics.”

Every one of these statements may be true. The vendor might deliver. The redesign might work. The revised terms might change the economics. But at the point of reporting, none of these future states are evidence. They’re predictions. And the confidence they create is based on anticipated evidence, not actual evidence.

We found temporal displacement in 62% of the executive status updates we reviewed. Problems were acknowledged but paired with forward-looking statements that restored confidence. The executive team would note the problem, absorb the reassurance, and maintain their confidence rating. The problems didn’t trigger investigation or resource reallocation because the narrative assured everyone that the fix was already in motion.

In our insurance partner, we tracked six problems that had been “being addressed” through temporal displacement for over nine months. In four of those cases, the promised future resolution had itself been displaced further into the future at each reporting cycle. The data migration vendor that was going to fix things in Q3 became Q4, then Q1 of the following year. The confidence score never adjusted because there was always a future event that promised resolution.

Mechanism 4: Survivorship reporting

This is the mechanism that surprised us most. Strategic portfolios shrink over time as initiatives are completed, merged, or quietly discontinued. The initiatives that get discontinued are almost always the ones that weren’t working.

But they don’t get formally killed with a post-mortem and a lessons-learned review. They get “deprioritised” or “absorbed into” another initiative or “paused pending resource availability.” They fade from the portfolio dashboard. And when the executive team reviews the remaining portfolio, what they see is the survivors: the initiatives that are performing well enough to still exist.

This creates a statistical illusion. The portfolio looks healthier over time because the unhealthy members are quietly removed. It’s the same bias that makes mutual fund performance look better than it is, because the underperforming funds are shut down and their track records disappear from the averages.

At one of our financial services partners, the strategic portfolio had contained 19 initiatives at the start of the year. By Q3, it contained 12. Seven had been removed through various mechanisms: three “completed” (though none had fully delivered their thesis), two “merged” with other initiatives, and two “paused.” The quarterly review looked at 12 initiatives and rated the portfolio health as strong. Nobody asked what had happened to the other seven. Nobody measured whether the original 19-initiative portfolio was delivering on its collective thesis. The surviving 12 were fine. The seven casualties were invisible.

The compound effect

These four mechanisms don’t operate in isolation. They compound.

A problem gets narratively smoothed as it moves up the reporting chain. The smoothed version gets aggregated with other metrics, burying it further. The aggregate shows a mild concern, which gets temporally displaced with a forward-looking commitment. And if the initiative eventually fails, it gets quietly removed from the portfolio, improving the survivorship statistics.

The result is an executive team that genuinely believes, based on the information available to them, that strategic execution is going well. They’re not ignoring evidence. They’re receiving evidence that has been structurally processed to support confidence. The illusion isn’t deliberate. It’s architectural.

What live evidence shows

In each of our four engagements, we built an alternative evidence view: a direct connection between the strategic bet’s thesis and the operational data that would confirm or deny it. No management summarisation layer. No aggregation. No narrative construction. Just the evidence streams that mattered, updated continuously.

The contrast was stark. Where quarterly reporting showed a strategic bet “on track,” live evidence showed confidence scores in the 30s and 40s. Where narrative smoothing had eliminated warning signals, the raw evidence streams showed clear negative trends that had been developing for months. Where aggregation had hidden variance, the disaggregated view showed that success in one area was masking failure in four others.

Two specific examples stand out.

Our healthcare technology partner had a strategic bet around reducing time-to-value for new customers. The quarterly report showed “implementation timelines improving, average reduced by 11 days.” Live evidence showed that the improvement was entirely driven by a change in customer mix: they were onboarding more small customers (who were faster to implement by nature) and fewer enterprise customers. The actual implementation process hadn’t improved at all. The underlying capability bet was failing while the metric was succeeding.

Our insurance partner had a bet around reducing claims processing costs through automation. The quarterly dashboard showed a 9% cost reduction, rated green. Live evidence showed that the cost reduction was coming from a temporary headcount freeze in the claims team, not from automation. The automation platform was behind schedule and hadn’t yet processed a single claim in production. The cost metric was green. The bet was red.

Closing the gap

The confidence-evidence gap isn’t a problem that better reporting can solve, because the reporting system is the mechanism that creates it. Narrative smoothing, aggregation bias, temporal displacement, and survivorship reporting are features of how organisations package information for leadership consumption. You can’t fix them by asking people to report more honestly. The people reporting are already being honest. They’re being honest within a system that structurally inflates confidence.

Closing the gap requires a different information architecture. Strategic bets need to be connected directly to evidence streams, with explicit thresholds that define what “working” and “not working” look like. That evidence needs to be visible without passing through summarisation layers that smooth it. Disaggregated data needs to be available alongside aggregates. And when initiatives are removed from the portfolio, the removal needs to be visible and accounted for in overall portfolio health.

This is harder than it sounds. It requires data infrastructure that most organisations don’t have. It requires governance models that are built around evidence rather than narrative. And it requires cultural tolerance for a less comfortable version of reality, one where confidence scores sit in the 30s and 40s rather than the 70s and 80s, but where those scores actually mean something.

The executives we worked with universally preferred the uncomfortable truth to the comfortable illusion, once they could see the gap. The problem was never willingness. It was visibility. They couldn’t close a gap they couldn’t see.