Your Dashboard Is Green. Your Customers Are Leaving.

The meeting ends with congratulations. Response times are down. More tickets are closed. Delivery milestones are green. Every department has a reason to celebrate.

Meanwhile, a customer is explaining the same problem for the third time. An operations team is rebuilding a supposedly completed handover. A supplier’s cheaper quotation is generating expensive corrections.

This opening is an illustrative scenario, but the management question is real: what happens when the number improves and the experience gets worse?

A dashboard is a selection of evidence. Its usefulness depends on what was selected, how the measures were defined, and whether anyone is willing to challenge the story they tell.

When the target becomes the product

In September 2016, the US Consumer Financial Protection Bureau fined Wells Fargo $100 million over unauthorized account-opening practices. The regulator linked the conduct to sales targets and compensation incentives. According to the bank’s analysis cited in that action, more than two million deposit and credit-card accounts may have been opened without customers’ authorization. [1]

This was a documented case of misconduct, not simply an imperfect dashboard. Yet it exposes a wider design problem: an account count can rise without a corresponding increase in legitimate customer value.

The lesson for leaders is to examine the behavior a measure rewards. When an activity earns recognition, bonuses, or relief from pressure, ask how someone could increase it without achieving its intended purpose. Then build the checks that would reveal that gap.

A target can be achieved while the business outcome gets worse.

When confidence outruns readiness

Crossrail offers a different lesson. In its 2019 review, the UK Public Accounts Committee found that fixation on the planned December 2018 opening contributed to warning signs being missed or ignored. It described failures to understand cumulative risks and to act sufficiently on information challenging the timetable. [2]

The National Audit Office separately identified an unrealistic schedule, an inadequate plan, and contractual and cost pressures among the factors driving unnecessary cost. Its assessment was of the delivery problems at that time, rather than a final judgment on the completed railway’s value. [3]

For a project sponsor, this raises a practical distinction. Physical progress, contractual completion, and operational readiness answer different questions. A finished-looking station does not, by itself, demonstrate that the railway’s integrated systems are ready to carry passengers.

Apply that distinction to a hospital, factory, software platform, or logistics hub. The steering committee needs evidence that the whole service works under realistic conditions: integration tests, trained operators, resolved critical defects, and an accountable owner for acceptance.

Klarna and the importance of reading beyond the headline

Klarna’s AI customer-service story deserves a careful reading. In correspondence filed with the SEC in July 2025, the company reported approximately $39 million in cost savings during 2024 and said its assistant handled 66% of customer-service chats in the twelve months ended March 31, 2025. These were company-reported figures. [4]

The same correspondence reproduced an SEC staff question about reports of lower service quality. Klarna disputed that characterization, saying the comments concerned improvements to human support and had been misreported. It also said AI-handled chats matched human agents on customer satisfaction and that customers retained the option of human assistance. [4]

This source does not establish that Klarna’s AI caused customers to leave. It illustrates why an executive discussion should examine cost, quality, resolution, and access together—and why a compelling headline should never outrun its evidence.

Automation should face the same test as any operating change: what improved for the person using the service, and what work or inconvenience moved elsewhere?

The invisible transfer of work

Consider a hypothetical support team that reduces average handling time by ending conversations sooner. If more customers then call again, some of the apparent efficiency has become repeat work. If a chatbot requires a customer to re-enter information when escalating to a person, the organization has transferred effort to its user.

The same pattern can appear between departments. Procurement records a saving; operations absorbs the defects. A delivery team marks a feature complete; support handles the confusion. A project closes its handover milestone; the receiving team pays to make the asset usable.

These are illustrative mechanisms, not claims about the companies above. They suggest a useful management rule: follow the work beyond the boundary of the team reporting the improvement.

Keep the metric. Add the missing question.

Activity measures still matter. Teams need to understand speed, cost, volume, and progress. Pair each with evidence of the outcome and a check for unwanted consequences.

B2D Advisory recommendations. These are proposed checks, not findings or prescriptions from the cited case studies. Definitions and review periods should fit the service.

Five changes for the next performance review

1. Define success from the receiving end. Before setting the target, finish the sentence: “The customer or receiving team succeeds when…” Use that outcome to test whether the metric is meaningful.
2. Separate evidence from confidence. Give each reported result a definition, reporting period, source, and owner. Mark estimates, stale data, and unknowns explicitly. Missing information should not silently become green.
3. Inspect the exceptions. Sample repeat complaints, failed handovers, abandoned journeys, and difficult cases. An improving average can conceal a deteriorating experience for one group.
4. Make bad news usable. Ask what support or decision a red status requires. If early disclosure only brings punishment, leaders make it harder to obtain the information they need.
5. Check benefits after delivery. Return after an agreed interval to confirm whether the promised outcome materialized. Keep a named business owner responsible for that review.

These changes do not require another crowded dashboard. They require a better conversation about the evidence already being collected—and the evidence still missing.

The question that changes the meeting

At the next review, choose one green indicator and ask:

“What would the customer—or the team receiving our work—show us that could contradict this result?”

Then inspect that evidence. A reopened case, a failed acceptance test, or an unusable feature may tell you more than another month of favorable averages.

The dashboard should help leadership see reality early enough to act. Its color matters less than the decisions it makes possible.