Insightful's 92% Utilization Score Doesn't Add Up

Why a 92% Utilization Score Should Raise Eyebrows

On the homepage of Insightful, a company that sells employee-monitoring software, there’s a customer testimonial that reads roughly like this: productivity started in the high-70% range when the tool was first deployed, but it’s now climbed to 92%, and employees have started coming to ask what their score was that day. The testimonial even says the work has turned into something like a game.

And yet, in a Wall Street Journal interview, the same company’s CEO said the range most companies actually target is 60-80% of the workday. The advice that article ended up giving to office workers was this: if you’re going to artificially inflate your activity rate, be careful — hitting 90% is more likely to get you flagged as suspicious than praised.

Three numbers don’t line up here. The customer testimonial’s 92%. The vendor’s own stated normal range of 60-80%. And the 90% threshold where manipulation is suspected. In effect, the metric the company built is tallying employees’ ability to “look like they’re working” and counting it as performance.

What Exactly Does a 92% Utilization Rate Measure?

First, we need to look at what this number actually counts. Utilization rate¹ is the share of working hours spent on activities classified as “work.” The operative phrase is “classified as.” It doesn’t measure what you actually produced — it measures whether your cursor lingered on apps and sites that a manager pre-designated as work-related.

So there are only two ways to raise this number: do more work, or look like you’re working.

The workarounds the Wall Street Journal compiled are, tellingly, almost all of the latter kind. The first tip is to keep your calendar meticulously filled. When you’re on a call or in an in-person meeting, your messaging status switches to “away” — and monitoring tools cross-check whether something is scheduled during that window to decide whether the absence is legitimate. If your calendar is empty, it reads as slacking off. There’s also advice to use a physical mouse jiggler² rather than a software one, since security systems have started blocking the software kind.

What these tricks have in common is that none of them make you better at your job — every single one is a way to get the monitoring tool to log your time as work.

Worth looking at the size of this monitoring market too. Insightful claims in its own materials that nearly 80% of major companies have adopted employee monitoring, and that more than 5,100 teams use its product. The product pitch itself is telling: it promises “complete visibility without surveillance or keystroke logging,” while also offering tracking of app and website usage, idle time, and — AI adoption patterns and unauthorized AI use. The word “surveillance” is conspicuously absent, even as the list of things being counted keeps growing.

None of this is new, really. The prototype is often traced to the “boss button,” built in the early 1980s by a developer named Roger Wagner — one keystroke switched your screen to a spreadsheet the instant your boss approached. Wagner himself says it was a joke: not a way to shield lazy employees, but a jab at overbearing managers. 40 years later, that joke has become a survival skill.

AI Token Usage at the Center of Meta’s Layoff Lawsuit

Up to this point, it’s mostly a matter of employees and managers eyeing each other warily. The real problem is that these numbers allegedly made their way into hiring and firing decisions — and that claim has turned into a lawsuit.

On July 13, 26 current and former employees filed suit against Meta in federal court in Oakland. The backdrop is the roughly 8,000-person layoff — 10% of the workforce — announced in May. The complaint describes the company deploying a “constellation of internal artificial intelligence systems,” alleging that keystroke and activity monitoring data, algorithm-assisted performance rankings, and an AI token³ usage dashboard were used to select who would be cut. Meta’s position is that humans make the final call.

There’s one sentence in the complaint I kept coming back to. It states that these scores and rankings are, by design, ones that employees on protected leave — or whose output dropped due to a disability — cannot accumulate. All 26 plaintiffs had taken medical, parental, or family leave, or had requested disability accommodations. A temporary restraining order was denied on July 17, and a preliminary injunction hearing was held yesterday, August 24.

Regardless of how the lawsuit turns out, there’s something here worth paying attention to: the plaintiffs’ claim that the company used AI usage as a metric for deciding who to cut.

The context matters too. In April 2026, Reuters reported that Meta had been logging employees’ keystrokes, mouse movements, and clicks for AI training purposes. The program reportedly ran into potential conflicts with European privacy regulations and was shut down after it emerged that other employees could access conversations and records the system had collected. The layoff announcement came the following month. It’s a timeline that shows just how quickly data collected for observation can slide into other uses.

Worker surveys reveal the same metric’s limits. In a Visier survey of 1,000 U.S. workers cited by The Wall Street Journal, 48% admitted to having inflated their own AI usage numbers. The researchers’ interpretation: it’s a defensive move against looking like you’re falling behind within the organization.

Now overlay the two facts. The company counts AI usage. Half the employees inflate that number. The company ranks people by the inflated number. And that ranking becomes the layoff list. Anyone without the room to inflate gets pushed to the bottom — someone on parental leave who simply never logged in being the clearest example.

There’s one more twist. Companies have started scrutinizing token costs lately, so heavy usage now reads as waste too. The zone of “correct” behavior keeps narrowing — and keeps moving. In last July’s issue, I called the invisible time spent verifying AI output a “verification tax.” This time, what’s being billed isn’t verification — it’s the time spent performing.

Korea Has a Legal Right to Explanation for Automated Decisions

In some U.S. states, companies aren’t even required to disclose that monitoring is happening. Korea’s situation is different.

Article 37-2 of Korea’s Personal Information Protection Act (“PIPA”) spells out data subjects’ rights regarding automated decisions. The provision branches into four parts. If a fully automated system’s decision has a significant effect on someone’s rights or obligations, that person can refuse the decision (Paragraph 1); they can demand an explanation of the decision (Paragraph 2); if such a demand is made, the business must take remedial measures like human-involved reprocessing (Paragraph 3); and the business must disclose the criteria and procedures behind automated decisions in an easily accessible way (Paragraph 4).

Here’s the catch. The right to refuse in Paragraph 1 comes with conditions attached. You can’t refuse if consent was given, if it’s a statutory obligation, or if it’s necessary for performing a contract. There’s plenty of room for “performance of an employment contract” to fall under this exception. Kwon Seok-hyun, a lawyer at Lawyers for a Democratic Society (“Minbyun”), points out that the blind spot in Korea’s current personal-data legal framework is treating ‘consent’ as a master key that justifies any form of surveillance. It fails, he argues, to account for the power imbalance between employers and employees at all.

But if you read the provision to the end, something survives. The right to demand an explanation under Paragraph 2 and the disclosure obligation under Paragraph 4 carry no such conditions. Your refusal might be blocked, but asking what criteria you were scored on, and forcing the company to disclose those criteria, is a separate matter entirely. As far as I can tell, these two provisions are barely used in Korea.

Put it in question form and it looks like this: which items in my work data feed into the evaluation, how are those items weighted, and does the calculation pause during leave or sick leave. If a company is using automated judgment, these three things fall under the disclosure requirement.

Nor is the field quiet on this issue. A report on electronic workplace surveillance published in October 2025 by Gabjil 119 (“Workplace Bullying 119,” a Korean labor rights group) found that corporate solutions already on the market capture PC screens second by second and play them back like video. The report also catalogs what’s being collected at Korean workplaces: internet-usage logs during work hours, messenger and email records, CCTV footage, GPS-based location, personal social-media activity, and even detection logs for PC power state and mouse/keyboard activity. This overlaps almost exactly with the items at issue in this lawsuit. The report includes cases where an employee who refused a recommended resignation had their CCTV-tracked movements thrown back at them, and where a whistleblower’s personal messenger conversations were restored from their work PC and used against them. A note: this piece isn’t legal advice. If you’re facing an actual dispute, consult a professional first.

Oswarld’s Lens

I’ve designed dashboard KPIs many times while working on GTM strategy, and the purpose of a metric always drifts in the same order. At first, it’s for observation — you build it to see what’s happening. Then it becomes a reporting tool, showing up on slides for executives. Finally, it becomes an evaluation tool.

A number that was reasonably accurate when it was just for observation tends to climb once it starts being used for evaluation. But actual performance doesn’t move. This is exactly what Goodhart’s Law⁴ describes.

That’s why I can’t read a figure like 92% as a success story. If that number shows up, I don’t see it as cause for celebration — I see it as a signal to start checking. The vendor itself said the normal range is 60-80%.

AI usage, in particular, is close to the worst possible metric. It counts input, not outcome; it takes almost no effort to inflate; and on top of that, there’s now a counter-signal in play — that using it too much is wasteful. Tying a metric like this to someone’s employment is a design failure.

I’m not against monitoring tools as such — they’re useful for finding bottlenecks. But whenever a measured number gets used for evaluation, a human being has to make one more judgment call before it counts.

Closing

Here’s the summary.

First, utilization rates and AI usage are activity logs, not outcomes. Once they’re tied to evaluation, knowing how to leave a good record starts to matter more than actual competence. Second, the problem the Meta lawsuit raises is the claim that these records were used in employment decisions—and that people on leave, or whose output drops due to disability, are structurally disadvantaged because they can’t rack up the same score. Third, Korean workers already have a channel to demand explanations and disclosure of criteria. It’s just not being used.

If you’re a manager, I’d suggest opening your dashboard today and asking just one question. Is the easiest way to raise this number to actually do good work—or just to look like you are?

What number are you being evaluated on at your company right now? If you’ve ever taken actions that had nothing to do with real performance just to protect that number, tell me what they were in the comments. Once I’ve collected enough cases, I’ll break them down by metric in a future issue.

---

💬 Tell me in the comments what metric you’re being evaluated on right now · 📨 If you have a colleague who watches a dashboard, pass this piece along to them

---

References & Further Reading

Primary sources

• Cordilia James, “How to Outsmart AI When It’s Tracking Your Workday”, The Wall Street Journal, 2026. ··· This is where this issue started. The article’s strength is that it interviews both the makers of surveillance tools and the people who game them.
• Sanders et al. v. Meta Platforms, Inc., U.S. District Court for the Northern District of California (Oakland), filed July 13, 2026. ··· The key section, in a 71-page complaint, addresses the non-cumulative nature of AI scores. Regardless of how the case turns out, it’s a well-organized document on the structural blind spots of automated evaluation.
• Personal Information Protection Act, Article 37-2 (Rights of Data Subjects Regarding Automated Decisions), Korea Law Information Center. ··· I’d recommend checking the proviso in Paragraph 1 against the differences in Paragraphs 2 and 4 yourself. Reading the clause takes about 3 minutes.

Background

• Workplace Gapjil 119 (Jikjang Gapjil 119, a Korean labor watchdog NGO), “Policy Report on Electronic Workplace Surveillance: Status and Legal-Institutional Improvements,” October 12, 2025. ··· This contains concrete domestic case studies. Reading only foreign coverage, it can feel like someone else’s problem — this report shows it isn’t.
• On Charles Goodhart’s law of metrics, Marilyn Strathern, “Improving Ratings: Audit in the British University System”, European Review, 1997. ··· The oft-quoted phrasing of Goodhart’s Law that we all cite actually traces back to this paper.

Related past issues worth reading

• What the Luddites Smashed Wasn’t the Machine ··· This issue covered the concept of a “verification tax.” It connects to this issue’s “time spent performing.”
• Thousands of IBM Consultants Are Taking a Test ··· This covered how AI usage becomes a credentialing signal inside organizations. This issue asks: what happens when that signal becomes an evaluation metric?

---

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

📝 Glossary

Footnotes

1. Utilization rate: The share of work hours classified as work-related activity. It’s a metric that counts where time went, not what got produced — a different concept from output. ↩

2. Mouse jiggler: Software or a USB device that automatically moves the mouse cursor to make it look like someone is at their desk. It’s used to prevent a status from switching to “away.” ↩

3. Token: The smallest unit an AI model uses to process text. Since it’s the basis for measuring usage and cost, how many tokens were used is treated as a number that shows how much AI was actually used. ↩

4. Goodhart’s Law: The principle that once a measure becomes a target, it ceases to be a good measure. It originated in British economist Charles Goodhart’s commentary on monetary policy. ↩