Why Your Safest Sites Might Be Your Riskiest Ones

Blog by: Bryan Statham, CEO at LifeBooster

 

Every year, somewhere in your organization, a site wins the safety award. Lowest recordable rate in the network. Recognized on the quarterly call. Held up as the model for everyone else.

The award is sincere, and the site may well deserve it. But consider what it actually measures. It measures the absence of a particular kind of paperwork over a particular window of time. It does not measure how much strain that workforce absorbed to produce the quarter.

Those two things can move in opposite directions for years before anyone notices.

Which raises an uncomfortable question. Your lowest-reporting sites and your lowest-risk sites are not always the same sites. When you cannot tell them apart, you are allocating capital, headcount, and leadership attention on a signal that was never built to carry that weight.

 

The injuries are getting rarer and more expensive at the same time

Here is the pattern almost every industrial operator is living through right now, whether or not anyone has named it.

Reported injuries keep falling. Lost-time claim frequency dropped again last year. But the cost of each one that does surface keeps climbing: medical severity and indemnity severity both rose in the same period, and the long decline in frequency has started to flatten out.[1]

So the scoreboard improves while the bill goes up.

That is not a paradox once you understand what each number is measuring. Frequency counts events. Severity reflects what was actually happening to people’s bodies before the event. One of those is getting better. The other is not.

Musculoskeletal injuries sit at the center of this. When someone goes down with an MSD, they are away roughly twice as long as the average injured worker.[2] Overexertion remains the single most expensive cause of serious workplace injury in the United States.[3] These are not freak accidents. They are the predictable end point of exposure that built up over months, in plain sight, and went unmeasured.

 

Three reasons a clean record is a weak signal

The number is statistically unstable. The most rigorous study of recordable rates analyzed 3.26 trillion worker hours across fifteen years and concluded that the metric is “96% to 98% random.” There is no meaningful association between recordable rates and fatalities. The authors’ conclusion was blunt: it is statistically invalid to use it to compare companies, business units, projects, or teams.[4]

Put that in operational terms. When you rank twelve sites by recordable rate and act on the ranking, you are very likely acting on noise.

The number is filtered before it reaches you. Injury counts are not observations of risk. They are the output of a long chain of human decisions. Whether a worker names the discomfort. Whether a supervisor logs it. Whether it meets a recordability definition.

Each link can suppress the signal, and none of them fails randomly. In one large manufacturing survey, roughly half of workers with musculoskeletal symptoms said they had not reported them. The reasons were not indifference. The symptom seemed minor. Reporting felt like a burden on the team. Nobody believed it would change anything.[5]

This is the part that should trouble an operations leader most. Reporting culture varies enormously between sites. A strong speak-up culture generates more paperwork. A heads-down culture generates less. If your comparison metric is paperwork, you will systematically reward the quieter site and investigate the honest one.

The number arrives too late to act on. Musculoskeletal injury is cumulative. Load, repetition, posture, and insufficient recovery build for weeks or months before tissue fails. By the time a claim exists, the exposure that caused it has been running a long time and the window to intervene has largely closed. Delay compounds financially too: claims reported a month or more after the injury cost roughly half again as much and are far more likely to end up in litigation.[6]

 

What silence actually costs you

There is a version of this conversation that treats unreported discomfort as a data problem. It is not. It is an operational problem, and it is expensive long before it ever becomes a claim.

A worker with a sore shoulder does not stop working. They work differently. They favor the other arm, which loads a joint that was not designed for it. They slow down on the second half of the shift. They stop volunteering for the harder station. They take the break they are entitled to and come back no more recovered than when they left.

None of that shows up in your injury log. All of it shows up in your numbers.

The research here is unambiguous and it is larger than most executives expect. When the cost of pain in the workforce is measured properly, more than three quarters of the lost productive time comes from reduced performance while people are at work, not from absence.[7] The days someone is missing are the small part. The days they are present and diminished are the large part.

Follow that through an operation and you get a familiar list. Output drifts below standard without an obvious cause. Quality slips at the end of the shift. Overtime creeps up to cover the gap. Your most experienced people start bidding away from the physically demanding stations, which concentrates that demand on newer workers who are already the most likely to get hurt. Attrition rises among exactly the crews you can least afford to lose.

Ask a plant manager why throughput softened last quarter and fatigue will rarely be the first answer, because nothing in the reporting system measures it. But it is often the real one.

Unreported discomfort is not a gap in your safety data. It is a cost already flowing through your P&L under other names.

 

The comparability problem nobody names

Even if you accepted incident counts as a risk signal, you could not use them to compare sites.

Two plants in the same network differ in reporting culture, workforce tenure, occupational health staffing, claims administration, jurisdiction, and case management philosophy. Every one of those moves the injury count independently of the physical demand placed on workers.

Put those two plants side by side on a recordable rate and you are not comparing risk. You are comparing two reporting systems.

This is where enterprise safety programs stall. Leadership asks a reasonable question, which is where should we invest first across the network, and the organization cannot answer it with evidence. So it defaults to the loudest site, the newest incident, or the most persuasive plant manager.

What is missing is a common risk currency. A consistent unit of exposure that means the same thing in Ohio as it does in Guadalajara, on a packaging line as on a loading dock.

 

Building a common risk currency: five requirements

  1. Anchor it to a validated exposure standard. The unit has to mean something outside your organization or it becomes another number people argue about. Established occupational exposure standards already do this work. Research validating exposure thresholds against injury outcomes shows a clear dose-response relationship: the higher the measured exposure, the higher the incidence of injury.[8] That is the property incident counts lack, and it is what will hold up in front of your insurer and your board.
  2. Measure work as it is actually done, across the full shift. Traditional assessment samples a few minutes of a job, scored by an analyst, at a moment the worker knows they are being watched. It is valuable and it does not scale. It also disagrees with itself, often substantially, when different analysts assess the same task using different established methods.[9] Continuous sensing changes the sampling problem entirely. Wearable sensors are small, low-power, and can run for an entire shift, which lets them capture what a ten-minute observation never will: the fatigue curve, the cycle-time drift, the overtime hour.[10]
  3. Normalize before you compare. Raw exposure totals penalize large sites and flatter small ones. Express the currency in units that survive differences in headcount, shift pattern, product mix, and season. Exposure per hour worked, per task, per role. Without this, comparability is an illusion and your site leaders will be right to reject the ranking.
  4. Rank by consequence, not volume. The most frequent exposure is not always the most costly one. Weight it by the severity and duration of the injuries that exposure is known to produce, and by what those injuries cost operationally in downtime, replacement labor, and overtime. This is what turns a safety metric into an operational one, and what earns the capital request.
  5. Validate controls against exposure, not against the next injury. Most organizations implement a control and then wait for the injury data to move. That takes quarters, and it may never produce a clean read. If your unit of measure is exposure, you re-measure the same task a month later and see whether risk actually came down. Controls that work get scaled. Controls that do not get stopped early. The program stops running on faith.

 

The inputs are already in the building

Source What it contributes What it cannot do alone
Wearable and sensor exposure data Continuous measurement of posture, repetition, force, and heat Explain why exposure changed
Operational systems (MES, WMS, scheduling, HRIS) Production intensity, staffing, overtime, tenure, shift design Measure physical demand on the body
EHS and incident systems Historical outcomes, near-miss and hazard reports Predict where the next injury forms
Claims and occupational health data Severity, duration, and true cost of outcomes Arrive in time to prevent anything
Ergonomic assessments Task-level expert judgment and control design Scale across thousands of jobs

The gap is rarely data availability. It is the intelligence layer that reconciles these sources into one comparable unit and puts it in front of a decision-maker while there is still time to act.

 

Four questions for your next operations review

  • 1. If our lowest-incident site also has our lowest reporting rate, would we be able to tell? If not, reporting culture is currently masquerading as performance.
  • 2. What is our common unit of risk, and does it mean the same thing at every site? If the answer is “recordable rate,” we do not have one.
  • 3. When we implemented our last major control, how did we confirm it worked, and how long did that take? If it involved waiting for the next injury cycle, we are managing a lagging system.
  • 4. Which site has the highest physical demand per hour worked? This is the question a common risk currency answers and the incident log cannot.

 

The reframe

The site with the clean record may be genuinely well run. It may also be a site where people have quietly stopped reporting, where a long-tenured crew is absorbing demand that will surface as claims in three years, or where good fortune has simply held for eleven quarters.

You cannot tell from the scoreboard. That is not a failure of the people running the site. It is a limitation of the instrument.

Low incident counts are worth celebrating. They are not evidence of low risk. The organizations pulling ahead are the ones that stopped treating the two as the same thing, and started measuring exposure with the rigor they already apply to throughput, quality, and cost.

Risk that is measured becomes risk that can be governed. Everything else is hoping the quarter holds.

LifeBooster helps industrial organizations measure workplace exposure continuously and compare risk consistently across sites, so safety decisions rest on evidence rather than incident counts.

To see how a common risk currency would look across your sites, reach out to our team.

 

Sources

[1] NCCI, 2026 State of the Line Report (2025 accident year data): lost-time claim frequency fell 2%, a more moderate decline than the long-term average, while medical and indemnity claim severity each rose 4%. https://riskandinsurance.com/workers-compensation-remains-profitable-as-premium-dips-and-severity-climbs/

[2] US Bureau of Labor Statistics, Employer-Reported Workplace Injuries and Illnesses, 2023–2024, Table 2: median 14 days away from work for overexertion, repetitive motion and bodily conditions, versus 8 days for all days-away cases, private industry. https://www.bls.gov/news.release/osh.t02.htm

[3] Liberty Mutual, 2025 Workplace Safety Index (2022 data): overexertion involving outside sources ranks first at $13.7 billion in annual direct cost. https://www.libertymutualgroup.com/about-lm/news/articles/us-companies-spend-50.87b-year-top-ten-causes-serious-workplace-injuries-according-2025-liberty-mutual-workplace-safety-index

[4] Hallowell, M., Quashne, M., Salas, R., Jones, M., MacLean, B., & Quinn, E. (2021). “The Statistical Invalidity of TRIR as a Measure of Safety Performance.” Professional Safety, American Society of Safety Professionals, April 2021. Analysis of 3.26 trillion work hours across ten large construction organizations over fifteen years. https://www.eei.org/-/media/Project/EEI/Documents/Issues-and-Policy/Power-to-Prevent-SIF/PSJ—TRIR-Paper.pdf

[5] Park, J.-T., & Yoon, J. (2021). “Why Workers Hesitate to Report Their Work-Related Musculoskeletal Symptoms: A Survey at a Korean Semiconductor Company.” International Journal of Environmental Research and Public Health, 18(21), 11221. 49.1% of 1,580 symptomatic respondents had not reported at least once. https://www.mdpi.com/1660-4601/18/21/11221

[6] Liberty Mutual, Workers Compensation Claim Reporting Lag Study (national account data, 2016–2018): claims reported 29+ days after injury cost 52% more on average and are 152% more likely to be litigated than those reported within three days. https://business.libertymutual.com/wp-content/uploads/2021/05/NI_LagStudy.pdf

[7] Stewart, W. F., Ricci, J. A., Chee, E., Morganstein, D., & Lipton, R. (2003). “Lost Productive Time and Cost Due to Common Pain Conditions in the US Workforce.” JAMA, 290(18), 2443–2454. Survey of 28,902 working adults; 76.6% of lost productive time was explained by reduced performance while at work rather than absence. https://pubmed.ncbi.nlm.nih.gov/14612481/

[8] Bonfiglioli, R., Mattioli, S., Armstrong, T. J., et al. (2013). “Validation of the ACGIH TLV for hand activity level in the OCTOPUS cohort: a two-year longitudinal study of carpal tunnel syndrome.” Scandinavian Journal of Work, Environment & Health, 39(2), 155–163. Incidence rate ratios of 2.43 between action limit and TLV, and 3.32 above TLV, in a cohort of 3,860 workers. https://www.sjweh.fi/article/3312

[9] Kee, D. (2022). “Systematic Comparison of OWAS, RULA, and REBA Based on a Literature Review.” International Journal of Environmental Research and Public Health, 19(1), 595. Agreement between methods ranged from null to 100%, with most studies showing correlations below 0.5. https://www.mdpi.com/1660-4601/19/1/595

[10] National Institute for Occupational Safety and Health (2020). “Assessing Lifting Risk Factors Using Wearable Motion Sensors.” NIOSH Science Bulletin, October 13, 2020. https://www.cdc.gov/niosh/bulletin/2020/sensors.html