Safety technology ROI: building an honest business case

An honest business case for safety technology starts from your own cost of the hazard the technology targets, counts every cost of owning the system, and shows a range of outcomes, including one where injuries do not fall at all. Its most useful output is usually the break-even reduction: how much the targeted costs must fall for the investment to pay for itself. Vendor ROI figures can inform the high end of that range, but they are claims, and a case that depends on them will not survive a skeptical finance review.

Why do safety technology business cases fail?

Most fail in one of three ways. They borrow outcome figures from vendor case studies and treat them as likely results. They count license fees but leave out installation, integration, alert review and governance work. Or they convert national injury statistics into site savings, which those statistics were never designed to support.

HM Treasury's Green Book, the UK government's appraisal guidance, describes optimism bias as a demonstrated, systematic tendency for appraisers to be over-optimistic: costs turn out higher, benefits lower and delivery slower than planned ^[7]. It tells practitioners to correct for this at the outset by raising cost and time estimates and lowering benefit estimates ^[7]. The guidance is written for public projects, but the bias is not limited to government, and safety technology cases are exposed to it because the benefits are uncertain and the sales material is optimistic.

Step 1: Price the status quo with your own numbers

National figures tell a board the problem is real. US private-industry employers reported 2.5 million nonfatal injuries and illnesses in 2024, a rate of 2.3 cases per 100 full-time equivalent workers ^[3]. OSHA's business case page cites Liberty Mutual's estimate that employers pay more than $1 billion a week in direct workers' compensation costs for disabling, non-fatal injuries ^[1]. In Great Britain, HSE puts the total cost of workplace injuries and new cases of work-related ill health at around £22.9 billion in 2023/24, with employers bearing about £4.3 billion of that, individuals £13.4 billion and government £5.2 billion ^[4].

That last split matters for a business case. Most of the cost of injury falls on workers and the public purse, so a case built only on employer savings will understate the value of prevention. It is still the figure a finance team will look at, so be clear which costs you are counting.

For the site itself, collect three to five years of data for the hazard category the technology addresses:

• Workers' compensation or employer liability claim costs, paid and reserved.
• Lost time, restricted duty and overtime to cover absence.
• Property, equipment, racking and product damage. For vehicle hazards this is often more frequent and better recorded than injury cost.
• Downtime and the hours spent on investigations.
• Insurance premiums, deductibles and any insurer requirements.
• Regulatory exposure: citations, improvement notices and legal costs.

Be careful with indirect cost multipliers

Many calculators add "indirect costs" by multiplying direct costs. OSHA's $afety Pays tool uses ratios that fall from 4.5 for claims under $3,000 to 1.1 for claims of $10,000 or more, taken from a Business Roundtable construction publication based on a Stanford University study ^[5]. OSHA notes that the multiplier varies with each employer's circumstances and that the figures are general estimates based on limited research ^[5]. If you use a multiplier, label it as an assumption and test the case without it. Better still, replace it with your own overtime, downtime and damage records.

Step 2: Count the full cost of ownership

Alert review time is easy to miss and easy to estimate. Multiply expected alerts per shift by the minutes each takes to review, then by shifts per year and a loaded hourly rate. Ask the vendor for alert volumes at comparable sites, then measure your own during a pilot.

"Acting on the data" deserves its own line. If the system works, it will find problems that cost money to fix. A business case that assumes benefits but no remediation spending is internally inconsistent.

Step 3: Estimate the effect as a range

The expected effect is where most cases overreach. Vendors publish outcomes such as percentage reductions in unsafe events, incidents or claims. Treat each as a vendor claim: usually from self-selected customers, without a control group, often measured in detected events rather than injuries, and not independently verified. Before using any such figure, ask what the baseline period was, what denominator was used, whether a comparison site showed the same change, and who did the calculation. Sites that adopt technology after a bad year will tend to improve anyway through regression to the mean ^[8], which inflates before-and-after comparisons.

It helps to know what well-designed evidence looks like. Levine, Toffel and Johnson compared 409 randomly inspected California workplaces with 409 matched controls and found a 9.4% decline in injury rates and a 26% reduction in injury costs, with no evidence of harm to employment, sales, credit ratings or firm survival ^[2]. The confidence intervals around both estimates were wide ^[2]. That is a randomized study of a regulatory intervention, not of safety technology, but it shows that credible effects on injuries tend to be in the tens of percent, with real uncertainty. A business case assuming a 70% fall in injuries from software alone should prompt hard questions.

Peer-reviewed economic evaluations exist for some intervention types. A systematic review by the Institute for Work & Health found strong evidence that ergonomic and other musculoskeletal injury prevention interventions in manufacturing and warehousing are worth undertaking on financial grounds ^[6]. The same review noted that the methodological quality of economic evaluations in this field needed to improve ^[6]. No comparable body of independent evidence yet exists for AI video safety analytics specifically.

A defensible structure:

• Zero scenario: no reduction in injury or damage costs. Shows the pure cost of the investment.
• Low scenario: a small reduction in the targeted costs, consistent with modest behavior change.
• Central scenario: based on leading-indicator changes you measured in a pilot, with the assumption linking them to cost reduction written out.
• High scenario: the most optimistic case you can defend. Vendor-reported results belong here at most, labeled as vendor claims.

Be explicit about the link between leading and lagging measures. A pilot might show fewer pedestrian entries into vehicle aisles, but a 2025 scoping review found the evidence connecting leading indicators to lagging safety outcomes to be weak overall ^[10]. The link is an assumption, and the case should say so.

Step 4: Find the break-even point

The Green Book recommends sensitivity analysis and switching values: the value a variable would need to reach for an option to stop representing value for money ^[7]. For a safety technology case, the most useful switching value is the break-even reduction in the targeted costs.

Here is an illustrative example. The figures are invented to show the method and do not describe any real site or product.

Five-year cost: $502,500. Suppose the site's own records show $240,000 a year in vehicle-related claims, damage and downtime, or $1.2 million over five years. To break even without discounting, those costs would need to fall by about 42%.

In this illustration the investment does not pay back on cost savings even in the high scenario. That does not make the investment wrong. It means the decision rests on other grounds, which Step 5 covers, and that the scope or price may need to change. A finance team will trust that conclusion more than a single ROI percentage. Ask your finance team for their discount rate and add a present-value line; discounting makes later savings worth less, so it raises the break-even reduction further.

You can also turn the question around: what would make it pay? Narrowing coverage to the highest-risk zones, negotiating license fees, or reducing alert volume through better configuration all lower the break-even point. Show those options.

Step 5: State the non-financial reasons plainly

Some reasons to invest are real but hard to price: legal duties to control a known serious risk, insurer requirements, better evidence for investigations, visibility on night shifts and remote areas, or a group commitment following a serious incident. OSHA's leading indicators guide also notes savings beyond claims, such as repair and production costs when hazards are fixed early ^[9]. State these in words, with whatever evidence you have. Do not convert them into invented dollar figures to close a gap in the numbers.

Also be careful with survey-based return figures. OSHA's business case page reports that over 60% of chief financial officers in a 2005 Liberty Mutual survey said each $1 invested in injury prevention returns $2 or more ^[1]. That is a measure of what executives believe, not a measured return, and it says nothing about any particular technology.

Step 6: Plan how you will check the case

A business case is a forecast. Write down how you will test it:

• Baseline the targeted costs and leading indicators before go-live, using the same definitions you will use afterwards.
• Use a comparison area or staggered rollout where possible.
• Agree pilot acceptance criteria with measurable thresholds, and an exit if they are not met.
• Review actual costs, including alert review time, at six and twelve months, and update the scenarios.

Summary

A defensible safety technology business case uses your own cost baseline for the targeted hazard, counts the full cost of ownership, corrects for optimism bias ^[7], and presents a range of outcomes with a zero-benefit case and a break-even reduction. Credible evidence on safety interventions points to real but modest effects ^[2], and independent evidence specific to AI safety technology is still thin, so vendor ROI figures belong at the optimistic edge of the range, labeled as claims. When the numbers do not pay back on savings alone, say so and make the case on the legal, risk and operational grounds that actually drive the decision.