How an experience mod is calculated: primary losses, excess losses and the split point

Ask ten business owners what their experience mod is and most can answer. Ask what it's made of and almost none can — including some of the brokers who sell against it. That's a problem, because the mod is the single largest controllable factor in a workers' comp premium, and the mechanics of the formula decide what "controllable" means.

Every number in this article was computed, not asserted. The example is a composite staffing account; the arithmetic is reproducible from the figures shown.

The formula

The NCCI experience rating formula compares your actual losses to what the bureau expected from an employer of your size and class mix:

        Ap + W·Ae + SV
Mod = ────────────────────        where SV = B + (1−W)·Ee
          Ep + Ee + B

• Ap — actual primary losses. Each claim counts only up to the split point (we use $18,500, the NCCI value in our 2026 modeling; it varies by state and year).
• Ae — actual excess losses. The portion of each claim above the split point.
• Ep / Ee — expected primary and excess losses. The bureau's expectation, from your payroll, your class codes' expected loss rates, and a D-ratio that splits expected losses into primary and excess.
• B — ballast and W — weight. Stabilizers that grow with employer size. Small employers get a large stabilizing cushion (so one claim can't wreck them); large employers get more credibility (their own losses count for more).
• SV — stabilizing value, the ballast plus the unweighted share of expected excess.

One sanity check worth knowing: if actual losses exactly equal expected losses, the formula returns exactly 1.000 — for any ballast and any weight. A mod above 1.00 means you ran worse than expected; below 1.00, better.

A worked example

Three unit-statistical years of payroll produce expected losses of $607,240. With a D-ratio of 0.28, that splits into expected primary of $170,027 and expected excess of $437,213.

The account has eleven claims totaling $594,000 incurred — just under expectations. Capping each claim at the split point gives actual primary of $147,050 and actual excess of $446,950.

With a ballast of $210,000 and a weight of 0.22 (representative values for this account size — real rating reads them from the bureau's published tables):

SV          = 210,000 + 0.78 × 437,213            = 551,026
Numerator   = 147,050 + 0.22 × 446,950 + 551,026  = 796,405
Denominator = 170,027 + 437,213 + 210,000         = 817,240
Mod         = 796,405 / 817,240                   = 0.975  →  filed as 0.97

Total actual losses ran 98% of expected, and the mod lands at 0.97. The formula behaves.

Why frequency hurts more than severity

Here is the part of the mechanics that changes behavior. Because each claim is capped at the split point before it enters the primary bucket, a $214,000 claim and a $19,000 claim contribute almost identically to primary losses — $18,500 each. The rest of the big claim lands in the excess bucket, where it's multiplied by the weight (0.22 here) before it touches the numerator.

Run the comparison on this account's parameters:

Roughly the same total dollars. The twelve small claims push 3.6 times as much into the numerator. That is the formula telling you what it believes: frequency is predictive of future losses; a single severe claim is substantially bad luck.

The underwriting consequence: a safety program that prevents ten strains does more for the mod than one that would have mitigated a catastrophic injury. If you manage to the mod, you manage frequency first.

What actually moves a mod

1. Close claims before the valuation date. Open-claim reserves count as incurred losses. A claim reserved at $60,000 that would close for $25,000 is carrying $35,000 of phantom loss into the calculation. (Our loss run forensics post covers how to find these.)
2. Fix coding errors. Claims coded to the wrong policy year or the wrong entity flow straight into Ap.
3. Attack frequency. Per the table above.
4. Know your dates. The unit stat data that drives next year's mod is reported about six months after each policy year ends — months before renewal. By renewal time, the mod is largely already determined. The window to influence it is earlier than most brokers engage.

HALO IQ's ModIQ runs this projection from your loss runs before the bureau publishes — the same formula, your actual data, with every open claim's effect priced. Try the X-Mod estimator with your own figures.

Representative parameters are used for ballast and weight; actual rating values come from the rating bureau's published tables for your state and account size, and the split point varies by state and year.