Every programme is behind loss trend, and the rate monitor understated a 12% cut as a 1.4% one
Restated at realistic scale, and re-based on a rate monitor. Rate here is premium per $100 of payroll, the exposure base contractors' GL is actually written on. Renewal Rate Change is measured on renewal business only, from the rate-action flag — the industry's rate monitor. A loss-cost trend of 7.0% is a stated assumption, not a field in these datasets. Rate adequacy is rate change minus trend.
Rate looks firm and it is not adequate. The book took on renewals against a 7.0% loss-cost trend — a 2.3-point shortfall — and every one of the fifteen programmes is behind, from at the top to at the bottom. Contractors GL, the firmest-looking programme, took . Inside it, one slice in Texas and Florida has cut rate four times in eighteen months: the monitor printed while the book actually wrote , because 36% of those renewals were flagged Decrease and those decreases averaged against a band that assumes five.
So what? A rate monitor that counts categories cannot price them. This one understated a 12% cut as a 1.4% one, and at programme level the rest of the book halved what was left, to . The proof is not in any month — it is in the cohorts. Business written in 2024-Q3 runs . The same slice eight quarters later runs . Each cohort has been priced a little worse than the last, and they are all still on the books.
against a 7.0% loss trend — every programme is behind
what the book actually wrote, per $100 of payroll
seven times the band the rate-action field assumes
eight cohorts of steady deterioration, still on the books
Four numbers the rest of the brief explains.
Renewal rate change against loss-cost trend · June 2026
The bar is what each programme achieved on renewal business. The dashed line is what loss cost is doing. Anything below the line is a rate cut in real terms.
Renewal rate change is weighted by policy count across the rate-action bands: decrease −5%, flat 0%, up single digit +5%, up double digit +12%. Loss-cost trend of 7.0% is a stated assumption.
Not one programme clears the line. The spread across the whole book is 1.15 points — 4.15%% to 5.30%% — against a shortfall of 1.7 to 2.9 points for every one of them. Reading rate change on its own says the book is raising prices. Reading it against trend says the book is cutting them by about two points a year, everywhere, and has been for some time.
The slice against the rest of Contractors GL
TX and FL placed through RT Specialty and Risk Placement — 32% of the programme's written premium — set beside the other 68%.
Realised rate · soft months
Realised rate · calm months
Loss ratio · rolling 12 months
Audit premium capture
Bars are scaled within each measure; the printed figure on the right is the true value. Rate is dollars of premium per $100 of payroll.
In calm months the slice prices within a whisker of the rest of its programme. In the four soft months it does not. And it carries the consequence: against 66.6% for the other two-thirds of the same book, with audit capture at . Same programme, same class, same carrier — different price and a different result.
Rate change, eighteen months · the slice
The blue line is the rate monitor — renewal rate change from the rate-action flag. The red line is what the book actually wrote, premium per $100 of payroll. The band between them is the part nobody is measuring.
Both series are percentage change against the slice's own calm-month level. The monitor is a count of rate-action categories; the realised rate is money divided by exposure.
The monitor is not blind — it dips in every one of the four months. It is mis-calibrated. In April 2026 it printed −2.3% while the book wrote −18.7%. A field that sorts renewals into four buckets can say how many accounts took a decrease; it cannot say how large the decrease was, and here the answer was .
Where the renewals actually landed
The rate-action mix on renewal business — soft months, calm months, and the book.
Renewal policies only, weighted by policy count. The book writes essentially no decreases; this slice writes 36% of its renewals as decreases, four months a year.
The book as a whole files almost no rate decreases at all. This slice files them on more than a third of its renewals, four months out of eighteen, and then returns to the book's own pattern. That is not market pressure — market pressure does not arrive and leave on a schedule.
Combined ratio by policy year · the slice
Losses belong to the year the policy was written. This is the only view that proves rate inadequacy, because it holds the pricing decision and its outcome together.
2026-Q3 is shown in grey because it is too immature to read — the newest cohort runs hot in almost every programme, and that is development, not a finding.
Eight cohorts, one direction. 2024-Q3 came in at and every cohort since has been worse, crossing break-even at 2024-Q4 and reaching . A single soft month is an incident; eight cohorts walking the same way is a pricing position. Note too what this view does that no monthly loss ratio can: premium written in April earns over the following twelve months, so no calendar month can show a rate cut and its consequence together. The cohort can.
Every KPI behind this brief, with its verdict
Moved and flat metrics together, so the ruled-out explanations stay visible. The last three are here because they did nothing — and that is what makes the rest readable.
| Metric | Actual | Against | Δ | Verdict | Reading |
|---|---|---|---|---|---|
| Renewal Rate Change · book | +4.7% | +7.0% trend | −2.3 pp | outbreak | Every programme is behind loss cost. This is the finding, and it is book-wide. |
| Renewal Rate Change · slice, soft | −1.4% | +3.7% | −5.1 pp | driver | The monitor did fire in all four soft months. Nobody reads it at slice grain. |
| Realised Rate · slice, soft | −11.8% | −1.4% | −10.4 pp | driver | What the book actually wrote. More than double what the monitor reported. |
| Implied decrease depth | −34% | −5% band | 7× | mechanism | The rate-action field records that a decrease happened, never how large. Here it was a third. |
| Realised Rate · programme | −5.6% | −11.8% slice | half | dilution | At programme grain the other 68% of the book halves the signal before it reaches anyone. |
| Combined Ratio · 2026-Q2 cohort | 110.8% | 96.1% in 2024-Q3 | +14.7 pp | proof | Eight cohorts of steady deterioration. This is rate inadequacy once it has earned through. |
| Loss Ratio · slice, 12m | 71.4% | 66.6% | +4.8 pp | outcome | Five points worse than the rest of its own programme, on the same class and carrier. |
| Audit Premium Capture · slice | 88.0% | 89.6% | −1.6 pp | leakage | The cheap accounts also fail to complete audit, so the shortfall is collected twice over. |
| New Business Mix · slice, soft | 43.7% | 36.5% | +7.2 pp | context | More new business in the soft months — but new and renewal both priced down, so mix is not the cause. |
| Exposure per policy | 5,322 | 5,301 | flat | rule-out | Account size did not change. This is a price cut, not a shift to smaller risks. |
| Quote-to-Bind · slice | 42.2% | 41.4% | +0.8 pp | rule-out | Conversion barely moved. The discount bought no measurable volume. |
| Policy Retention · slice | 94.9% | 94.5% | +0.4 pp | rule-out | Nobody was leaving. This was not a defensive cut against a competitor taking the book. |
Rate is premium per $100 of payroll. Audit capture is measured on policies reaching expiry, so it lags the month they were written by roughly a year — the 88.0% relates to business bound in the earlier soft months, not this one.
Three rule-outs carry a lot of weight. Exposure per policy was flat, conversion was flat, and retention held. This was not a smaller class of risk, it did not win business, and it was not defending a book that was walking out the door. It was a discount given for no return.
The whole book is behind loss trend, not just this slice
Renewal rate change runs across the book against a 7.0% loss-cost trend. The best programme is and the worst is . Not one of the fifteen clears the line.
Why it matters: Everything else in this brief is a one-million-dollar problem in one slice. This is a book-wide one, and it is invisible if rate change is read on its own. A +4.7% renewal increase reads as discipline right up until it is set against what claims are doing, at which point it is a 2.3-point annual give-away compounding across $651M of premium.
The monitor is not blind — it is mis-calibrated
In the four soft months the monitor printed while the book wrote . It dipped every time; it just dipped a tenth as far. The rate-action field flagged 36% of renewals as Decrease, and to produce the realised fall those decreases averaged .
Why it matters: This is worse than a blind spot, because a monitor that fires weakly gets dismissed as noise. The fix is not a new report — it is to carry the magnitude of each rate action rather than the category, and to reconcile the monitor against realised premium per unit of exposure every month. The gap between the two lines is the control.
The cohorts are the proof, and no monthly view could have given it
Combined ratio by policy year walks from to across eight cohorts, crossing break-even at 2024-Q4.
Why it matters: Premium written in April earns over the following twelve months, so a calendar month can never show a rate cut and its consequence together — any same-month correlation is an artefact. The cohort view holds the decision and the outcome in one row, and it shows this is not four bad months but a two-year direction.
Programme grain halves the signal before anyone sees it
The slice wrote in the soft months. The same months at programme level read , because the other 68% of Contractors GL was pricing normally.
Why it matters: A monthly rate review reading at programme grain sees a 5.6% dip four times in eighteen months and reasonably calls it noise. The signal only survives at state-by-producer grain. That is a reporting-design conclusion rather than a pricing one: the review has to run at the grain the decision is made at.
The discount bought nothing, which makes re-rating cheap
Quote-to-bind moved 0.8 of a point. Retention held at 94.9%. Exposure per policy was unchanged at 5,322 units. And the accounts leaked again at audit, with capture at against 89.6% for the rest of the programme.
Why it matters: The pricing team will assume re-rating this slice costs volume. The evidence says it will not: a 12% cut moved conversion by less than a point, so reversing it should cost about as little. That turns a contentious renewal conversation into a straightforward one, and it is the most actionable thing on this page.
Which renewals were flagged Rate_Decrease in the four soft months, and at what price each?
The field records that a decrease happened. The policy records how big — that is the review list.
Which months in the last 24 did this slice write below 3.40 per $100 of payroll?
Establishes whether the four-month rhythm is a pattern or three coincidences.
How has each policy-year cohort developed since it was written?
Tests whether the walk from 96% to 111% is still moving or has settled.
Do any other programmes show a gap between renewal rate change and realised rate?
The mis-calibration is a measurement problem, so it will not be confined to one slice.
What did the accounts that took a decrease look like at their next renewal?
Shows whether the discount was recovered later or has become the new base price.
How much of the uncaptured audit premium sits on accounts still in force?
Separates what is genuinely recoverable from what has already lapsed.