Commercial property and casualty premiums fell 1.2% in the first quarter of 2026, the first overall decline the Council of Insurance Agents and Brokers has recorded since Q3 2017. Property catastrophe reinsurance then dropped roughly 20% at the June 1 renewal. This is the first soft market to begin with algorithmic pricing already installed at the largest carriers, and the open question is whether that precision holds the cycle up or takes it down faster.

Key Takeaways

  • 9 of the tracked lines posted premium decreases in the CIAB Q1 index, commercial property leading at -5.8%. Commercial auto was the sole holdout, up 5.8% for a 59th consecutive quarter.
  • 91.9 combined ratio for the personal auto industry in Q1 2026, the best since 2006 and the quarter in which each of the seven largest personal auto writers cleared $1 billion of underwriting gain, is the catalyst for the softening rather than a defense against it.
  • 86% of the top ten's premium growth in 2025 went to Progressive, $8.9 billion of a combined $10.4 billion, which is share redistribution rather than market expansion.
  • 12 of the 16 largest P&C carriers are expected by sell-side analysts to post year-over-year combined ratio deterioration in 2026, against an industry combined ratio S&P projects near 99%.

The Correction Is Uneven by Design

The headline decline is small and the distribution underneath it is not. Large account premiums fell 2.7% and medium accounts 1.9%, while small accounts still rose 1.1%, itself a 60% deceleration from Q4 2025's 2.8% increase.

MetricQ1 2026Prior PeriodChange
CIAB average premium change-1.2%+1.1% (Q4 2025)First decline since Q3 2017
Commercial property premiums-5.8%-3.2% (Q4 2025)Accelerating declines
S&P GMI underlying growth (H1 2026)-3.7%+5% (2025)Sharpest deceleration in years
Personal auto industry CR91.996.8 (Q1 2025)Best since 2006
Property cat reinsurance rates (June 1)-20%-10% (Jan 1)Accelerating softening

Personal lines tell the same story from the other side. Progressive's trailing-twelve-month direct written premiums reached $70.2 billion, passing State Farm as the largest U.S. auto insurer after 84 years, yet its own growth decelerated from 21% in full-year 2024 to 6% in Q1 2026. State Farm answered with $4.6 billion of rate reductions across 40 states and a $5 billion customer dividend in February 2026.

Those two responses are not the same action at different scales. Progressive slowed a growth rate; State Farm set a price floor that competitors match or concede share against. The distinction between them is the subject of the rest of this piece.

Class-Cell Pricing Cannot See What ML Pricing Selects

Traditional ratemaking treats risks inside a territory-class cell as homogeneous. Loss costs come off historical triangles, trend and expense and profit load go on top, and the resulting rate applies uniformly to everyone in the cell. When the market softens the carrier has two moves: cut the whole book, or hold and lose new business volume.

An ML pricing stack scores risks individually and adjusts continuously. The same carrier can cut price on the segments its models identify as carrying margin while tightening the segments where loss ratios are deteriorating, and the portfolio premium per exposure can look flat while the composition shifts toward higher-margin risks.

State Farm's $4.6 billion is the illustration. Portfolio-wide reductions cut price equally on the best and worst risks inside each class cell. A competitor pricing at the risk level takes the best of those risks at a marginally lower price and declines the rest, and what the class-cell carrier retains is the residual. This is adverse selection running at the carrier level rather than the policyholder level, and Progressive's capture of 86% of the top ten's premium growth is what it looks like from outside.

The reserving consequence is the part without precedent. A carrier whose Q1 loss ratio improves because its model declined risks it would previously have written has a book with no development history at that composition. The improvement is real and its persistence is unestablished, because it depends on a selection advantage that erodes as competitors deploy comparable models. Reserving through a soft market normally means judging rate adequacy against a stable book; here the rate level and the book composition are moving at once, and the historical development factors describe neither.

Scale sets who can do this. Progressive's annual IT spending is estimated at $2.2 billion, supporting more than 500 ML and data science staff across over 100 distinct models, trained on more than 14 billion miles of Snapshot telematics data accumulated since 2008. A regional writer working from bureau rates with modest internal adjustments is not behind on a project. It is holding a different instrument.

Precision Argues Both Ways and Nobody Can Audit It

The discipline case is the intuitive one: a model that says a segment needs a 95 combined ratio to break even stops an underwriter pricing it at 102 for volume. Algorithmic guardrails against the competitive pressures that drove prior cycles below technical thresholds.

The same capability supports the opposite behavior. A carrier that identifies 10 points of margin above breakeven in a segment has both the ability and the incentive to give back 5 of them, and that reduction is technically adequate while signalling to the market that lower prices are sustainable. Repeat across hundreds of segments and the aggregate rate decline is faster and better synchronized than the 1998-2001 or 2006-2009 entries, which moved at the speed of quarterly reviews and annual filings.

Travelers ran an 88.6% consolidated combined ratio in Q1 2026 against 102.5% a year earlier, with core return on equity of 19.7%; Chubb ran 84%. Margin at that level is what a model reads as room to move.

The supervisory side has not kept pace with either reading. The NAIC's AI Systems Evaluation Tool pilot opened across 12 states on March 2, 2026 and runs through September, and the NAIC's March 2026 issue brief asks that models be transparent, explainable and auditable. Gradient-boosted models and neural networks are difficult to explain at the individual risk level, which is exactly the level at which the selective rate reductions are being made. Regulators currently lack the technical infrastructure to independently validate carrier ML models at scale.

That gap has a second effect nobody filed for. Documentation, bias audits and AI governance staffing cost roughly the same in absolute terms at $2 billion of premium as at $50 billion, so the compliance burden lands as an expense ratio penalty inversely proportional to size. The regime built to check algorithmic pricing prices the smaller carriers out of using it.

Further Reading

Sources