The record is real. The growth is not

The record is real. The growth is not.

US annuity sales reached $464.1bn in 2025, up 7%. In the first half of 2026 they reached $228.7bn, a record, but only 1% above the same period of 2025. The second quarter’s preliminary $123.9bn was revised down to $121.2bn in LIMRA’s final survey. Growth sits in market-linked products: in the quarter, registered index-linked annuities rose 22% and variable annuities 24%, while fixed-rate deferred fell 9%.

Two points of caution. Since January 2026, LIMRA counts all money paid into an individual annuity, including exchanges between contracts of the same company. It has not said how much that adds, so a board cannot separate new money from recycled money from the headline.

And demand is not the constraint: this was the eleventh straight quarter above $100bn, and 4.1 million Americans turn 65 each year. Share gains must come from distribution and the insurer’s own book, not from the market.

Where conversion is decided

In annuities the insurer does not own the lead. The advisor does. The insurer controls three conversions: whether an appointed advisor becomes an active producer and how much of that advisor’s business comes its way; whether an expiring contract renews or exchanges within the company; and whether a recommended exchange can be defended to a regulator.

The expiry date is the largest conversion the insurer controls

The SOA and LIMRA study of fixed-rate deferred annuities for 2015 to 2022 found that surrenders peak in the year the surrender charge expires. The 2023 to 2024 update shows the same peak.

Insurers wrote $164.9bn of fixed-rate deferred annuities in 2023 and $165.3bn in 2025. Contracts with guarantees of three to five years written in those years reach expiry between 2026 and 2030. The date is known at issue. The advisor’s status and the rate the owner could earn elsewhere sit in other systems.

The advisor decides what happens at expiry

Cerulli reports that advisors who switch broker-dealers typically lose about 22% of their assets, and that advisors holding 41% of industry assets plan to retire within a decade. What does not get measured is annuity leakage. When an advisor retires or moves, the contract stays on the insurer’s books, but the person who would act at expiry is gone. BCG describes a North American insurer that  that contacted half of the customers left without an advisor before their contracts expired.

Why this is a data problem

An expiry conversation needs contract dates and values, advisor status, recent advisor activity, the competing rate and the suitability history. These sit in policy administration, distribution, pricing and compliance, with different owners and definitions. BCG notes that for most insurers the data in core legacy systems is not ready for AI use cases, and that the answer is a layer that decouples front and back systems. Without one, each new use case tends to rebuild the same connections, and this is why AI programs stall at pilot stage.

A DaaS; data-as-a-service layer is potentially that answer: a small set of governed data products, each with a named owner, service level, quality tests and access policy, sitting between the core platforms and the people and models that act on them. It reads from existing systems and serves APIs, feeds and dashboards. Nothing is replaced, but integrating older policy administration platforms is where most effort will go.

Four data products, in this order

•    Contract event calendar: expiry, rate reset, rider and minimum distribution dates by contract, with the advisor of record and the gap to market rates. It decides which contracts to work, when, and with what offer.

•    Advisor profile: production by product and channel, application activity, contact history, licensing and appointment status. It shows who to call and who has gone quiet.

•    Exchange evidence file: source of funds, replacement history, riders forfeited and surrender charges, held once for sales and compliance. In April 2026 FINRA fined Ameriprise $450,000 and ordered more than $993,000 in restitution over its supervision of variable annuity exchanges, covering 114 customers between 2015 and 2018. Ameriprise noted the settlement did not state the transactions were unsuitable. Supervision needs evidence assembled at the time of the recommendation.

•    Value ledger: net flows against a control group, with acquisition, hedging and retention cost, separating new money from internal exchanges. It shows whether growth is real.

What data will not fix

A data layer does not replace competitive products and pricing. It improves how well an insurer uses them. BCG attributes about 70% of the effort in AI programmes to change management. In practice, every lead carries a reason and talking points, wholesalers get fewer and better leads, compliance co-designs the exchange rules, and a value office reports net results.

Sequence and cost

•    Months 0 to 6: baseline expiry cohorts and advisor activity; agree net-value metrics and control groups with finance and compliance; build the exchange evidence file first.

•    Months 4 to 12: build the contract event calendar and advisor profile for one product and one region; run a retention test against a holdout group.

•    Months 12 to 24: extend to further channels and products, embed in CRM and wholesaler workflow, complete model risk validation, and bring run and monitoring under insurer control.

The program doesn't move to the next phase on a date. It moves only when the current phase has proven three things:

•    Retained net value against control: the customers, that the insurer contacted kept more money with the insurer than a similar group that the Insurer deliberately didn't contact (the holdout or control group). Net means the value left after costs, commissions and any extra benefits offered to keep the customer.  

•    Flat conduct exceptions: Complaints, unsuitable-sale findings and compliance breaches haven't gone up. This matters because retention activity can slide into pushing customers into replacement products that aren't in their interest, which is the behaviour FINRA has been fining.

•    Data quality above an agreed threshold: the contract dates, customer details and advisor records feeding the model are accurate and complete enough to act on. Finance, compliance and the business agree the target (for example 95% of records complete) at the start.

The maths behind the estimates: 50 to 60 people at full scale. Fully onshore at about $160,000 per loaded role, that is $9.6m a year. A hybrid with 12 onshore roles and 48 in an offshore global capability centre at $55,000 to $70,000 each runs at $4.6m to $5.3m, a saving of 45% to 52%.

This holds only if the GCC owns data products and outcomes rather than tickets.

Three questions for the board

•    Expiry: which contracts reach surrender expiry in the next 90 days, who is the advisor of record, and who is calling them?

•    Advisors: which advisors who placed business with us in the last two years have retired, moved or gone quiet, and what happened to their contracts?

•    Growth: how much of last quarter’s growth was new money, and how much was exchanges and renewals within own book?

If the answers take weeks, the constraint is the data, not the models.

The $10bn illustration and the cost figures are estimates built from public benchmarks, for illustration only.