Insurance marketing budgets, set by what a policy is worth.

The sections below cover how to turn policy value into a budget, which outcome to model, how to handle several products under one brand, and how comparison sites, partners, premium changes and TV fit in.

Insurance has one advantage most industries lack: customers renew. A new policy is worth its first-year premium plus every renewal that follows, minus claims and costs. Insurers know their retention and margins well, so the value of a new policy over a chosen horizon is a number they can put on the table.

That number is the ceiling for what the next policy may cost to win. The model estimates what the next policy costs in each channel at today's spend; the budget level is where that marginal cost meets the ceiling. Because renewals stretch the value over years, the payback horizon you choose matters a lot. A short horizon gives a cautious budget, a long one an aggressive budget, and the scenario planner shows both side by side.

Quotes are easy to count and close to the marketing, and they are not revenue. Signed policies are revenue, and they come later. New premium weights each policy by its size, which matters when a car policy and a home policy are worth very different amounts.

When it's unclear which to use, both go in and the first model is built on both. The choice is then made on what each shows, and it can differ by market. The delay between marketing and a signed policy is part of the model setup, so each policy is credited to the marketing that started it.

Most insurers market car, home, travel and personal insurance under one brand, and much of the advertising is for the brand rather than one product. It is tempting to build one model per product. In practice that splits the data thin and forces the model to decide which product a brand campaign was for, which it can't know.

The usual answer is one model per market on total new business, with product as a dimension in the data. Campaigns aimed at a single product go in as their own channels where the spend is large enough. How much a product-specific campaign lifted that one product is then an analysis on the data, outside the budget model.

Insurers sell through several routes besides their own site: comparison sites, partner agreements, car dealers, banks and agents. Two questions decide how each is handled.

First, how you pay. A comparison site paid per sold policy behaves like pay-per-sale affiliate: the cost follows the sales, so a model that treats it as an ordinary channel credits it with sales it did not cause. It stays in the model at its historical spend and is locked in the recommendations. A comparison site or partner bought as media, at a fixed cost or per click, can be modeled as a channel.

Second, whether marketing drives the sale. Policies sold through partners and banks belong in the outcome when your marketing influences them. Left out, the model can't see a large part of what the advertising does.

Premium increases and campaign prices move volume on their own. Large, planned price changes are marked in the model, so their effect is not credited to the media that ran at the same time.

Beyond that, we keep controls few. Every extra variable takes explanatory power away from the media, and anything used to explain the past also has to be known when the model forecasts. Your own planned premium changes are known ahead. Competitors' prices and your ranking on comparison sites usually are not, so they stay out of the forecasting model.

Insurance demand has seasons, and they differ by product. Travel insurance follows holiday planning, car insurance follows car sales, and many people review their policies when a renewal letter arrives. A flat monthly budget ignores all of it.

The model separates seasonal demand from the effect of the marketing, so a strong month is not read as a strong channel. It then shows in which months the next part of the budget works hardest, and the plan can follow that. See how channel response can change through the year.

Insurers are among the largest TV advertisers, because trust and recall decide where people go when it is time to switch. TV's near-term pull, the extra searches, quotes and policies in the weeks after a burst, shows up in the model as TV's own effect.

The slower effect, a brand people think of first years from now, is harder. For that, the model can be extended with a brand component that captures sales arriving months later. Many insurers also lock TV at the planned level and let the model optimize the rest; the scenario planner then shows what the lock costs in forecast sales. See how we model brand effects or read how to measure TV with MMM.

Odins is a managed marketing mix modeling platform. We connect the ad platforms, set up the pipeline for TV and other offline media with you or your media agency, structure your new business by market and product, build and maintain the model, and review every recommendation before it reaches you. Data flows in continuously and the model is retrained once a month, while scenarios and forecasts are yours to run whenever you need them, in the platform or from your own AI assistant.

Moving budget toward channels with headroom typically gives 5 to 15% more effect from the budget you already have. See the managed service.