The Seduction of CPL as a Metric

Cost per lead as a performance measure has a lot going for it, it's easily understandable, it's relatively easy to track and attribute across channels and property boundaries, easily comparable, computation is fast and the outcome period between marketing nudge and lead outcome is fairly short. Unfortunately it is also extremely misleading as a target for insurance acquisition because it prices the top of the funnel while all the economics sit at the bottom. The right target is allowable acquisition cost per segment, derived from expected lifetime underwriting contribution. CPL should fall out of that calculation as a derived constraint, never lead it.

That is the whole argument. The rest of this note is the machinery for actually doing it, and the reasons your current reporting fights you.

What CPL gets right

CPL is a great metric in exactly one sense: simplicity. You launch a campaign Monday, by Tuesday you know whether leads cost $18 or $42, and by Friday someone has made a budget decision off it. Every platform reports it natively. Every agency is happy to be measured on it, because it is the one number they can most directly move.

The problem is what CPL cannot see. A lead is not a policyholder, and a policyholder is not profit. The gap between lead and profit is vast. Between a form fill and a profitable customer sits a chain of conditional events:

• the lead gets contacted, or quoted, or starts a checkout
• the quote converts to a bound policy
• the policy survives past the early-cancellation window
• the policyholder renews, once, then again
• the claims experience of that cohort lands somewhere sane

Each link has a rate, and those rates vary wildly by source. When you optimize CPL, you are implicitly assuming the rates downstream are constant across sources. They never are. We have yet to see a single account where the cheapest lead source and the most profitable lead source were the same source. Not once.

The chain, written down

Here is the spine of the framework. Nothing in it is exotic. The discipline is in writing it down per segment and refusing to collapse it back into one blended number.

Stage	Quantity	Typical owner of the data
Lead	CPL	Lead Gen/Ad platform
Quote	lead-to-quote rate	Marketing automation / CRM
Bound policy	quote-to-bind rate	Policy admin system
Earned premium	annual premium per policy	Policy admin / finance
Retention	expected policy lifetime in years	Actuarial / finance
Losses and expenses	loss ratio, expense ratio	Actuarial / finance

The economic value of an acquired policyholder, in its simplest usable form:

Lifetime contribution ≈ annual premium
                       × expected policy lifetime (years)
                       × (1 − loss ratio − variable expense ratio)

This is deliberately the same shape as how an actuary thinks about a policy: premium in, expected losses and expenses out, over a duration. The marketing twist is that you compute it per acquisition segment rather than per rating cell. We wrote up the pricing analogy in more detail in Pricing a customer the way an actuary prices a policy.

From lifetime contribution you derive allowable CAC. If you want a 12-month payback you cap it harder; if you can fund a 24-month payback you loosen it. Either way, the cap is an output of the economics, not a number someone anchored on in a planning meeting two years ago.

And only then, at the very end, do you derive an allowable CPL:

Allowable CPL = allowable CAC × lead-to-bind rate

Notice what happened. CPL did not disappear. It became a per-segment derived constraint. A $45 lead from a source that binds at 12% is cheaper, in the only sense that matters, than a $19 lead from a source that binds at 2%.

A worked example, with dummy numbers

The numbers below are illustrative, picked to be roughly the right shape for a direct-to-consumer personal-lines product. Substitute your own.

Take a product with $480 annual premium, an expected policy lifetime of 2.5 years, a 62% loss ratio and 12% variable expenses. Lifetime contribution is 480 × 2.5 × 0.26 = $312.

Say you are willing to spend up to 55% of lifetime contribution to acquire a customer. Allowable CAC is about $172.

Now two lead sources:

• Source A: $19 CPL, 2.1% lead-to-bind. Effective CAC: $905.
• Source B: $45 CPL, 11.8% lead-to-bind. Effective CAC: $381.

Both are underwater against the $172 cap, which is itself a useful and slightly uncomfortable finding. But the ranking is the point: the "cheap" source is more than twice as expensive as the "expensive" one. A team managing to CPL would scale Source A and kill Source B. The economics say to do exactly the opposite, and then to go fix the bind rate or the funnel before scaling either.

This inversion is not a corner case. It is the normal situation, because the things that make leads cheap (broad targeting, incentivized clicks, vague creative that attracts the curious rather than the in-market) are the same things that make them not bind.

The latency objection, and the actuarial answer

The standard objection: lifetime contribution takes years to observe, and you have a budget meeting on Thursday.

True, and actuaries solved this problem a century ago. You do not wait for the ultimate; you estimate it from early development and you weight your estimate by how much data you actually have. In actuarial language this is credibility weighting. In practice for marketing it looks like:

1. Use early funnel transitions (lead-to-quote, quote-to-bind) as same-week signals. These stabilize fast.
2. Use early retention (30-day, 90-day cancellation) as a leading indicator of policy lifetime. A source whose policies cancel at 90 days at twice the book average will not magically develop a normal lifetime.
3. For loss ratio, start with the book-level assumption for the product, and only let a segment's own claims experience pull the estimate away from book once the segment has enough exposure to mean anything.

Point 3 matters more than people expect. Early claims experience on a small cohort is mostly noise, and reacting to it produces thrash. Anchoring to book and letting credibility grow with exposure is the boring, correct answer.

We will be honest about the part we are least sure of: how much of the segment-level variation in loss ratio is real versus selection noise that washes out at scale. We have seen segments where it persisted for years and segments where it evaporated in a quarter. The framework handles both cases gracefully as long as you let credibility do its job, but anyone who tells you they can predict which case you are in is guessing.

Why your reporting fights you

The framework needs five or six numbers per segment. None of them are exotic. Almost nobody has them in one place.

Marketing data lives in ad platforms and a web analytics tool. Quotes live in your insurance core, marketing automation or the CRM. Policies, premium, retention, and claims live in the policy admin system or a finance warehouse, usually keyed in a way that has never been joined back to a click. The join is the actual work, and it is unglamorous: identity resolution between lead records and policy records, consistent campaign taxonomies, and an honest reconciliation between what the platforms claim and what the CRM observed. That reconciliation is its own swamp, which we cover in Why your ad platform numbers will never match your CRM.

If you take one operational action from this note, make it this: get policy-level outcomes joined to acquisition source, even crudely, even for just one product line. Even if your attribution rate is low to start. A crude join beats a precise CPL every time.

What to do differently tomorrow

A realistic sequence, in order of effort:

1. This week. Stop reporting blended CPL as a headline. Report CPL by segment next to lead-to-quote and quote-to-bind rates for the same segment. The inversions will be visible within a month.
2. This month. Write down the lifetime contribution formula for each product with finance's numbers, even at book level. Derive allowable CAC and allowable CPL per segment. Put the allowables in the same report.
3. This quarter. Build the lead-to-policy join. Start tracking 90-day cancellation by acquisition segment.
4. Ongoing. Let segment-level loss experience earn credibility slowly. Revisit allowables when product pricing or retention assumptions move.

None of this requires new tooling. It requires deciding that the unit of account is the policyholder, not the lead, and rebuilding the weekly conversation around that.

The teams that do this end up making strange-looking decisions with complete confidence: paying $60 CPLs while competitors brag about $20s, killing "efficient" campaigns, concentrating budget in segments their old reports ranked last. Then the loss ratios and renewal numbers come in, and the decisions stop looking strange.