What if your retention strategy starts working only after the customer has left?

PROBLEM ONEThe rule does not retain

The first is timing. The penalty only takes effect once the customer has already gone. For it to change anyone's mind it has to be known in advance, and mostly it isn't.

The largest test of a pure forfeiture rule ever run is not in telecoms. It is in American company pensions, where leaving a job too early costs you your employer's contributions — real money, on average 40% of the balance. Vanguard studied 4.7 million people leaving jobs across roughly 1,500 plans between 2010 and 2022. The effect on whether people quit was zero. Separation rates did not move around vesting dates.

The reason is in the same study. Only 33% of participants could correctly identify their own vesting schedule.

You cannot scare someone with a punishment they have never heard of.

That transfers directly. A customer who has never read the clause voiding their insurance at port-out is not held by it. They find out afterwards — at which point it is not a retention device, it is a grievance.

And there is no evidence for it either. No operator, in any industry, has ever published a measured retention result for a forfeiture rule. It is defended in public with assertion — loss aversion, breakage, "a natural reason to contact members" — and with accounting and legal rationale. Not with measurement.

That is a claim about the public record, not about what sits in anyone's internal decks. Operators may well have measured this. None of them has shown the working.

PROBLEM TWOAnd it does not have to be this way

The second problem is that we have been running two decisions on one axis, and the word hiding them is bundle.

We talk as if products must either live together or live apart. But selling them together and killing them together are completely different decisions, and only one of them earns anything.

Sell it together. Don't kill it together.

Selling the add-on together with the main product creates value. The UK's FCA measured this twice. In a 2013 controlled experiment, 65% of people who were offered an add-on during the main purchase accepted the first offer. Only 17% did so when they had to shop for it separately. Selling at the same time also increased the price paid by 15%.

In 2018, the FCA studied what happened after it required a pause between buying a car and buying GAP insurance. Add-on sales fell by 19.8%. Separate sales increased only from about 6% of the market to 9%. Most of the lost add-on sales did not move to another seller; they disappeared.

My conclusion is simple: selling together works. Ending together is another decision. I could not find a published measurement showing that this second decision creates value.

What this evidence cannot tell you

Here is where most articles on this subject quietly cheat, and where I nearly did.

The case against fusion has two halves. One half has telecom evidence behind it. The other half has none at all.

The figures usually quoted here — 42% of cancelling customers returning within twelve months, a second subscription lasting 98.7% as long as the first — are from streaming and newspapers. They are real measurements. They are not measurements of mobile.

And mobile is exactly the category where that distinction has already bitten. Ofcom found the bundling effect in fixed broadband and not in mobile, in the same dataset, the same year, from the same respondents. If the effect does not even survive the move from fixed to mobile, it is not safe to move a number from Netflix to a SIM.

Nor is the direction obvious. Porting out might make returning harder — more friction, a new contract, a device commitment. Or easier, because you keep your number and never have to re-register it with your bank, your contacts and every service that texts you a code. Nobody has measured which, in either direction.

So this note is not going to tell you the answer. It is going to tell you the threshold you would have to clear, and how to find out whether you clear it.

How to find out in one month

The keep-rate is unusually cheap to measure, for a reason that is easy to miss: under fusion it is zero by construction. The add-on is terminated, so nobody can keep it. There is no counterfactual to build and no control group to run. You unfuse, and you count.

That is why it reads in weeks rather than a year. Measuring a difference between two arms needs roughly four times the sample of measuring one number, and the return rate is defined over twelve months, so most of it has not happened yet when you want to decide.

1. Stop terminating the add-on at port-out. For a random half of leavers, or for everyone — either works, because there is nothing to compare against.
2. Let two billing cycles run. Anyone still paying in cycle two is a keeper.
3. Count them. Compare against the threshold above for your churn rate.
4. If it clears, you are done. The surviving add-on alone outweighs everything fusion was buying you. You never need the return data.
5. If it does not clear, keep the cohort running and read the return rate at six and twelve months. You lose nothing by having started with the fast test — the sample accumulates either way.

One important exception

I found one strong argument for fusion, but it is not a retention argument.

If customers can keep an insurance product after leaving, the people who choose to keep it may be more likely to make a claim. Several studies show this effect. In the United States, claims under health insurance continued after employment were 46% higher than claims for current employees. In group life insurance, mortality was 18% higher among people who chose additional cover. This made actual costs 9.7% higher than they would have been with a random group. The FCA also found a 52% claims ratio for travel insurance sold as an add-on, compared with 42% for separately purchased insurance.

For insurance products, simply allowing the policy to continue at the same price may therefore be a poor design.

However, automatic termination is not the only solution. It removes the higher insurance risk, but it also removes all future revenue from the customer. The more direct solution is to set a new price and assess the risk again for customers who want to continue.

This problem does not apply in the same way to products without insurance claims and with low additional cost: cloud storage, content, software, data allowances or loyalty balances. For these products, a customer who continues after leaving mobile is mainly an additional source of revenue.

What would change my mind

Three numbers would settle this, and none of them is public.

A keep-rate from any operator that has stopped terminating an add-on at port-out.

One number, two billing cycles, no control group. It would be the first published anywhere, and it would make this whole note redundant — which is the point.

A mobile-specific return rate.

What share of customers who port out later come back. No operator discloses it, no regulator tracks it, and no analyst house publishes it. Until someone does, every quantified claim about the value of preservation in mobile — including the ones I would like to make — is borrowed from another industry.

A claims ratio for device insurance continued after the mobile line ends.

The only place the adverse-selection objection can be settled rather than bounded.

If you have run any of this, I would rather see your number than keep reasoning around the absence of one.