Customers rarely leave at the moment they cancel. They leave weeks or months earlier, quietly, and the cancellation is administration catching up with a decision already made. If you cannot see that earlier moment in your data, you are not managing retention. You are recording it.

Where it actually breaks

In every subscription business I have worked in, the real churn points sat somewhere unglamorous. A failed payment that was retried badly and never followed up in a human way. A delivery that slipped, and a message that never acknowledged it. A second order that arrived before the customer had finished the first, because the cadence was set for the business rather than for consumption.

None of those are pricing problems. All of them are CRM problems, in the broad sense. Data, timing, and whether anyone owns the moment.

The cancellation is administration catching up with a decision already made.

The trust problem underneath

There is usually a layer below that, and it is the one that keeps retention programmes from working. The team does not fully trust the data. Two people can produce two different churn numbers, both defensible, and so every conversation about retention starts with a debate about measurement rather than about customers.

When that is true, people build spreadsheets beside the platform. The platform investment stops returning anything, and nobody says so out loud because it was expensive.

What I would do first

• Agree one churn definition, write it down, and stop calculating it any other way.
• Find the last positive interaction before cancellation, not the reason given at cancellation.
• Look at payment failures separately. Involuntary churn is often a large slice and it is the easiest to reduce.
• Cut the segment list back to the ones that actually change a decision. Most CRM setups carry dozens nobody maintains.
• Put one owner on the lifecycle, end to end. Split ownership is why the gaps persist.

None of that is exciting, and all of it compounds. Retention work is unglamorous by nature, which is precisely why it is usually available. The offer gets attention because it is visible. The plumbing gets attention when someone decides to look.

A test worth running

Take last month's cancellations. For each one, find the moment you would have wanted to intervene. Then ask whether your current setup could have spotted that moment, and whether anyone would have acted on it.

If the answer is no on either count, you have your roadmap, and it is shorter than the one you were about to write.

So the handover matters more than anyone treats it. Done well, it buys back three or four weeks. Done in the usual way, which is a folder of documents and a calendar full of introductions, it buys back very little.

The five things I actually want

• The numbers as the team uses them day to day, not as the board sees them. Those are usually different, and the difference is informative.
• The three decisions currently stuck, and who they are stuck between.
• An honest account of who is leaving, or thinking about it.
• The contracts and commitments that expire inside my window.
• The thing nobody says out loud in meetings.

That last one is the most valuable and the one most handovers leave out, usually because the person doing the handover has stopped noticing it. Every organisation has a subject that gets discussed in corridors and not in rooms. An interim who does not know what it is will walk into it in week two.

Every organisation has a subject that gets discussed in corridors and not in rooms.

What I do not need

A full strategy document. A tour of the tooling. Thirty introductory meetings scheduled before I have any basis for asking useful questions. All three feel like good onboarding and all three consume the week when attention is cheapest to spend well.

I would rather have two days of reading the actual reporting and sitting in on real meetings than a week of curated context.

The mandate question

One thing to settle before the start rather than in week three. What can this seat decide alone, what needs a conversation, and what is off limits. Interim roles fail more often on unclear authority than on unclear priorities.

If the answer is that everything needs a conversation, that is worth knowing on day one, because it changes what is realistic and it changes what I should be hired to do.

And the handover out

The last two weeks are part of the engagement, not a formality after it. The point of an interim is not to keep the seat warm. It is to hand your permanent hire a situation that is better than the one you found, with the ground prepared rather than the mess redistributed.

That means writing down what was tried and rejected, not just what worked. The rejected list saves your next leader a quarter.

If the honest answer is none, you are not running a loyalty programme. You are running a permanent margin reduction with a database attached, and paying for the administration of it.

Why so many end up there

Almost nobody sets out to build a discount scheme. It happens by accretion. A points mechanic launches, it underperforms, so the earn rate improves. Still slow, so a welcome bonus appears. Then a birthday reward, then a tier, then a partner offer. Each step is defensible on its own and the sum is a programme that rewards behaviour customers were going to exhibit anyway.

The measurement usually keeps pace with the drift. Enrolment goes up, so the programme reports well, and nobody asks whether enrolment was ever the point.

Almost nobody sets out to build a discount scheme. It happens by accretion.

What a good programme is actually doing

It changes one specific behaviour, deliberately chosen. Frequency, basket composition, category spread, referral, or the move from a single purchase to a repeated one. One of those, named up front, measured against a group who are not in the programme.

That last part is the bit that gets skipped, and without it you cannot tell the difference between a programme that works and a programme that has attracted the customers who were already loyal.

Design questions worth arguing about

• What is the one behaviour this changes, in a sentence with no conjunctions in it?
• Who is in the control group, and how long do we leave them alone?
• What does it cost per member per year, fully loaded, including the operations of it?
• What happens at the moment of reward, and is that moment memorable or administrative?
• Would a customer describe this programme to a friend without prompting?

The fourth question is the one that separates the programmes people talk about from the ones they merely belong to. A reward that arrives as a line on a receipt is not an experience. It is a rounding error the customer notices once.

The uncomfortable version

Sometimes the right answer is fewer rewards, better placed. Occasionally the right answer is to close the programme and put the margin back into the product or the delivery promise, both of which customers feel more directly than points.

That is a difficult conversation to have about something with a launch behind it and a team attached. It is also, in my experience, the conversation that has been avoided for about two years.

The questions that actually decide it

Three, and none of them appear on a feature matrix.

• What does it cost to operate, in people and hours, once the project team has gone?
• What does it prevent us from doing in two years, and how confident are we that we will not want to do that?
• Can the team we actually have run it on a Tuesday afternoon in December, when two people are on holiday and something has broken?

The third one sounds flippant. It is the one that determines whether the platform is a capability or a dependency. A stack that requires a specialist for every routine change will slow the commercial team down permanently, and that cost never appears in the business case because it arrives as friction rather than as an invoice.

A stack that needs a specialist for every routine change becomes a permanent tax on the commercial team.

Optionality is the expensive part

Every platform choice closes doors. That is not an argument against choosing, it is an argument for knowing which doors. Pricing flexibility, market expansion, subscription mechanics, the ability to run a marketplace or a shop in shop later, how easily data leaves the system. Those are commercial questions dressed as technical ones.

The failure mode is not choosing badly. It is choosing without anyone in the room whose job is to think about year two.

How I would run it instead

Start with the commercial roadmap for the next three years, written before any vendor is contacted. Then ask each option what it costs to deliver that roadmap, including the parts it makes awkward. Feature comparison comes last and settles far less than people expect, because at this point most platforms in a category do most things.

And put an operating cost estimate in the business case. Licence plus implementation is the number everyone quotes. Licence plus implementation plus five years of running it is the number that decides whether this was a good idea.

If you are already mid project

The useful question is not whether to continue. It is what has already been cut from scope, and whether any of it was the reason for doing this in the first place. That happens more often than anyone admits, and it is recoverable if you catch it before launch rather than after.

That loop is the thing that makes a team good. Strategy sets direction, which matters, but direction alone does not teach anyone anything. A team that decides on Monday and sees the consequence by Friday builds judgement at a rate that no planning cycle can match.

What a trading rhythm actually is

A short standing meeting, the same day each week, with the same handful of numbers, and a decision at the end of it. Not a status update. Not a dashboard review. A decision, taken by the people in the room, with a name attached.

• The same five or six measures every week, in the same order.
• Last week's decision revisited first, before anything new is discussed.
• One person who owns the call when the room disagrees.
• Forty five minutes, and it ends on time even when it is interesting.

The second point is where most attempts fail. If last week's decision is not revisited, the meeting becomes a place where decisions are made and then quietly abandoned, and people learn that the meeting does not matter.

A team that decides on Monday and sees the consequence by Friday builds judgement faster than any planning cycle.

Why documents lose

A quarterly plan operates at a timescale where cause and effect are almost impossible to attribute. By the time the quarter closes, twenty things have changed, and the plan is assessed on whether it was delivered rather than whether it was right.

Worse, planning cycles reward comprehensiveness. Every stakeholder gets a line, nothing is dropped, and the team arrives at the quarter with more initiatives than capacity. Then the rhythm that would have caught this does not exist, so the discovery happens in month three.

You need both, but build this one first

I am not arguing against strategy. A team with rhythm and no direction optimises its way into a corner. But if you can only build one thing this quarter, build the rhythm, because the strategy will be better written by a team that has spent twelve weeks learning what actually moves.

It is also the cheaper of the two to start. A rhythm needs a calendar invite and someone willing to hold the standard. That is genuinely most of it.

So the new process runs alongside the old one. Both get maintained, because nobody has been told they can let the old one go and nobody wants to be the person who dropped it. Capacity halves. Delivery slows. And when the programme underperforms, the diagnosis is adoption.

Adoption is usually the symptom

When people are described as resistant to change, it is worth checking whether they have been given the change or simply given more. Those look identical from a distance and feel completely different from inside the team.

A team running two systems is not resisting. It is coping. The resistance framing is comfortable because it locates the problem in people rather than in the plan.

When people look resistant to change, check whether they were given the change or just given more.

The question for every transformation meeting

Which existing activity does this replace? If nobody in the room can answer, the initiative has just added work, and it should not leave the meeting until someone can.

Asked consistently, that question does two useful things. It exposes initiatives that were never going to change anything, and it turns the plan into something a team can actually absorb.

Write the subtraction list

• What stops, specifically, and on what date.
• Who currently spends time on it, and what they do instead.
• What breaks if we stop, and whether we can live with that.
• Which reports nobody will miss, which is always more than expected.

It is a shorter document than the roadmap and more predictive of whether the programme works. It is also considerably harder to write, because every item on it belongs to someone.

Sequencing follows from it

Once you know what stops, the order becomes clearer. Start with the change that frees capacity, then spend that capacity on the next one. Programmes that sequence by strategic importance rather than by capacity tend to stall in the middle, with everything half done and nobody available.

Value arriving early is not only good for morale. It is what keeps the funding conversation on your side in month nine.

I have built attribution models, and the useful thing I learned from doing it is how much of the value came from the arguing rather than the output. Building the model forces a team to say out loud what it believes about how customers decide. That conversation is worth more than the number it produces.

Where precision misleads

The risk with a well built model is not that it is wrong. It is that it is precisely wrong, and precision reads as authority. A dashboard with two decimal places invites decisions that the underlying data cannot support, and the confidence is manufactured entirely by the formatting.

The risk is not a model that is wrong. It is a model that is precisely wrong, because precision reads as authority.

What works better

• Contribution thinking. What would we lose if this channel went to zero, rather than what did it get credit for.
• Holdout tests. Turn something off in one market, deliberately, and watch. Uncomfortable and extremely informative.
• Directional agreement across several imperfect views, rather than one authoritative view.
• A stated tolerance. Decide in advance how wrong a number can be and still support the decision.

Holdouts are the ones teams resist, because switching off spend feels like deliberately losing money. It is, briefly. It is also the only method that answers the question directly, and the cost of not knowing compounds every quarter you postpone it.

The standard to hold

The goal is not the correct number. The goal is a decision you would make again, given the same information. That is a lower bar than accuracy and a far more useful one, because it is achievable this quarter.

It also changes the tone of the conversation. Teams arguing about whose number is right tend to stop arguing when the question becomes what they would each do differently if their own number were true.

One practical habit

Write the decision down with the number that drove it, then revisit it a quarter later. Not to score anyone, but to calibrate. Teams that do this develop a feel for which of their numbers deserve trust, and that feel outperforms any model.

Deep expertise in one channel is the easiest thing to buy in this market, and rarely what the role needs in its first year. What it needs is someone who can read a commercial picture, decide what to stop, and carry a number in front of a board without flinching.

What the job is on a bad week

Worth designing the role around this rather than around the good weeks. On a bad week the job is telling a founder that the channel they are proud of is unprofitable, telling a team that two of their five projects are being cancelled, and then keeping both groups working.

That is a temperament question more than a skills question, and it is very hard to assess from a CV.

Deep expertise in one channel is the easiest thing to buy, and rarely what the role needs in year one.

Questions that reveal something

• Tell me about something you shut down. What did it cost you internally?
• Describe a number you were wrong about, and how you found out.
• What did you inherit that you decided to leave alone, and why?
• Where does your judgement come from in an area you are not expert in?

The third question is the most underrated. Leaders who change everything are easy to spot and often expensive. Leaders who can identify what is already working and protect it from their own instinct to improve it are rarer and considerably more valuable.

If you do not yet know what good looks like

This is common and there is no shame in it, particularly if the role is new. The honest position is that you are trying to hire permanently for a job you cannot yet specify.

Running the seat on an interim basis first is a cheaper way to find out. You get the work done, you learn what the role actually needs day to day, and you write a far better brief for the permanent search. The alternative is a permanent hire, a mismatch discovered in month five, and eighteen months lost.

One structural note

Decide before the search whether this role owns a number or advises on one. Both are legitimate. Advertising the first and delivering the second is how good people leave in year one, and it is entirely avoidable at the job description stage.

The question that matters

Is this capability part of how the business competes, or is it a service the business buys?

If customers experience it, if it compounds with knowledge of your specific business, or if you need to change it weekly, it belongs inside. If it is specialist, cyclical, or benefits from seeing many companies rather than one, it belongs outside.

Most functions sit clearly on one side once the question is asked plainly. The arguments happen when the question is never asked and cost stands in for it.

If it compounds with knowledge of your business, it belongs inside. If it benefits from seeing many businesses, it belongs outside.

What insourcing really costs

• Recruitment, and the months before the seat is filled.
• Management attention, which is the cost nobody budgets.
• A career path, because specialists leave when there is nowhere to go.
• Cover, since one person is a single point of failure with holidays.
• The knowledge that leaves when they do, unless someone made them write it down.

The middle item is the one that sinks otherwise sound plans. Six specialists need a leader who understands their work well enough to prioritise it. If that leader does not exist, insourcing produces a group of individuals rather than a team, and the output is worse than the agency you replaced.

What agencies really cost

Not the retainer. The cost is in context. An agency starts every quarter slightly further from your business than your own team, and closes that gap with meetings you pay for in time. On fast moving work that gap is the whole problem.

There is also the incentive question, and it is worth being adult about it. An agency paid as a share of spend has a view on whether spend should fall. That does not make them dishonest, it makes them human, and the arrangement should account for it.

The hybrid that usually works

Own the judgement, buy the execution. Keep strategy, prioritisation, data and the customer relationship inside. Buy production capacity, specialist depth and peak load from outside.

The failure version of that is owning neither, where an agency sets direction and the internal team coordinates it. Nobody is accountable and everybody is busy.

Nobody is lying. The belief is assembled from a journey map drawn two years ago, a dashboard measuring the happy path, and the experience of colleagues who all have accounts, employee context and the ability to ask someone internally when something goes wrong.

Why internal teams cannot see it

Proximity. After six months inside a business, you cannot use its product like a stranger. You know which button is misleading, so you no longer misread it. You know the delivery message is wrong, so you discount it. The friction becomes invisible precisely to the people with the power to remove it.

The friction becomes invisible to exactly the people with the power to remove it.

How I go looking

• Buy something. Properly, with my own money, on a phone, on a normal connection, without telling anyone.
• Then break it. Change the order, fail a payment, ask for a refund, contact support and time the reply.
• Read the last two hundred support tickets. Not the categories, the actual text.
• Read the reviews between three and four stars. Five stars tell you nothing and one star is often not about you.
• Sit with the support team for a morning and listen to what they explain repeatedly.

The support inbox is the most under used research asset in most companies. It contains a prioritised list of your problems, written by customers, at no cost, and it is usually read only for volume and response time.

What the gap looks like when you find it

Rarely a broken page. Usually a promise made in one place and quietly not kept in another. A delivery window on the product page that fulfilment cannot meet in practice. A cancellation described as easy in the marketing and routed through a form in reality. A subscription that says pause anywhere and offers it in one place only.

Those are not UX problems and they cannot be fixed by the team that owns the page. They are organisational gaps that show up as customer experience, which is why they persist through several redesigns.

The habit worth building

Once a month, someone senior buys something and writes down what happened. Not a report, half a page. It is the cheapest research a company can run and the first thing to disappear when things get busy.