Our Ecosystem
← Blog·CMO

How to Build an LTV:CAC Model Finance Trusts

David Manela··9 min read
A set of brass calibration weights in their fitted case beside an open logbook, with a pen resting across the page.

A model earns trust the way an instrument does. Someone checked it against something known.

Most LTV:CAC models are arithmetically correct and still get discounted the moment they reach the CFO. The formula is rarely the problem. The problem is that nobody can say where the assumptions came from, when they were last checked, or what would have to be true for the number to be wrong.

An LTV:CAC model finance trusts is one that has been validated, not just built. That means three things: the design and its assumptions are written down, the model is monitored against the changes that would invalidate it, and its predictions are compared with what the cohorts actually delivered. Ownership and independent challenge matter as much as the inputs.

Why does finance discount a model that is arithmetically correct?

Your CFO is not checking your multiplication. They are asking whether the number survives contact with the general ledger, and whether it will still be true next quarter. Most of the argument about inputs is already settled: lifetime value has to be built on contribution margin rather than revenue, and acquisition cost has to be fully loaded. That ground is covered in the metrics that actually decide whether ecommerce growth is profitable. The live question is different. It is whether anyone can show their working.

A discount applied to a marketing number is a rational response to missing provenance. The CFO has no way to audit a figure whose assumptions live in one analyst's head and whose last sanity check was the day it was built. So they haircut it before it reaches the board, and the haircut gets larger each quarter the model goes unexamined.

What does validating a model actually mean?

It is worth borrowing the vocabulary from the people who do this for a living. In April 2026 the US banking agencies issued revised interagency guidance on model risk management, which superseded guidance that had stood since 2011. It describes validation in three parts: conceptual soundness, ongoing monitoring, and outcomes analysis.

Two caveats before anyone over-applies it. That guidance is aimed at banking organizations with more than $30 billion in total assets, and it says in terms that it does not set forth enforceable standards or prescriptive requirements. Your company is not in scope. Borrow the structure, not the obligation.

The second caveat is more uncomfortable. The guidance excludes simple arithmetic calculations of the kind found in spreadsheets from what it calls a model at all. By that definition, a great many LTV:CAC models would not qualify as models. If yours is a tab with a retention percentage typed into a cell, the honest first step is not validation. It is building something there is a point in validating.

How do you establish conceptual soundness?

Conceptual soundness means assessing and documenting the model's design: the key modelling choices, the assumptions, and the data selected to support them. For an LTV:CAC model, the practical form of this is an assumption register. One page, one row per assumption, four columns: what the assumption is, where the number came from, who owns it, and when it was last reviewed.

The assumptions that most often go unwritten are the ones that move the answer most:

  • The shape of the retention curve beyond the period you have data for
  • The time horizon the model looks forward, and why that horizon and not a longer one
  • Whether future margin is discounted at all, and at what rate
  • The margin percentage applied, and whether it is current or historic
  • How refunds, returns and failed payments are treated
  • Whether organically acquired customers sit inside or outside the denominator

There is a simple test for whether this has been done properly. Hand the model to someone who did not build it and ask them to list its assumptions. If they cannot, the model is not documented, whatever the file says.

How do you monitor the model as the business changes?

Ongoing monitoring is an evaluation of whether a model is still performing as expected given changes in products, exposures and market conditions. An LTV:CAC model is fitted to a particular customer mix. When the mix moves, the model keeps producing numbers, and they quietly stop describing the business.

So write the trigger list alongside the review cadence. A quarterly scheduled review suits most companies. The triggers matter more than the calendar: a pricing change, a change to refund or cancellation policy, a new acquisition channel passing a meaningful share of spend, entry into a new market, a change in payment terms. Any one of those is a reason to revalidate before the quarter is up.

How do you prove the model was right?

Outcomes analysis compares model outputs with the corresponding real-world outcomes. For an LTV:CAC model that means taking a cohort the model valued twelve or eighteen months ago and setting the predicted contribution margin beside what that cohort actually delivered. How long you have to wait before the comparison means anything is set by your payback window, which is why payback benchmarks by stage are worth establishing before you start.

This is the step almost nobody completes, and the reason is mundane. It requires having written the prediction down at the time, in a form you can go back to. Teams that have not kept that record cannot run the comparison at all, however good their current model is. If your cohorts are too young to support the comparison yet, predicting lifetime value before you have years of cohort data is the interim approach, and the predictions it produces are exactly the ones you should be recording for later.

Keep every comparison. A single favourable result proves very little. A two year run of comparisons, including the ones where the model was wrong and by how much, is what converts a spreadsheet into something a board will underwrite.

Who owns the model, and who is allowed to reject it?

Model governance asks for clear roles with well-defined accountability, and for attention to conflicts of interest between the people who build a model and the people who validate it. It also asks for effective challenge, meaning review by objective experts with enough independence to stay objective and enough standing to actually change something.

Translated into a growth team, that is uncomfortable but simple. In most companies marketing builds the LTV model and is also measured on what it produces. That is precisely the conflict the principle exists to manage. It does not mean marketing should not build it, because marketing holds the channel and cohort knowledge that makes the model any good. It means the person who signs it off cannot be the person whose budget depends on the answer.

The workable split is finance owning validation, marketing owning the build, and both names appearing on the assumption register. If nobody in the business is permitted to reject the number, the number is not validated. It is just published.

How much rigour does your model actually need?

A risk-based approach scales the oversight to what is at stake. Model purpose together with model exposure determines materiality, and more material models warrant more comprehensive and rigorous oversight. The translation is direct: govern the model in proportion to the capital it directs.

A model steering two hundred thousand dollars a year does not need quarterly independent review. A model steering forty million dollars a year needs considerably more than one analyst and a shared drive. Companies tend to fail this in both directions at once, over-engineering a small decision while a nine-figure allocation runs on a file nobody has opened since the planning cycle.

What does a model finance trusts look like in practice?

Six artifacts, none of them sophisticated:

  1. One page defining lifetime value and acquisition cost, agreed by both functions and dated
  2. An assumption register with an owner and a review date against every line
  3. A version history, so a number in an old board deck can be traced to the model that produced it
  4. A monthly reconciliation between the revenue the model assumes and the revenue finance recorded
  5. A quarterly outcomes comparison on matured cohorts, kept whether it flatters the model or not
  6. A named validator who does not report to the person who built it

What makes this work is not the sophistication of any single piece. It is that the same six things are produced on the same cadence for long enough that the CFO stops applying a discount, because the record of the model being checked is now longer than their doubt about it.

Exactius has measured $3.98B+ in revenue through this kind of infrastructure. Almost none of the difficulty sits in the mathematics. It sits in agreeing the definitions once, writing down what the model assumes, and then being willing to check it against what actually happened.

Sources

If your LTV:CAC number gets discounted every time it reaches the board, the gap is usually governance rather than mathematics. Book a call to walk through what your model would need to survive that review.

Exactius is a full-funnel growth agency accountable for its clients' P&L. Its AI-enabled senior operators provide performance marketing, strategy, creative, and whole-business analytics and data science, engaged one function at a time or as a full team. It serves consumer and B2B companies where paid marketing is a main growth lever, through two practices: one for companies from $5M to $100M and one for companies from $100M to $1B.

Tags:LTV:CACmodel validationunit economicsCFOcontribution margincohort analysismarketing financegovernance
D

David Manela

David Manela is the founder of Exactius and creator of the Growth Operating System — a framework for deploying capital-efficient, compounding growth inside scaling companies.

FAQ

Frequently asked

What is the right way to build an LTV:CAC model finance can actually trust?

Build it to be validated rather than merely correct. Three things do most of the work. Document the design and every assumption it rests on, including who owns each one and when it was last checked. Monitor it against the changes that would invalidate it, such as a shift in customer mix, pricing or refund policy. And compare what the model predicted against what the cohort actually delivered once it matured. Add one more: someone who is not measured on the output has to be able to reject it.

Why does my CFO discount our LTV:CAC number?

Usually not because the arithmetic is wrong. A CFO discounts a number when they cannot see where its assumptions came from, when they were last tested, or what would have to be true for the number to be wrong. A model with no assumption register and no record of being checked against actuals is a point estimate with a decimal place on it. The discount is a rational response to missing provenance, not a lack of faith in marketing.

How often should an LTV:CAC model be revalidated?

On a fixed cadence and on triggers, not on whichever comes to mind first. A quarterly review suits most companies for the scheduled part. The trigger list matters more: a pricing change, a change in refund or cancellation policy, a new acquisition channel passing a meaningful share of spend, entry into a new market, or a change in payment terms. Each of those alters the customer mix the model was fitted to, which means the model is describing a business that no longer exists.

Who should own the LTV:CAC model?

Split the build from the validation. In most companies marketing builds the model and is also measured on what it produces, which is the conflict of interest that model governance exists to manage. That does not mean marketing should not build it, since marketing holds the channel and cohort knowledge. It means the person who signs off on the model should not be the person whose budget depends on the answer. In practice finance owns validation, marketing owns the build, and both sign the assumption register.

What is outcomes analysis for an LTV:CAC model?

Comparing what the model said a cohort would be worth against what that cohort actually delivered. It is the step almost nobody completes, because it requires having written the prediction down at the time and then waiting long enough to check it. The waiting period is set by your payback window. If CAC pays back in nine months, a comparison run at three months tells you very little. Keep the record of each comparison, because the history is what eventually earns the model its credibility.

Does bank model risk guidance actually apply to a company our size?

No. The revised interagency guidance issued in April 2026 is aimed at banking organizations with more than $30 billion in total assets, and it states plainly that it does not set forth enforceable standards or prescriptive requirements. It is worth borrowing anyway, because it is the vocabulary your CFO and your board already use for deciding whether a model can be relied on. One detail is worth sitting with: that guidance excludes simple arithmetic calculations of the kind found in spreadsheets from what it calls a model at all.

Related Reading

Keep going

Ready to fix the system?

Your growth system is either compounding or degrading.

Book a diagnostic call. We'll identify where your growth system is breaking and what it's costing you.

Book a call← More articles