The Case for a Single Source of Growth Truth

A single source of growth truth is one agreed set of definitions, with one owner each and one place to read the figures. It is worth saying first what it is not, because that is where most attempts go wrong: it is not another dashboard. A growth team with four disagreeing numbers that builds a fifth place to look has acquired a fifth number. Last updated: October 2026.

Omniconvert has measured how growth programmes are steered across the CROBenchmark dataset of 7,000+ websites in 15+ industries, against 248+ audit criteria, over 13 years in eCommerce. The teams that move quickly are not the ones with the best instrumentation. They are the ones where nobody argues about which number is the number, and that is a smaller and more political achievement than a data project.

This piece sets out what the disagreement actually costs, why four systems can all be right at once, and the five definitions worth standardising before anything is built. It sits alongside the argument that the growth disciplines share one input in CRO, creative & AI visibility tie into one growth system, and the wider picture in state of DTC growth in 2026.

The reconciliation tax

It helps to describe the symptom precisely, because teams rarely name it as a problem. They experience it as a meeting that runs long.

The shape is familiar. A weekly growth meeting opens with a figure on a screen. Somebody from performance marketing says their platform shows something different. Somebody from finance says neither matches what was invoiced. Twenty minutes go on establishing which number everyone will accept for the purposes of the next thirty, and a decision that needed one input has consumed two.

That is the direct cost. The indirect one is larger. A decision that requires a number nobody trusts gets deferred, and it is deferred by a mechanism nobody notices: not a refusal, but a request for more analysis. The analysis arrives, it disagrees with something, and the cycle repeats. An organisation in this state is not short of data. It is short of agreement, and agreement cannot be gathered.

There is a third cost, which I think is the most damaging. When numbers are contestable, arguments are won by seniority rather than by evidence, because the evidence is always arguable. A team learns that the way to get a decision is to be persuasive rather than to be right, and that habit survives long after the instrumentation is fixed.

Why four systems disagree, and why none of them is broken

This is the part that has to be understood before any policy makes sense, and it is usually skipped because it sounds like an excuse.

An ad platform attributes a conversion to an impression or click it served, within a window it chose, to a person it recognised by its own identity graph, reported against its own timezone. Every one of those four choices is reasonable from where it sits, and every one differs from your store's.

Your store counts a completed order against the session that produced it, recognises the person by the account or the device it can see, and reports on your clock. Also reasonable, also different.

Your analytics tool makes a third set of choices, usually about sessions and channel grouping, and then applies them retroactively when its model updates. The differences between the common attribution models are set out in first-click vs last-click vs data-driven attribution, and they are large enough on their own to explain most of a reconciliation argument. Your finance system counts money that settled, net of refunds that arrived weeks later, which is the most conservative and the slowest of the four.

Put that way, the striking thing is not that they disagree. It is that anybody expected them to agree. Four instruments measuring different quantities under different rules produce four answers, and a reconciliation project premised on making them match is attempting something arithmetic rather than organisational.

Which is why the answer is not to pick the most accurate one. It is to decide which definition the business will use, and to be clear that the others remain correct for their own purposes.

A single source of truth is an authority decision

I have watched several versions of this project and the ones that failed all failed in the same way, which is that they were handed to the right people in the wrong order.

The technical version starts with a warehouse, a pipeline and a modelling layer. It is competently built, it produces a figure, and the figure is immediately contested by whichever department it disadvantages, because the definition it encodes was chosen by whoever wrote the transformation rather than by anybody with the authority to settle it.

The version that works starts with a short document. Five definitions, each with a named owner, each written in a sentence a non-specialist can read. Then the pipeline implements the document. The sequence matters because the document is where the argument happens, and having the argument deliberately, once, in a room, is much cheaper than having it implicitly every week in a growth meeting.

Naming the owner is the step teams find uncomfortable and it is the one that does the work. An owner is not an author. An owner is the person whose answer stands when two departments disagree about whether a particular case counts, and whose job it is to decide quickly rather than convene a working group.

One practical note on who. Acquisition cost should usually be owned by finance and conversion by whoever owns the store experience, which splits the authority rather than concentrating it. That is deliberate: a single owner for everything becomes a bottleneck and invites the suspicion that the definitions serve one department's narrative.

The five definitions worth standardising

The table sets out each one with the question it settles, who should usually own it, and the edge case that forces the definition to be explicit.

The edge-case column is the most useful one to work through, because an edge case is what reveals that two people have been using a word differently for a year. Acquisition cost is the standard example: performance marketing means media spend, finance means media spend plus agency fees plus creative production plus the tools, and the two figures can differ by a third. Nobody was wrong and nobody had written it down.

Contribution is the one most worth getting right, because it is the input to every decision about what a customer is worth. Measured across the relationship rather than on a single order, after the cost of goods, the discount, the acquisition and the servicing, that figure is what we call True Profit, and it is the only number that answers what you can afford to pay for a customer. Bain and Reichheld's long-cited finding that a five percent retention improvement can raise profits by twenty-five to ninety-five percent only becomes actionable once contribution is defined, because until then nobody can say which customers the retention is worth buying. The mechanics of that figure are covered in customer lifetime value, and the speed at which it is recovered in CAC payback period.

What to do with the platform numbers

There is a tempting overreach at this point, which is to declare the platform numbers wrong and ban them. That fails, for a practical reason.

The platform's figures are what the platform optimises against. A performance team steering a campaign needs the platform's view of what happened, because that is the view the bidding system is using. Telling them to steer by a warehouse number computed overnight makes their job harder and does not make the platform behave differently.

So the rule is about job rather than accuracy. Platform numbers steer platforms. The agreed definitions report the business, set budgets and settle arguments. Both are kept, and each is used for one thing.

The gap between the two is then worth watching as a figure in its own right. A platform claiming substantially more conversions than your own definition recognises is telling you something about attribution windows and identity matching, and the trend in that gap is more informative than either number alone. Shopify's analytics documentation is reasonably clear about what its own figures do and do not include, which makes it a good place to start when you are writing down the store-side definitions.

This is also where Nexus by Omniconvert earns its place in the argument. Nexus is the AI for eCommerce growth engine that unifies your commerce data, prioritises experiments by True Profit, and generates campaigns and creative you approve before they go live. What matters for this article is the first clause: the unification is the precondition for everything after it, and the approval step is the reason a unified view does not quietly become an autonomous one. A system that proposes and a person who decides is the arrangement that keeps the definitions meaningful.

What this asks of an operator

The work is smaller than it sounds and the resistance is larger, which is the usual shape of anything that reallocates authority.

• Write the five definitions on one page. A sentence each, in plain language, with the edge case named. If it runs to four pages, you are standardising too much.
• Name one owner per definition and say what owning it means: their answer stands, and they answer within a day rather than convening a group.
• Publish one place to read the figures. Not necessarily a new tool. The requirement is that it is the only place quoted in a growth meeting, which is a rule rather than a feature.
• Narrow the platform numbers' job in writing: they steer platforms, they do not report the business. Keep them, and stop arguing with them.
• Diarise a definitions review. Twice a year. Definitions drift as the business changes, and an unreviewed document gets quietly worked around rather than updated.
• Decide which metrics you will not standardise and say so out loud, so the omission reads as a choice rather than an oversight. The category-level list in CRO metrics is a useful place to mark what stays local to a team.

One closing observation about why this is worth the political cost. The benefit is not accuracy. The benefit is speed: a team that does not have to establish which number is real before a decision makes more decisions per quarter, and over a year that compounds in a way no individual metric improvement does. Baymard Institute's usability research is a good reminder of how much is available to a team that can actually act on what it finds, and acting is the part that definition disputes quietly prevent.

FAQ: a single source of growth truth

What is a single source of growth truth?

One agreed set of metric definitions, with one named owner per definition and one authoritative place to read the figures. It is not a new dashboard, and building a dashboard without agreeing the definitions underneath it produces a fifth disagreeing number rather than a resolution. The deliverable is a short document plus an owner, and the technology follows from it.

Why do our platform numbers never match our store?

Because they are answering different questions with different rules and both answers are defensible. An ad platform attributes a conversion to an impression it served within its own window, counts the person it recognised, and reports on its own clock. A store counts a completed order against the session it saw. Neither is wrong. They are measuring different things, and expecting them to agree is the error.

Which metrics should be standardised first?

Five: what counts as a customer, what counts as an order, what counts as a conversion, how acquisition cost is calculated, and what contribution includes. Those five appear in almost every growth decision and they are the ones defined inconsistently most often. Standardising fifty metrics is a project that never finishes and nobody reads the output of.

Is this a data engineering problem or a management problem?

Management, with an engineering consequence. The hard part is deciding whose definition wins when finance and performance marketing disagree about acquisition cost, and no pipeline resolves that. Once the decision is made and written down, implementing it is ordinary work. Teams that start with the pipeline build an accurate system that two departments still refuse to quote from.

What should you do with the platform numbers afterwards?

Keep them, and use them only to steer the platform. An ad platform's own figures are the right input for its own optimisation decisions and the wrong input for reporting the business. Writing that distinction down removes most of the remaining argument, because the disagreement stops being about who is right and becomes a question of which decision the number is for.