What Is a
Revenue Leak?
Money already coming in that the marketing stack loses or misreports. There are four kinds. Each one has a symptom you notice, a unit it is measured in, a first diagnostic, and a fix with a number on both ends.
By Daria Dovzhikova · Updated September 2026
TL;DR
- A leak has a number attached. A preference does not. That is the whole test.
- The four kinds: attribution and measurement, AI search visibility, demand systems, launch infrastructure. Each lives in a different tool, which is why nobody owns it.
- You measure a leak in its own unit, before and after: attributed revenue, unbranded citation share, qualified conversations, the day-30 read.
- A four-week pilot finds it in week one and runs two fixes as experiments, decided every Friday.
What a revenue leak is
A revenue leak is money that was already on its way in, and that the marketing stack lost or misreported on the way. The customer existed. The demand existed. The spend happened. Somewhere between the ad click and the order, or between the buyer's question and your name, the stack dropped it.
The word gets used loosely, so here is the test. A leak has a number attached: purchases the platform did not see, answers that named a rival, conversations that never got a reply, a launch with no day-30 read. If you cannot put a number on it, you are looking at a preference, and preferences are a different conversation.
Four kinds show up in almost every marketing stack with revenue behind it. They map one to one onto the systems a marketing engineer owns, and onto the experiment cards a Growth Pilot runs.
The four kinds
| Leak | Symptom you notice | Unit | First diagnostic | Fix |
|---|---|---|---|---|
| Attribution and measurement | The ad platform reports a fraction of the purchases the order system shows, or more than it. Finance stops trusting the dashboard. | Match rate and attributed revenue | Reported purchases against the order of record for the same window, per platform. | Server-side events, dedup, one reporting view Runs as an experiment card in a four-week pilot |
| AI search visibility | A buyer asks ChatGPT, Claude, Gemini or Perplexity for the best tool in your category and gets a competitor, sometimes with a citation to your own page. | Unbranded citation share, per engine | Ten unbranded buyer questions across four engines. Count who gets named. | Extraction fixes and comparison pages, re-measured Runs as an experiment card in a four-week pilot |
| Demand systems | Pipeline depends on one person's inbox. When they are on holiday, outbound stops. Nobody can say how many qualified conversations last month produced. | Qualified conversations per month | Count the conversations that started from a system versus from a person, for the last ninety days. | A signal-scored pipeline with human review Runs as an experiment card in a four-week pilot |
| Launch infrastructure | A release ships with a changelog and a post. No messaging hierarchy, no comparison page, no enablement, and no read at day thirty to say what it did. | The day-30 read against the baseline | For the last launch: who owned it, what page it produced, what number it was meant to move. | Messaging hierarchy, comparison page, day-30 read Runs as an experiment card in a four-week pilot |
Where each leak is covered on this site
Attribution and measurement
AI search visibility
Demand systems
Launch infrastructure
One example of each
Two of these are named engagements. One is anonymized at the client's request. The fourth is the shape, because it is the same at every company.
Attribution
A DTC hardware brand running a marketing site plus four regional Shopify Plus stores. Meta Ads Manager reported 7 purchases against 151 in the order system, and a cost per purchase of $2,440 against a real range of $700 to $1,200. Events Manager showed why: the Conversions API was not firing, the event ids did not match across the marketing site and the stores, and 28 unverified domains were polluting the pixel. The ad platform had been optimizing against a signal that saw one purchase in twenty.
The rebuild guideAI search visibility
Radar by Skyhook, an open-source Kubernetes UI. On unbranded prompts across four answer engines, the July baseline named Radar in 3 of 48 answers. The page the engines cited most when recommending a competitor was Radar's own comparison page. After that page was rewritten and the canonical positioning shipped, the September re-run named Radar in 8 of 46.
The Radar caseDemand
The same client's community talks about it in eight places at once. A daily listening system swept those channels, triaged 1,635 conversations over the engagement, filed 174 for a human reply with an owner and a priority, and posted nothing by machine. Before it, mentions surfaced days late or never; after it, every conversation worth answering reached a person once, with a draft attached.
The listening systemLaunch
A release with a date and no owner. The changelog goes out, the founder posts, and the comparison page that would have caught the buyers weighing the incumbent never gets written. Thirty days later nobody can say what the launch moved, so the next one starts from zero again. The fix is an experiment card with a baseline at the start and the same read on day 28.
The launch checklistHow to find yours
Two of the four you can check yourself in an afternoon. Put platform-reported purchases and actual purchases for the same month into the Attribution Leak Calculator and read the gap. Run the free AI visibility check on your category and count how often you are named on unbranded questions, with your own name kept off the prompt.
Week one of a Growth Pilot does all four lanes in one baseline: what your attribution layer sees versus what it reports, where four answer engines name you against competitors, where your demand depends on a person instead of a system, and what the last launch moved. The two leaks with the largest number become the two experiment cards.
Why leaks get ignored
Each leak lives in a different tool, and each tool has a different owner. The pixel belongs to whoever set it up two agencies ago. The comparison pages belong to content. Outbound belongs to sales. The launch belongs to product. Nobody looks at all four, so the number that would expose the leak never lands on one desk.
The role that owns the systems end to end is the marketing engineer. Most companies do not have one yet. Until they do, the leak sits between four job descriptions and shows up as a finance question nobody can answer.
FAQ
What is a revenue leak?
Money already coming in that the marketing stack loses or misreports. An ad platform that sees a fraction of your conversions and optimizes against the wrong signal. An answer engine that recommends a competitor when a buyer asks an unbranded question. A pipeline that stops when one person is away. A launch nobody measured. Each has a number attached, which is how you tell a leak from a preference.
How is this different from churn, or from revenue leakage in billing?
Churn is customers leaving after they bought. Billing leakage is invoices that were never raised or collected. Both sit downstream of marketing. A revenue leak in the sense used here sits upstream: demand that existed and was lost, or spend that was steered by a wrong number, before the customer ever reached a contract.
How do you measure a revenue leak?
In the unit the leak is denominated in, before anything changes and again after. Attribution: match rate and attributed revenue. Visibility: unbranded citation share across four engines, with branded prompts reported on their own line. Demand: qualified conversations per month. Launches: the day-30 read against the baseline. If the number did not move, the readout says so.
What does it cost to find the leak?
Two of the four leaks you can measure yourself with the free tools on this page. The paid route is the Growth Pilot: $5,000 fixed for four weeks, where week one is a baseline of the current funnel and two fixes run as experiments from week one to week four, with a decision every Friday. If the baseline finds nothing leaking, the first Friday review says so.
Which leak is the most common?
Attribution, by a distance. Most companies with real ad spend have a gap between what the platform reports and what the order system shows, and most of them found out from a finance question rather than a dashboard. AI search visibility is the fastest-growing one, because the first stop in a buyer's research moved to an answer engine and almost nobody measures it on unbranded prompts.
Can AI fix it cheaper?
AI already runs most of the delivery, which is how a pilot runs two experiments in four weeks. It does not diagnose the leak, decide which number matters, or take the readout to finance. You pay for the diagnosis and the fix. If your first question is whether it can be cheaper, the free tools are the honest answer.
Who owns a revenue leak inside the company?
Usually nobody, which is why it persists. Attribution lives with whoever set up the pixel, visibility with whoever writes the pages, demand with sales, launches with product. Each one is a system, and the role that owns systems end to end is a marketing engineer. Most companies do not have one, so the leak sits between four job descriptions.
How long does a fix take?
Inside a pilot, four weeks: the baseline and two experiment cards in week one, both cards running from week one to week four, a readout on day 28. A fix that needs longer keeps running as Continuation, month to month, with one new card a month.
The Growth Pilot
Four weeks, two real growth experiments, a decision every Friday.
Week one of a pilot is the baseline across all four leaks; the two with the biggest number become the experiment cards. For developer-tool and AI-product founders who don't have a growth lead yet. $5,000 fixed, half at kickoff. How a pilot runs · What an experiment card looks like
Book a 30-minute callRelated guides
Not sure which leak it is?
Week one of a pilot reads all four lanes. If nothing is leaking, the first Friday review says so.