DOI : 10.17577/Two days before a board meeting, the CEO of a mid-sized software company received two revenue figures for the same quarter. Finance reported $4.81 million. The data team’s dashboard said $4.63 million. Both teams were confident. It took most of a day to find the cause: the dashboard’s pipeline pulled invoices from the billing system but ignored credit notes issued after disputes, and the two teams used different time zones to decide when the quarter ended.
Neither number was wrong by its own rules. That was the problem.
Why Companies End Up With Two Truths
Billing systems and analytics pipelines usually grow up separately. Finance picks a billing tool to get invoices out the door. Engineers or analysts later build pipelines to feed dashboards. Each side makes reasonable choices, and the definitions quietly drift apart.
The research on this is not encouraging. Industry estimates commonly put revenue leakage from billing errors, missed charges, and unapplied price increases at somewhere between 1 and 5 percent of revenue for subscription businesses. One widely cited analyst estimate puts the average annual cost of poor data quality at around $12.9 million per organization. Surveys of data practitioners consistently report that a large share of their week goes to finding, cleaning, and reconciling data rather than analyzing it.
For a company with $20 million in annual revenue, even the low end of that leakage range is $200,000 a year.
Billing Should Work Like a Ledger
The most important architectural decision in software for billing systems is treating every financial record as permanent. An invoice, once issued, never gets edited. If something was wrong, the system issues a credit note or an adjustment that references the original. Every change becomes a new record with a timestamp and a reason.
Systems built this way can answer questions that edited records can’t. What did this customer owe on March 31, as of March 31? Why did this account’s balance change last Tuesday? Who approved this discount? Auditors ask questions like these, and so do customers disputing a charge.
The underlying events follow the same logic: subscription created, plan upgraded, usage recorded, invoice issued, payment failed, payment retried. Store them in order and never overwrite them.
Should a growing SaaS company build this itself? For standard subscriptions, no. Hosted billing platforms handle invoicing, proration, tax, and payment retries better than most internal teams will. Build only what’s unique to your product, typically the metering that counts usage and the rules that turn usage into charges.
Pipelines Need the Same Discipline
The dashboard error in the opening came from a pipeline that captured some financial events and not others. Pipelines reading from billing data need to pull every event type, including credits, refunds, chargebacks, and write-offs.
Define revenue once. A single shared definition, with its time zone, recognition rules, and treatment of credits written down and implemented in one place, prevents the slow drift that produced two different board numbers.
Then test the pipeline against the billing system daily. If the pipeline’s totals differ from the billing ledger by more than a dollar, an alert should go to both finance and data teams before anyone builds a chart on it.
Prophecy is one way teams approach this. Pipelines are built visually, but the tool generates standard code stored in version control, so a finance analyst can follow the revenue logic step by step while engineers review every change before it ships. It doesn’t decide what revenue means for you, and it doesn’t replace reconciliation tests. What it does is make the logic visible to the people most likely to spot a missing credit note.
Where Scale Breaks Things
Late-arriving data is the quiet failure. Usage reported three days late, a payment that settles after month-end, a dispute resolved weeks after the original charge. Architectures that assume data arrives on time produce numbers that change after they’ve been reported. Design for corrections from the start: load data incrementally, make every load safe to rerun without duplicating records, and mark reports as preliminary until the close is final.
The company eventually traced the $180,000 gap to roughly 300 credit notes and a five-hour time zone difference. Fixing the pipeline took a week. The harder conversation was agreeing, for the first time, on a single written definition of revenue that both teams would use.
