Two Reports, Two Different Numbers
Ask three people in a lower middle market company what last month's revenue was and you can get three answers. Sales quotes one figure from the CRM, finance quotes another from the accounting system, and the number the owner repeats to the bank is a third one, assembled in a spreadsheet nobody else has seen. None of them are lying. They are reading different systems that were never taught to agree.
Most owners learn to live with this. They pick the number they trust, mention the gap in passing, and move on. The habit is more expensive than it looks, because every decision downstream inherits the disagreement. You can't price confidently, forecast honestly, or explain your own business to a lender when the foundation shifts depending on who you ask.
Where the second number comes from
The gap almost always traces back to hand work. Despite a decade of software built to replace it, the spreadsheet is still where the real numbers get made. A 2025 survey by the Association for Financial Professionals found that 96 percent of finance professionals still use Excel for planning, and separate research puts finance teams at roughly 40 percent of their time spent on manual tasks that software could do. That is not a knock on the people. It describes what happens when your systems don't talk, so a human becomes the connection between them, copying figures from one screen into another.
Every copy is a chance to be wrong. Manual data entry carries an error rate estimated between 1 and 4 percent per field, which sounds trivial until you remember how many fields sit under a monthly close. One transposed figure, one row that did not get refreshed, and two reports that should match no longer do. At scale the damage compounds. Gartner has estimated that poor data quality costs the average large organization around $12.9 million a year. Your number is smaller, but the mechanism is identical, and at your size a bad number is more likely to be the one you actually acted on.
The temptation to buy a cure that's worse
I have sat across from owners who reached the same reasonable conclusion: the systems don't agree, so we'll buy one big system that holds everything. It is a natural instinct and often a costly mistake.
Large integrated platforms fail to deliver on their promise more often than they succeed. Independent research on ERP projects has found that somewhere between 55 and 75 percent miss their original objectives, that most run over budget, and that a majority take longer than planned. The common causes are not exotic. They are bad data moved into the new system, weak ownership, and scope that swelled past what anyone could manage. You don't fix a data problem by pouring the same messy data into a more expensive container.
The goal isn't one system. The goal is one version of the numbers. Those are different projects, and the second one is far cheaper.
One number, earned cheaply
Start by naming the source of truth for each figure that matters. Revenue comes from here, cash from there, headcount from that. Write it down, one line each, and make it a rule rather than a habit. Half the disagreement in most companies is simply that nobody ever declared which system wins.
Then close the seams by hand where the volume is low and by a simple connection where it is high. Plenty of the tools you already pay for can pass data to each other directly, which retires the retyping that introduces most of the errors. You are not trying to automate everything. You are removing the specific manual steps where a number gets reborn slightly different each month.
Last, agree on definitions. A surprising share of the two-report problem is not bad data at all. It is that sales counts a booking the day it signs and finance counts it the day it ships, so both numbers are correct and neither matches. Decide which one the company means when it says revenue, and put it in writing.
None of this is a transformation. It is a few decisions about where truth lives and a little plumbing to keep it there. Do it, and the next time someone asks what you made last month, there is one answer, and you can say it without checking three places first.