The Three Versions of Your Data (And Why They Never Match)

$25 Million That Was Never Real

Last year, I watched $25M go from "collectible" to "at-risk" in a single meeting. Not because anything changed in the real world. Because someone finally looked at the source data.

Our case tracking system showed 47 accounts as "pending payment." Finance reported it as collectible revenue. Operations believed the system. Everyone assumed someone else was validating the numbers.

Nobody was.

When we dug in, 35 of those accounts were uncollectible. Status fields that hadn't been updated in months. Cases marked "pending" that were actually closed. Reconciliation delays masquerading as open receivables.

The $25M didn't disappear. It was never real. The data just finally admitted it.

The Three Sources of Truth

Every company has this problem. The numbers don't match, and they never will, because you're dealing with three fundamentally different versions of reality.

Version 1: The System. This is what your database, ERP, or GL shows based on status fields, transaction records, and automated entries. It's precise. It's also often wrong, because it only knows what someone (or something) told it. A status field that hasn't been updated in 60 days is just a stale opinion stored in a database.

Version 2: Operations. This is what your team believes based on their conversations, emails, and gut feel. The sales rep knows that deal is dead even though the CRM says "negotiation." The collections manager knows that account is paying next week even though the aging report shows 90+ days. This version is often more accurate than the system but lives in people's heads, not in your reports.

Version 3: Finance. This is what you report to the board, investors, and leadership based on conservative assumptions and accounting standards. Finance doesn't care what Operations thinks will happen. Finance cares what the evidence supports. This version is the most defensible but also the furthest from operational reality.

You'd think these would converge. They don't.

Why They Diverge

The problem isn't bad people. It's misaligned incentives and timing differences.

Finance needs historical consistency. I can't restate last quarter every time someone updates a status field. The board expects numbers to make sense month-over-month. Auditors expect traceability. So Finance locks periods and moves forward, even when the underlying data was wrong.

Operations needs current reality. They can't work deals based on stale information. Their performance depends on knowing what's actually recoverable right now, not what the system said two months ago.

The system punishes accuracy. Update a case to "uncollectible" and Finance asks why the forecast changed. Leave it as "pending" and Operations makes decisions based on phantom revenue. Either way, someone's working with the wrong number.

This creates a predictable pattern: data quality degrades slowly over weeks and months, then someone finally reconciles and discovers a cliff. The $25M meeting. Every CFO has one eventually. The only variable is how big the gap is when you find it.

How to Close the Gap

You can't eliminate the three versions. That's a fantasy. But you can build systems that keep them close enough to make good decisions.

1. Never trust a status field older than 30 days.

If no one has touched a record in a month, it's probably wrong. Build automated alerts for stale data. This is one of the simplest things you can do, and it catches problems before they compound. Most CRM and ERP systems can flag records that haven't been updated within a threshold. If yours can't, a weekly report filtered by "last modified date" takes 10 minutes to build.

2. Trace every report to source data.

The dashboard looks clean. The underlying data is chaos. Before presenting any number to leadership, ask: where does this come from? Which system? Which table? When was it last updated? If you can't answer those questions, you're presenting a number you can't defend.

3. Build weekly reconciliation checkpoints.

Not monthly. Weekly. The problems compound fast. A weekly 30-minute sync between Finance and Operations on the top 20 accounts by dollar value catches 80% of divergence issues before they hit a board report.

This doesn't need to be a formal meeting. A shared spreadsheet where Operations updates the status and Finance confirms the reporting treatment works fine. The point is forcing the two versions to look at each other regularly.

4. Default to conservative for reporting.

When Finance and Operations disagree on a number, Finance wins for external reporting. Always. Operations can work their version internally for pipeline management and forecasting. But the board, investors, and auditors see the conservative number.

This protects the company and, frankly, protects you. Being pleasantly surprised by upside is always better than explaining why last quarter's number was wrong.

5. Use AI to automate the reconciliation layer.

This is where the game has changed in the last two years. AI-powered reconciliation can continuously compare your system data against bank feeds, operational reports, and historical patterns. It flags discrepancies in near real-time instead of waiting for month-end.

I've seen AI catch issues that would have taken a team days to find manually: duplicate entries, timing mismatches, status fields that contradict transaction data. The technology isn't perfect, but it's dramatically faster than a human combing through spreadsheets.

The setup takes time. You need clean data feeds and well-defined matching rules. But once it's running, you're catching the $25M problem at $500K instead of letting it grow for six months.

The Bigger Lesson

Data governance isn't glamorous work. Nobody gets promoted for clean status fields. But here's what I know after cleaning up several messes like the one I described: the cost of bad data isn't the bad data itself. It's the decisions you make based on numbers that were never real.

Hiring plans built on phantom revenue. Cash projections that assume receivables are collectible when they're not. Board presentations that paint a picture the underlying data doesn't support.

The three versions of your data will always exist. Your job isn't to make them identical. It's to make them close enough that the gap doesn't cost you money, credibility, or sleep.

Mine are closer now. Not perfect. Closer.

That's the job.