How confident can I be in each metric?
Your dashboard says gross margin is 62%. Your ROAS is 4.2x. But how much of that number is based on solid data and how much is based on guesses, gaps, and incomplete matching? Here's how to think about metric confidence.
The short answer
How reliable are your numbers? A metric is only as trustworthy as the data behind it. If the metric comes from a single system (like your P&L from QuickBooks), it is probably reliable. If it requires matching data across systems (like customer LTV or blended ROAS), its confidence depends on how well those systems are connected.
Making decisions on low-confidence metrics is worse than guessing
You see that your Google Ads ROAS is 4.2x. You decide to double your ad budget. But that ROAS was calculated by matching ad clicks to Stripe payments, and only 34% of those matches were exact. The other 66% were fuzzy matches or completely unattributed. Your real ROAS might be 2.1x or 6.0x. You have no idea.
At least when you guess, you know you are guessing. When a dashboard shows you a precise number with two decimal places, you assume it is accurate. That false precision is what leads to bad decisions.
The fix is not to stop measuring. It is to know the confidence level of each metric so you treat high-confidence numbers differently from low-confidence ones.
Three factors that determine how much you can trust a metric
- Data completeness. Does the metric use data from all relevant records, or just the ones that could be matched? If your customer LTV calculation only includes the 60% of customers who matched across CRM and payments, the other 40% could skew the real number in either direction.
- Match quality. Even among matched records, are the matches reliable? A metric based on exact email matches is more trustworthy than one based on fuzzy name matches. The percentage of exact vs. weak matches in the underlying data directly affects your confidence.
- Data freshness.Is the data current? A metric calculated from last week's sync is more reliable than one from a sync that ran three months ago. Stale data means the metric reflects a past reality, not the current one.
How to estimate metric confidence manually
There is no built-in confidence score in QuickBooks, HubSpot, or Stripe. To estimate confidence yourself, you need to work backwards from each metric to its data sources.
For every cross-system metric you rely on, ask three questions: What percentage of the relevant records were successfully matched? What quality were those matches (exact vs. fuzzy)? When was the last time the data was synced?
For single-system metrics (like gross margin from your P&L), the confidence is generally high as long as your books are up to date. For cross-system metrics (like customer LTV, blended ROAS, or marketing attribution), the confidence depends entirely on your match quality between the systems involved.
This is hard to do manually because most tools do not expose the connection between a metric and its underlying data quality. You would need to trace each number back to its source tables and evaluate the match quality yourself.
Confidence changes as your data quality changes
A metric that was high-confidence last month could become low-confidence this month if new records were added without proper matching, or if an integration broke and data stopped syncing. Without continuous monitoring, you never know when a metric you trust has quietly become unreliable.
Or see confidence scores on every metric automatically
Bottomline attaches a confidence indicator to every metric in your report. Metrics based on complete, well-matched data show high confidence. Metrics with gaps or weak matches show the confidence level alongside the number, so you always know how much to trust what you are reading.
Your gross margin is high-confidence because it comes from a single accounting system. Your blended ROAS is low-confidence because the match rate between ads and payments is weak. You can trust the first number to make decisions. The second one needs better data before you bet money on it.