Which actions are verified from data vs. estimated by AI?

Why the distinction between verified and estimated matters

You get a report that says “Collect $14,200 in overdue invoices” and “Re-engage 4 dormant customers for an estimated $3,100.” Both sound like good actions. But they are fundamentally different.

The $14,200 is real. Those invoices exist. The clients owe you that money. The only question is whether you can collect it. The $3,100 is an estimate based on historical patterns: those customers used to order regularly, they stopped, and similar re-engagement efforts in the past recovered that average amount.

If you treat both numbers with the same confidence, you will either over-invest in uncertain actions or under-invest in certain ones. The labels help you allocate your time wisely: high-confidence actions first, then estimated ones.

How to think about data confidence in action items

Every action recommendation falls somewhere on a spectrum from fully verified to fully estimated:

• Verified from data. The number comes directly from a connected system. Overdue invoice amounts from your accounting software. Ad spend figures from your ad platform. Deal values from your CRM. These are facts, not projections.
• Estimated from patterns. The number is calculated from your historical data plus assumptions. Customer re-engagement estimates, projected savings from renegotiating a contract, expected close rates on stale deals. The underlying data is real, but the outcome involves prediction.
• Why both are valuable. Verified actions give you high-confidence quick wins. Estimated actions help you see opportunities you might miss otherwise. The key is knowing which is which so you can calibrate your expectations.

How to assess data confidence manually (step by step)

If you are building your own action list, here is how to label each item by confidence level:

1. 1
   Start with your accounting reports

   Any action item that comes directly from a report in QuickBooks or Xero (AR aging, expense line items, revenue figures) is verified. The number exists in your books right now.

2. 2
   Check your CRM and ad platform for direct data

   Deal values in your CRM are verified (the deal exists, the value is set). But a projected close rate or re-engagement recovery rate is estimated. Same with ad spend: current spend is verified, projected savings from changes are estimated.

3. 3
   Label each action on your list

   Next to each action, write “Verified” or “Estimated.” If you are unsure, it is estimated. Be honest about which numbers are facts and which are projections.

4. 4
   Prioritize verified actions first

   When two actions have similar dollar values, prioritize the verified one. A $5,000 overdue invoice you can collect is more reliable than a $5,000 deal you might close.

Why most action lists skip confidence labels

When you are building your own action list, it is tempting to treat every number as equally reliable. It takes extra effort to label each item, and it can feel deflating to admit that half your projected recovery is estimated rather than certain.

But over time, tracking which estimates prove accurate and which miss teaches you to calibrate better. If your AI-estimated recovery rates consistently come in at 60% of the projection, you can adjust future estimates downward. That calibration loop only works if you label things honestly from the start.

Or see confidence labels on every action automatically

Bottomline labels every action in your monthly report as “Verified” or “Estimated” based on where the data came from:

No guesswork about which numbers are facts and which are projections. Bottomline traces every recommendation back to its data source and tells you upfront. Over months, it also tracks how accurate its estimates are, so the AI-estimated numbers get more reliable over time as the system learns your business patterns.