Most reporting on automated revenue systems measures the machine rather than the business. Here is the short version of what to keep.
The four numbers
1. Minutes to first contact. From a lead arriving to a human or an agent reaching it. It moves quickly, it is fully within your control, and it has a consistent relationship with conversion. If you report one number, report this one.
2. Percentage of leads that reach a qualified human conversation. Coverage and quality in a single figure. It catches two opposite failures: a system that reaches nobody, and one that reaches everybody with garbage.
3. Revenue traceable to its original source. Not modelled. Traceable, record by record, from money received back through the deal, the handoff, the conversation and the list. Report the percentage of closed revenue you can walk backwards. Most companies start in the low double digits and are surprised.
4. Hours of manual work removed. The number your team feels, and the one that keeps the system funded when budgets are reviewed. Count the steps that no longer happen, times their frequency.
What to ignore in leadership reporting
Executions, runs, workflow success rate. These describe the machine. Useful for the engineer, misleading for the business.
Emails sent, calls dialled, records processed. They rise whether or not anything improves. Twenty thousand calls is not a result.
Model level metrics. Token cost, latency, accuracy on a benchmark. Watch them for cost control. They do not belong in a revenue review.
The one check that catches silent failure
Every week, open ten records the system touched. Not a dashboard. The records.
Ask: is the business result there? Did the outcome get written? Does the next step make sense? Would a rep know what to do with this?
This catches the failures dashboards hide, because a green execution log means the automation ran, not that anything happened. Non deterministic systems in particular cannot be reasoned about theoretically, they have to be measured by running real data through them.
Setting the baseline before you start
The most common reporting mistake is having no before. Capture these the week before anything ships:
- Current minutes to first contact, measured not estimated
- Current percentage of leads contacted at all
- Current percentage of revenue you can trace
- The manual steps, counted, with their frequency
Without the baseline, every later number is an assertion.
How to review it
Monthly, with three questions: what moved, what did not, and what are we stopping. The third one is the point. A system that only ever adds gets expensive and unaccountable, and the discipline of naming one thing to stop each month is what keeps it honest.
If you want the attribution built properly
MitHub publishes open profiles of engineers who have built this chain inside real companies, with the deployments written out.