Measure whether Connect is working
Watch five things weekly: how old the oldest item in the queue is, how often a draft is sent unedited, whether overdue commitments are shrinking, the call review score with its model-limited count beside it, and allowance consumption against the plan. All five come from what the product already records — none requires a spreadsheet.
The five numbers#
| Number | Where it comes from | What moves it |
|---|---|---|
| Age of the oldest item in Needs You | The queue itself, which is ranked rather than chronological | Approval capacity, and autonomy set tighter than the team can service |
| Share of drafts sent without editing | The Decision Log, which records approvals and edits | Knowledge quality, standing instructions, and memory at the contact tier |
| Overdue follow-ups, trend not level | The follow-up summary | Whether commitments are being created with reasons and carried on channels that work |
| Call review score, with the model-limited count | The review written at the end of each call | Settings for the score; nothing for the model-limited part |
| Allowance consumed against the plan | Plan and Usage | Volume, reply length, and how much research each prospect gets |
Depth is the wrong reading on the first of those. A queue with forty items that all arrived this morning is healthier than one with six from last Tuesday, because entries drain by themselves as their causes clear. Age is what tells you something is genuinely stuck.
Reading each one properly#
- Queue age
- Oldest first, not average. An average hides the one item nobody will touch, which is usually the one that matters.
- Unedited send rate
- Rising means Connect is writing what you would have written. Falling means something changed — new products, a new market, or a standing instruction that has gone stale.
- Overdue trend
- A flat overdue count with different items underneath is the same debt with new names. The trend is the number; the level is decoration.
- Review score
- Two figures are published: one that excludes findings no setting can change, and the older all-inclusive one. A fall in the second and not the first is the model, not you.
- Allowance
- Consumption per working day, not per month. A month is too coarse to catch the day a discovery run changed the shape of your spend.
Numbers not worth watching#
- Total messages handled. It measures how busy your customers were.
- Queue depth on its own, for the reason above.
- Transcript-derived reply latency. Timing is measured on the wire; a transcript row arrives when a sentence completes and once showed millisecond figures beside waits of several seconds.
- Prospect counts taken from a fixed slice of the newest rows. Headline figures are computed over the whole set now, but a suspiciously round number is still worth checking against the counting page.
- Anything you have to assemble by hand every week. A measure that depends on somebody remembering to build it stops existing within a month.
Turning a number into a change#
Pick the one number that moved most, and look at its cause rather than the number.
Result Queue age points at approval capacity or autonomy; unedited send rate points at Knowledge and memory; overdue points at how commitments are being created.
Change one thing.
Result Two changes in one week produce a result neither of them explains.
Check the same five numbers next week, before deciding whether it worked.
Result Most of these move on a weekly rhythm; judging them daily produces noise and a lot of unnecessary tinkering.
Write down what you changed, as a standing instruction or a memory entry where it belongs.
Result Otherwise the improvement is not repeatable and nobody else can tell what produced it.
Questions#
Is there one number that says whether it is working?
No, and a single figure would hide the trade-offs. Queue age and unedited send rate move in opposite directions when autonomy is tightened, so either one alone can be improved by making the other worse.
How long before these numbers mean anything?
Enough volume to see a trend rather than a week's weather. Until then, read them as a description of what happened rather than as a measurement of whether it is improving.
Can I export these?
The underlying records are available through the Data grid and its CSV export, which neutralises formula characters on the way out. That is the route if somebody outside Connect needs the figures.