Introduction
The ROI view is the page someone senior will read, usually without you in the room. This page is about reading it accurately — and about the two things on it that are easy to quote wrongly.
The tiles
Tile | Is |
| Total value ÷ total spend, as a multiple |
| Counted sessions × what each is worth, across all outcomes |
| Spend of agents credited to an outcome — not all spend |
| Total value − total spend |
| Spend not attributed to any outcome |
All five follow the window picker (default: last 30 days). Two figures from two windows are not comparable.
The table
One row per outcome: Outcome (with its count), Owner, Sources, Total outcomes, Total value, Total spend and Net value. The row menu has Edit and Delete.
An outcome with dashes across the numeric columns counted nothing in this window. That is either a genuine zero or a rule that has stopped matching — and the two look identical here, which is why source maintenance matters.
The two things people quote wrongly
1 · Return on spend divides by measured spend
The denominator is the spend of agents credited to an outcome — not your AI bill. The Unmeasured spend tile shows the rest, and it is the most important figure on the page.
So a high multiple can coexist with an estate where most spend produces no measured outcome at all. Both facts are true; quoting only the first is how an ROI claim gets dismantled.
What to do: compare Total spend with Unmeasured spend (or with Cost Explorer over the same window), and report both. "3.2× on the 40% of spend we measure" is a statement that survives questioning. "3.2× return" is not.
"Volunteering the coverage gap is what makes the multiple believable. Somebody will find it either way — better that it comes from you."
2 · The verdict is arithmetic, not judgement
Net positive, Break even and Net negative compare value against spend. They say nothing about whether the value figure was well chosen.
A Net positive verdict on an outcome whose value nobody agreed is a confident-looking number resting on an assumption. The verdict inherits the quality of the input.
Value by score
On an outcome's page, the Value by score table breaks it down by the agents crediting it, and it is the most operationally useful view of an outcome.
One source dominating — Usually correct, and worth knowing — that agent is the one to protect.
A source contributing nothing — Either it does not do this work, or its rule is broken. Check the rule before removing it.
A source that used to contribute and stopped — Almost always a rule reading a signal that changed.
Reporting this upward
Name the window — Every figure is scoped to it.
Report Total spend alongside Unmeasured spend — Always.
Say where the value figure came from — Name who agreed it.
Include the outcomes that counted nothing — Omitting them makes the set look better than it is.