Work Understanding: AI adoption and impact
Objective. Open the AI adoption and impact view, separate observed AI-tool usage from the estimated impact of that usage, and avoid quoting modelled impact as proven ROI.
This screen shows where AI tools are being used in the work and estimates their effect on time or output. It answers 'is AI actually being adopted, and does it appear to help?' - with usage observed but impact modelled.
See it in the demo
What this screen answers
The customer question is: are the AI tools we rolled out actually being used in the real work, and does that use appear to change how long the work takes or how much gets done? Sales and CSM people open it when a prospect is spending on AI copilots and wants evidence of adoption and effect rather than a vendor promise.
There are two distinct things on this screen and keeping them apart is the whole job: adoption (observed - where and how much AI is used) and impact (estimated - the modelled effect of that use). Lead with adoption as the firmer signal.
The task
- Open the AI adoption and impact screen from the overview.
- Scope to one process and period.
- Read adoption first: where AI use is observed, by how much of the work, trending which way.
- Then read impact as a separate, softer claim: the estimated time or output difference, under assumptions.
- Narrate the distinction aloud: 'adoption is observed; impact is an estimate of effect, not a proven causal saving.'
Do not invent adoption percentages, impact figures or tool names. Describe at the level of observed usage versus modelled effect.
How to read it
Adoption is relatively firm because it is observed usage. Impact is an estimate and carries the classic misread: 'AI saved us N hours' stated as fact. The screen shows an association between AI use and a time or output difference - association is not proven causality. Faster work where AI was used may reflect easier cases routed to it, more skilled users, or a parallel process change.
Also watch the denominator: adoption as a share of eligible work is different from adoption across all work. And low measured impact can mean the tool is used on the wrong tasks, not that AI has no value - a reason to investigate, not to conclude.
Honest limits
This screen does not prove that AI caused a saving or a quality change. It cannot isolate AI from case mix, user skill, or concurrent changes without a controlled before/after. Impact figures are modelled under assumptions that must be surfaced.
Confirm with product how AI use is detected and how impact is estimated for a customer. Any ROI claim belongs to a scoped experiment against the customer's own baseline, not to the modelled number on this screen. Present adoption as evidence and impact as a hypothesis worth validating.
Interpretation check
Adoption is observed usage; AI impact is an estimate of effect, not proven ROI. An association between AI use and faster work is not causality - it needs a controlled before/after to claim a saving.
In the field
For an RCM team piloting AI claim summarisation, the defensible story is 'adoption reached X of eligible cases' plus a scoped before/after on turnaround - not a headline 'AI cut handling time' lifted from the modelled impact figure.
Practical task
On the demo AI screen, for one process record two separate lines: the observed adoption signal, and the estimated impact with the assumptions you think it rests on. Then write the one experiment that would turn the impact estimate into a claim you could defend.
Self-check
The screen shows work with AI was faster. Can you tell the customer AI saved them that time?
No. That is an association, not proven causality. Easier cases, more skilled users or a parallel change could explain it. A saving claim needs a controlled before/after against their baseline.
Which number on this screen is firmer, and why?
Adoption, because it is observed usage. Impact is modelled under assumptions, so it is the softer claim and must be presented as a hypothesis, not a result.
Impact looks low for a tool. What is the honest reading?
It is a signal to investigate - the tool may be applied to the wrong tasks or measured on the wrong denominator - not a conclusion that AI has no value.
What converts an impact estimate into a defensible ROI statement?
A scoped before/after experiment that isolates the AI change against the customer's own baseline, with assumptions stated - not the modelled figure alone.
Verified vs demo. Screen is live in the demo at /#/ai; exact controls, detection and impact-modelling logic and figures are SME-validation-pending, and how AI use and impact are measured for a customer must be confirmed with product.