Personal auto claims intake and settlement under CAD 10,000 — Northwind Insurance (illustrative)
709.9 hrs/week → 442.4 hrs/week -38%
18.9 → 11.8 full-time equivalent, across 12 assessed steps and 4 roles.
Those figures are the middle of a range. If every estimate lands at the same end, the answer is between 398.6 and 549 hrs/week.
136.3 hrs/week of the work after this change is work nobody does today — reviewing what the technology produces, and handling what it escalates.
Other options examined: Keep the process fully human and add two adjusters; Automate document extraction only, leave settlement untouched
Each block is one step, sized by the hours it takes. Both sides are drawn to the same scale, so the change in shape is the change in work.
12.2 → 7.2 full-time equivalent (-185.2 hrs/week)
| Kind of work | Hrs today | % | Hrs after | % |
|---|---|---|---|---|
| Data handling | 140 | 31% | 10.5 | 4% |
| Routine judgment | 163 | 36% | 51.7 | 19% |
| Complex judgment | 82.7 | 18% | 62 | 23% |
| Relationship | 70 | 15% | 70 | 26% |
| Supervision | 0 | 0% | 76.3 | 28% |
What shrinks: Extract facts from submitted documents (-129.5 hrs), Check coverage against the policy (-70 hrs), Assess the repair estimate for reasonableness (-41.3 hrs), Decide the settlement amount (-20.7 hrs)
What is new: Review AI settlement recommendation (+41.3 hrs), Review AI extraction output (+35 hrs)
3.5 → 0.2 full-time equivalent (-126 hrs/week)
| Kind of work | Hrs today | % | Hrs after | % |
|---|---|---|---|---|
| Data handling | 105 | 79% | 0 | 0% |
| Coordination | 28 | 21% | 7 | 100% |
What shrinks: Update the claim record and close (-58.3 hrs), Register the claim in the system (-46.7 hrs), Chase missing documents from the claimant (-21 hrs)
2.4 → 3.3 full-time equivalent (+32 hrs/week)
| Kind of work | Hrs today | % | Hrs after | % |
|---|---|---|---|---|
| Complex judgment | 90.4 | 100% | 122.4 | 100% |
What shrinks: Screen for fraud indicators (-28 hrs)
What is new: Handle additional AI-escalated exceptions (+60 hrs)
0.8 → 1.1 full-time equivalent (+11.7 hrs/week)
| Kind of work | Hrs today | % | Hrs after | % |
|---|---|---|---|---|
| Coordination | 7.5 | 24% | 7.5 | 18% |
| Supervision | 23.3 | 76% | 35 | 82% |
What grows: Review a quality sample of closed claims (+11.7 hrs)
| Hrs | Step | Who does it after |
|---|---|---|
| -129.5 | Extract facts from submitted documents | Claims Specialist |
| -70 | Check coverage against the policy | Claims Specialist |
| +60 | Handle additional AI-escalated exceptions new work | Senior Adjuster |
| -58.3 | Update the claim record and close | Intake Clerk |
| -46.7 | Register the claim in the system | Intake Clerk |
| -41.3 | Assess the repair estimate for reasonableness | Claims Specialist |
| +41.3 | Review AI settlement recommendation new work | Claims Specialist |
| +35 | Review AI extraction output new work | Claims Specialist |
| -28 | Screen for fraud indicators | Senior Adjuster |
| -21 | Chase missing documents from the claimant | Intake Clerk |
| -20.7 | Decide the settlement amount | Claims Specialist |
| +11.7 | Review a quality sample of closed claims | Team Lead |
| +136.3 | ai output evaluation | Claims Specialist, Senior Adjuster |
| +80.7 | exception reasoning | Claims Specialist, Senior Adjuster |
| +11.7 | quality assessment | Team Lead |
| -199.5 | policy knowledge | Claims Specialist |
| -111.3 | table lookup | Claims Specialist |
| -105 | data entry | |
| -105 | system navigation | |
| -94.5 | document reading | Claims Specialist |
| -41.3 | repair cost knowledge | Claims Specialist |
| -28 | fraud pattern awareness | Senior Adjuster |
| -21 | claimant communication | Claims Specialist, Intake Clerk |
| -21 | follow-up discipline | Intake Clerk |
| -9 | contextual policy interpretation | Claims Specialist, Senior Adjuster, Team Lead |
398.6 to 549 hrs/week (-44% to -23%)
That is every estimate landing at the same end at once, which is unlikely. It is the widest this answer gets, not a chance of anything.
Handle additional AI-escalated exceptions — how much new work this creates — decides 90 hrs of that width on its own. It is the number worth measuring rather than estimating.
| Hrs decided | Which number | Whose work |
|---|---|---|
| 90 | Handle additional AI-escalated exceptions — how much new work this creates | Senior Adjuster |
| 51.7 | Review AI settlement recommendation — how long each piece of new work takes | Claims Specialist |
| 8.8 | Screen for fraud indicators — share still passing through a person | Senior Adjuster |
Ten questions no calculation can settle. They are asked of every case, and the answers below are the analyst's, not the tool's.
Is work being taken off a role that still needs doing? Who does it now?
Chasing documents does not disappear when the assistant sends the reminders. Someone still has to judge when a claimant is unresponsive versus confused. Currently unassigned in the future state.
How many exceptions do you really expect, and who handles them?
The vendor projects 8% escalation. Our own disputed-claim rate is 13.6% today and nobody could explain why the assistant would lower it. Modelled at 120/week.
8 of 10 were not answered. They are listed here because leaving a question open is permitted and hiding that it was left open is not.
These are counts, not a grade. There is no score here and there will not be one — a single figure invites gaming and implies an opinion this tool does not have.
This summary does not count whether the decisions inside this workflow have owners with real authority; whether anyone is actually offering the work people would move into; whether the people doing this work have confirmed the description. Those need work this analysis has not done.