No result found
This report needs a scored result to generate.
This is a static sample and does not depend on a scored result.
Back to the course page
PRAXORALAB Illustrative example · fictional company, for format reference only
Redesigned Workflow, S.E.N.S.E. Framework Applied
Coral Bay Insurance
Built for one real team brought into the room. Coral Bay Insurance's example team: the Motor Claims Intake & Assessment Team, redesigning AI-assisted first-notice-of-loss intake and drafted assessment notes for motor claims. Walked through all five stages of the S.E.N.S.E. framework, Spot, Empathise, Narrow, Solution, Evaluate, not assigned from a template.
S — Spot
Coral Bay's original brief asked the AI tool to draft entire settlement recommendations end to end, across all seven steps of the motor claims process.
Walked against the team's actual day, the tool was reliably strong at exactly two of those seven steps: capturing first-notice-of-loss data from submitted forms and photos, and populating the factual fields of the standard assessment-note template. It was not reliable at severity banding or settlement figures.
E — Empathise
Structured interviews with all eight claims officers surfaced the real friction point: not drafting text, but re-keying the same claim data three times across the intake portal, the assessment tool and the payment system, interrupted constantly by phone-based first-notice-of-loss calls.
Not one officer named drafting speed as the bottleneck. The original AI-adoption brief had assumed it was, because nobody had asked the team directly before writing that brief.
N — Narrow
The redesign was narrowed to a single highest-value point: the intake-to-first-draft handoff, not the full seven-step claims lifecycle.
Settlement negotiation and fraud-pattern escalation were explicitly left untouched in this pass, pending the capability gap map on the next page, rather than redesigned on the same day as everything else.
S — Solution
The AI tool captures structured intake data and drafts the factual scaffold of the assessment note: dates, policy numbers, and damage descriptions read from submitted photos.
A claims officer verifies the photo-to-damage mapping, owns severity banding and the settlement recommendation, and handles any customer conversation. No step where AI output reaches a customer or a payment is unsupervised.
E — Evaluate
The redesigned draft was run side by side against the old process on 15 real closed claims, with a named officer signing off every one, not assumed correct because a demo looked convincing.
That comparison is not treated as final. It is re-run in Week 6 of the roadmap on the next page, against a baseline set in Week 1, before the redesign is extended past the three officers who piloted it.
Result — the redesigned intake-to-draft workflow
First-notice-of-loss capture — AI drafts from the submitted form and photos, officer verifies within four hours.
Assessment-note scaffold — AI drafts the factual fields, officer edits and corrects the photo-to-damage mapping.
Severity banding — fully human; the officer's judgement call, not the tool's.
Settlement recommendation — fully human; the team lead co-signs any recommendation above S$8,000.
Fraud-pattern flag review — AI flags statistical anomalies only; interpreting and escalating a flag stays fully human.
The detail worth sitting with on this page is not the redesigned workflow itself, it is how many steps of the original seven-step process Spot eliminated before anyone touched a solution. Coral Bay's leadership walked in wanting the tool to draft settlement recommendations end to end, and that instinct is not unusual, it is what most AI-adoption briefs default to when nobody has separated "the AI is good at this" from "this is where the team spends its time." Spot found the tool reliable at exactly two of seven steps. That is not a disappointing result. It is the result that keeps a claims team from putting an unreliable settlement figure in front of a customer, which is a considerably worse outcome than a slower rollout.
Empathise is the stage I watch most carefully in every session I run, because it is the one most teams skip under time pressure, and Coral Bay's result shows exactly why that is a mistake. The original brief assumed drafting speed was the bottleneck. Not one of the eight officers named drafting speed when asked directly. The actual bottleneck, re-keying the same claim data three times across three systems, is a problem an AI drafting tool does not solve at all; it is closer to a systems-integration problem than a workflow-redesign one. If Coral Bay had gone straight from brief to solution, as most vendor-led rollouts do, they would have shipped a well-drafted note generator that left the actual daily frustration completely untouched.
Narrow is where I think this team showed real discipline, and it is worth naming explicitly because it is the step most likely to get cut when a steering committee wants a bigger announcement. Deciding not to touch settlement negotiation or fraud escalation in the same pass, deferring both to the capability gap map on the next page, is the correct sequencing precisely because those two areas turned out to have genuine capability gaps behind them. Redesigning a workflow around a capability the team does not yet have is how a pilot looks successful in week one and quietly fails by week eight.
What I would flag as the one thing to watch going forward is Evaluate's own honesty: a 15-claim side-by-side comparison is a start, not a validated result, and the plan correctly re-runs it in Week 6 before extending past the three pilot officers. I would not sign off on this redesign as finished until that second evaluation exists on paper, not only as an intention stated here.
PRAXORALAB Capability Gap Map
What Motor Claims Intake & Assessment Team already has, built from evidence rather than a generic competency framework, and what is genuinely missing. Each area is scored against what the redesigned workflow on the previous page actually depends on, not against an abstract skills matrix.
Policy & coverage interpretation
Already have it. Officers already read policy wording against real damage claims daily. This transfers directly into the redesigned workflow; no further work needed here.
Photo-to-damage assessment judgement
Already have it, just never written down. The tacit skill of matching photo evidence to a repair-cost band is already there. It has never been documented as a checklist a newer officer, or an AI draft, could be checked against.
Fraud-pattern recognition
Real, but inconsistent. Two of the eight officers catch most flagged fraud through instinct built over years on the team. Neither can yet explain the pattern clearly enough for a colleague to learn it from them.
AI-output verification literacy
Genuine gap. Nobody on the team is currently trained to systematically check an AI-drafted assessment note against the source photos before it is signed off. This sits directly inside the redesigned workflow's second step.
Escalation judgement for AI-flagged anomalies
Genuine gap. The tool is new enough that no one owns the decision of when a statistical fraud flag is worth pursuing versus noise. Today it is whoever happens to notice the flag first.
Priority: AI-output verification literacy
Both genuine gaps matter, but this one sits directly inside the redesigned workflow's second step. Until someone can systematically check an AI-drafted assessment note against its source photos, that step cannot run safely at full team scale.
Five areas, two bands doing real work, and I want to be precise about the distinction between them, because collapsing "already have it" and "genuine gap" into one undifferentiated skills list is exactly the generic-competency-framework failure mode this method was built to avoid. Policy interpretation and photo-to-damage judgement both score as capability the team already holds; the only difference between them and a textbook "strength" is that the second one has never been written down. That is not a training need. It is a documentation task, and treating it as a training need is how organisations end up sending experienced staff to a course teaching them something they already know how to do.
Fraud-pattern recognition sitting in the middle band is the one I would spend the most time on if this were a live session. Two of eight officers catching most flagged fraud through instinct is a genuinely valuable asset, and it is also a single point of failure: if either of those two officers leaves, that capability leaves with them, uncoded and untransferable. Scoring it as partially-transferring rather than fully-have is the honest call, and it is also the more urgent one, because unlike the two fully-have areas, this one has a clock on it that has nothing to do with the AI rollout at all.
The two genuine gaps are where I would push back hardest on any team tempted to sequence them in parallel. AI-output verification literacy sits directly inside the redesigned workflow's second step, the assessment-note scaffold, which means it is not an abstract future need, it is a dependency the new process already has and does not yet have an owner for. Escalation judgement for AI-flagged anomalies matters, but the tool has not been running long enough yet for that gap to have caused a problem. Coral Bay's Month 1 plan correctly leads with verification literacy for exactly this reason: the workflow they redesigned on page one cannot actually run safely until this specific gap closes, and every week it stays open is a week an unverified AI draft could reach a customer file uncaught.
I would call this gap map honest rather than flattering, which is the standard I hold every one of these to. A flattering gap map tells a team what it wants to hear. This one tells Coral Bay exactly where its redesigned workflow is currently exposed.
Coral Bay Insurance and its capability ratings are invented for this sample only, to show the shape of the output, not a real client's actual team. In the session, this gap map is built from evidence gathered about the team a participant brings, not assigned from a template.
PRAXORALAB First-Quarter Learning Plan
The redesigned workflow and capability gap map above, brought together into one sequenced plan for Motor Claims Intake & Assessment Team, ready to hand to L&D, ordered by what each step actually depends on rather than by department.
- Document the photo-to-damage assessment judgement as a checklist, drawn from the two most experienced officers' own tacit knowledge, not a generic training module.
- Run a structured verification-literacy session: how to audit an AI-drafted assessment note against source photos before signing off.
- Set the baseline: track how many AI-drafted notes are accepted unmodified in the first two weeks, so the Week 6 evaluation has something real to compare against.
- Turn the two instinct-driven officers' fraud-pattern recognition into a checklist the rest of the team can be taught from.
- Name a specific escalation owner for AI-flagged anomalies, rather than leaving it as whoever notices first.
- Re-run the Week 6 evaluation against the Month 1 baseline, on the same 15 claims used in the original side-by-side comparison.
- Extend the assessment-note scaffold to the full eight-officer team, not only the three who piloted it.
- Re-score the capability gap map against these same five areas, and compare against this quarter's starting point.
- Report results to the Head of Claims Operations, with a recommendation on whether to extend AI use into settlement-recommendation drafting next quarter.
The reason this plan runs foundation, then structure, then embed, and not the reverse, is the same sequencing logic that made Narrow work on page one: each month's output is a direct input to the next, and skipping ahead produces work that has to be redone. Month 1's two documentation tasks, the photo-to-damage checklist and the fraud-pattern checklist groundwork, exist because Month 2's structural work, naming an escalation owner and teaching the fraud-pattern checklist to the rest of the team, has nothing to structure without them. I have watched organisations name an escalation owner before anyone had actually written down what that owner is meant to be judging, and the result is a title on an org chart with no method behind it.
The baseline captured in Month 1, tracking how many AI-drafted notes get accepted unmodified in the first two weeks, is the detail I would not let a client cut under time pressure, because without it Month 2's Week 6 re-evaluation has nothing honest to compare against. A team under pressure to show progress will often skip straight to a satisfaction survey instead, and a satisfaction survey measures how officers feel about the new process, not whether the process is actually performing better than the one it replaced.
Month 2 is where I think the real test of this plan sits, specifically in naming a single accountable owner for AI-flagged anomalies rather than leaving it distributed across whoever happens to be on shift. Distributed ownership of an escalation decision is not really ownership at all; it is a decision nobody is actually accountable for making well, which is precisely the gap identified on the capability gap map. Fixing that gap on paper and fixing it in practice are two different achievements, and Month 2 is where Coral Bay finds out which one they actually did.
Month 3's extension to the full eight-officer team, gated behind the Month 2 re-evaluation rather than launched on Month 1's optimism, is the correct discipline, and the re-scored gap map at the end of the quarter is what turns "we ran a pilot" into a number the Head of Claims Operations can actually hold the team to next quarter. What "done" looks like at week twelve is not a perfect gap map. It is a Coral Bay claims team that can explain, to their own management and to a regulator if asked, exactly why an AI-drafted note was trusted and who checked it.
PRAXORALAB What Must Be Deliberately Strengthened
As AI takes on more of Motor Claims Intake & Assessment Team's routine work, the evidence gathered above points to a specific, named answer, not a general call for more AI literacy.
5 of 8
claims officers had never had their photo-to-damage assessment judgement formally documented, despite averaging over six years on the team.
Internal skills audit, illustrative
62%
of intake-to-draft task hours were spent on repetitive re-keying across three systems, not on judgement calls, before the redesign.
Internal time study, illustrative
AI-output verification literacy and fraud-pattern recognition, not general AI literacy, are what must be deliberately strengthened. Both are named capabilities the redesigned workflow now depends on, and both currently have no formal owner: verification literacy because nobody has been trained to audit an AI draft against its source, and fraud-pattern recognition because it lives entirely inside two officers' individual judgement with no way yet for a colleague to learn it from them.
If neither capability is strengthened, the redesigned workflow keeps running on paper while its two real dependencies stay unmet: an unverified AI draft could reach a customer file uncaught, and the team's actual fraud-detection capability stays a single point of failure resting on two people.
The figure I would put in front of Coral Bay's leadership first is the 62 percent, not because it is dramatic, but because it directly contradicts the assumption their original brief was built on. If drafting speed were genuinely the constraint, an AI drafting tool would have been the right first move. Sixty-two percent of intake-to-draft hours going to repetitive re-keying across three systems means the team's actual constraint was never speed of writing, it was the absence of a single source of truth for a claim's basic facts. That is the finding that should have shaped the redesign brief in the first place, and it only surfaced because Empathise asked the team directly instead of assuming.
The five-of-eight figure matters for a different reason: it explains why "send the team to a generic AI-literacy course" was never going to be the right answer here, and I want to be specific about why, since that is the default most organisations reach for. A generic AI-literacy course teaches people how AI tools work in the abstract. It does not document the specific tacit judgement, photo-to-damage assessment, that five of eight officers already hold and have simply never had reason to write down. Sending experienced staff through a literacy module aimed at people who do not yet trust or understand AI at all would have wasted the exact capability this evidence proves is already there.
That is why the honest answer to what must be deliberately strengthened at Coral Bay is not "AI literacy" in general. It is two specific capabilities: AI-output verification literacy, because it sits inside a step the redesigned workflow already depends on and currently has no owner, and fraud-pattern recognition, because it currently lives entirely inside two people's judgement with no transfer mechanism behind it. Both are named, not generic, and both are traceable directly back to evidence gathered from this specific team rather than assumed from a competency framework built for someone else's organisation.
I would resist the temptation, which I have seen plenty of organisations give in to, to declare victory once the AI tool is drafting notes reliably. A tool performing well is not the same claim as a team being ready for what it changes about their work. Coral Bay's own numbers, the five officers with undocumented judgement, the two people who currently are the fraud-pattern-recognition function, say plainly that the harder work is still ahead of them, not behind.
Every score, quote and figure on these four pages is invented for Coral Bay Insurance, a fictional company, so the format of what a participant leaves with can be judged before enquiring. It is not a real client's deliverable, and no company named Coral Bay Insurance is a Praxora Lab client. The session itself redesigns your own team's own workflow, in the room, on the day.
3.5 hours, one session, facilitated by Michael Low.