PK °%‚\_rels/PK °%‚\ docProps/PK °%‚\ppt/PK °%‚\ ppt/_rels/PK °%‚\ ppt/charts/PK °%‚\ppt/charts/_rels/PK °%‚\ppt/embeddings/PK °%‚\ ppt/media/PK °%‚\ppt/slideLayouts/PK °%‚\ppt/slideLayouts/_rels/PK °%‚\ppt/slideMasters/PK °%‚\ppt/slideMasters/_rels/PK °%‚\ ppt/slides/PK °%‚\ppt/slides/_rels/PK °%‚\ ppt/theme/PK °%‚\ppt/notesMasters/PK °%‚\ppt/notesMasters/_rels/PK °%‚\ppt/notesSlides/PK °%‚\ppt/notesSlides/_rels/PK °%‚\Ù—²q00[Content_Types].xml PK °%‚\ðÜÈø]] _rels/.rels PK °%‚\È!”ßdocProps/app.xml 0 0 Microsoft Office PowerPoint On-screen Show (16:9) 0 8 8 0 0 false Fonts Used 2 Theme 1 Slide Titles 8 Arial Calibri Office Theme Slide 1Slide 2Slide 3Slide 4Slide 5Slide 6Slide 7Slide 8 PptxGenJS false false false 16.0000 PK °%‚\JDÄ@$$docProps/core.xml ShieldScore — Executive Recommendation PptxGenJS Presentation ShieldScore Analytics Team ShieldScore Analytics Team 1 2026-04-02T04:45:32Z 2026-04-02T04:45:32Z PK °%‚\Á¬g ppt/_rels/presentation.xml.rels PK °%‚\OÝ¨Í Í ppt/theme/theme1.xmlPK °%‚\ëõÚ§™ ™ ppt/presentation.xml PK °%‚\X›Âppt/presProps.xml PK °%‚\Øý¶¶ppt/tableStyles.xml PK °%‚\D >00ppt/viewProps.xml PK °%‚\H7ût¯¯!ppt/slideLayouts/slideLayout1.xml PK °%‚\ÕÑ’ñ77,ppt/slideLayouts/_rels/slideLayout1.xml.rels PK °%‚\¸‡tiŠŠppt/slides/slide1.xml ShieldScore A proactive risk identification system that can reduceErie Insurance's preventable auto claims by 15%,saving an estimated $27 million annuallyCONFIDENTIALErie Insurance Group | Analytics Team | April 2026PK °%‚\¯›3öÎÎ ppt/slides/_rels/slide1.xml.rels PK °%‚\.šš‘ppt/notesSlides/notesSlide1.xml 1PK °%‚\»:AËË*ppt/notesSlides/_rels/notesSlide1.xml.rels PK °%‚\–©ãžÜ,Ü,ppt/slides/slide2.xml Erie spends $180M annually on auto claims over $5K — an estimated 20% of which are preventable through proactive policyholder engagement$180MAnnual auto claims >$5KErie's single largest cost category after reinsurance. Growing at 4% annually driven by repair cost inflation.~20%Estimated preventableIndustry benchmark: insurers using proactive alerts see 15–25% claim reduction. IBM found 52% of alerted policyholders took preventive action.87%ML projects fail to deployThe biggest risk isn't model accuracy — it's never shipping the solution. We've built the deployment tool alongside the model.Source: Erie Insurance internal claims data (FY2025); IBM Institute for Business Value, "Insurance in the AI Era" (2025); VentureBeat ML deployment research (2019)1PK °%‚\‡2-®ÎÎ ppt/slides/_rels/slide2.xml.rels PK °%‚\ÙÀ·ppt/notesSlides/notesSlide2.xml 2PK °%‚\xÕ·ËË*ppt/notesSlides/_rels/notesSlide2.xml.rels PK °%‚\6\(xmxmppt/slides/slide3.xml ShieldScore assigns each policyholder a 0–100 risk score using 16 data points, enabling four tiers of proactive intervention before claims occur PPREDICTProbability of >$5K claimin next 12 months→ AACTSafe driving programs,weather alerts, telematics→ IIMPACTReduced claims cost,improved retention→ RRISKFalse positives, ZIP bias,privacy concernsRisk TiersLOW (0–29)Standard monitoringMEDIUM (30–49)Monthly newsletterHIGH (50–69)Proactive check-inCRITICAL (70–100)Immediate outreach2PK °%‚\ W/ÎÎ ppt/slides/_rels/slide3.xml.rels PK °%‚\K |Åppt/notesSlides/notesSlide3.xml 3PK °%‚\9 YÙËË*ppt/notesSlides/_rels/notesSlide3.xml.rels PK °%‚\'WÚTÖnÖnppt/slides/slide4.xml Four variables drive 60% of claim prediction power — prior claims history alone accounts for more than a quarter of model accuracyFeature importance (% contribution to prediction)Prior claims (3yr)22%At-fault accidents15%Credit score12%Age9%Annual mileage8%Payment timeliness7%Vehicle age6%Telematics enrolled5% Key Findings4×Policyholders with 3+ prior claims are 4× more likely to file a >$5K claim2×Late payers file claims twice as often as on-time payers−40%Telematics-enrolled customers show 40% lower claim ratesU-shapeDrivers under 25 and over 70 carry disproportionate riskSource: ShieldScore model analysis on 5,000 policyholder dataset; feature importance from gradient boosted classifier3PK °%‚\×`ÎÎ ppt/slides/_rels/slide4.xml.rels PK °%‚\vsœÿppt/notesSlides/notesSlide4.xml 4PK °%‚\¿JŒ ËË*ppt/notesSlides/_rels/notesSlide4.xml.rels PK °%‚\Ÿà\±è3è3ppt/slides/slide5.xml The model correctly identifies 72% of high-risk policyholders while maintaining a false alarm rate under 30% — well within the cost-optimal threshold72%of high-risk policyholderscorrectly identified<30%false alarm rate(unnecessary outreach)Cost asymmetry drives thresholdCost of missed claim: $5,000+Cost of false alarm: ~$20 250:1 cost ratio strongly favorscatching more risk, even at theprice of some false alarmsValidation: Model tested on held-out data the algorithm never saw. Performance is consistent (AUC 0.86 validation → 0.85 test), confirming no overfitting. Model is ready for production.4PK °%‚\ð5ŸÎÎ ppt/slides/_rels/slide5.xml.rels PK °%‚\ä¸W8ppt/notesSlides/notesSlide5.xml 5PK °%‚\þQeËË*ppt/notesSlides/_rels/notesSlide5.xml.rels PK °%‚\!oý„Ç/Ç/ppt/slides/slide6.xml ShieldScore deploys through two channels — a nightly batch run that generates the outreach list, and a real-time lookup for in-call agent decisions BATCH SCORINGRuns nightly at 11 PM on all active policyholdersGenerates prioritized outreach list by 6 AMFeeds retention team dashboard and CRMUpdates customer risk segments for marketing REAL-TIME SCORINGFires when agent opens a policyholder accountDisplays ShieldScore + tier + recommended actionEnables in-call decision (offer telematics discount?)Sub-second response via simplified decision rulesDeployment tool: An Excel-based scorer replicates the model's decision logic using simplified rules. Underwriters can score individual policyholders manually — no API required. This ensures the tool ships on day one, not "when engineering has bandwidth."5PK °%‚\ج+ÇÎÎ ppt/slides/_rels/slide6.xml.rels PK °%‚\âz«ppt/notesSlides/notesSlide6.xml 6PK °%‚\=|”ÖËË*ppt/notesSlides/_rels/notesSlide6.xml.rels PK °%‚\M*“Œšpšpppt/slides/slide7.xml Projected impact of $27M in annual savings requires a $200K implementation investment — a 135× return, with measurable results within 6 months$27Mestimated annualclaim reduction15%fewer preventable>$5K claims135×ROI on $200Kimplementation cost6 moto measurablepilot resultsRisks & mitigations Risk Mitigation Credit score bias Quarterly disparate impact audit; alternative models for ban states ZIP-code proxies for race Evaluate driving-behavior features as replacement; monitor by demographic Model degradation over time Weekly AUC monitoring; quarterly retrain; auto-alert if AUC < 0.78 Telematics privacy concerns Opt-in only; transparent data usage policy; customer consent governance 6PK °%‚\ÿÉFÎÎ ppt/slides/_rels/slide7.xml.rels PK °%‚\)±lppt/notesSlides/notesSlide7.xml 7PK °%‚\|g¸ËË*ppt/notesSlides/_rels/notesSlide7.xml.rels PK °%‚\­¿à-/-/ppt/slides/slide8.xml RecommendationApprove a 500-policyholder pilot in the Erie, PA region for Q2 2026, with full-territory rollout contingent on achieving 15% claim reduction in the pilot group1Deploy ShieldScore to 500 policyholders in Erie region (May 2026)2Integrate risk scores into agent dashboard for real-time visibility3Launch safe-driving intervention program for Critical and High tiers4Measure claim reduction vs. matched control group over 6 months5If 15% target met: scale to full 12-state territory by Q1 20277PK °%‚\6Â¥ÎÎ ppt/slides/_rels/slide8.xml.rels PK °%‚\iºÞppt/notesSlides/notesSlide8.xml 8PK °%‚\p÷O¨ËË*ppt/notesSlides/_rels/notesSlide8.xml.rels PK °%‚\Kà ——!ppt/slideMasters/slideMaster1.xml PK °%‚\NÇ)Ž¾¾,ppt/slideMasters/_rels/slideMaster1.xml.rels PK °%‚\6á½TT!ppt/notesMasters/notesMaster1.xml 7/23/19Click to edit Master text stylesSecond levelThird levelFourth levelFifth level‹#›PK °%‚\s® **,ppt/notesMasters/_rels/notesMaster1.xml.rels PK °%‚\_rels/PK °%‚\ $docProps/PK °%‚\Kppt/PK °%‚\ mppt/_rels/PK °%‚\ •ppt/charts/PK °%‚\¾ppt/charts/_rels/PK °%‚\íppt/embeddings/PK °%‚\ ppt/media/PK °%‚\Bppt/slideLayouts/PK °%‚\qppt/slideLayouts/_rels/PK °%‚\¦ppt/slideMasters/PK °%‚\Õppt/slideMasters/_rels/PK °%‚\  ppt/slides/PK °%‚\3ppt/slides/_rels/PK °%‚\ bppt/theme/PK °%‚\Šppt/notesMasters/PK °%‚\¹ppt/notesMasters/_rels/PK °%‚\îppt/notesSlides/PK °%‚\ppt/notesSlides/_rels/PK °%‚\Ù—²q00P[Content_Types].xmlPK °%‚\ðÜÈø]] ±_rels/.relsPK °%‚\È!”ß7docProps/app.xmlPK °%‚\JDÄ@$${!docProps/core.xmlPK °%‚\Á¬g Î$ppt/_rels/presentation.xml.relsPK °%‚\OÝ¨Í Í -ppt/theme/theme1.xmlPK °%‚\ëõÚ§™ ™ Nppt/presentation.xmlPK °%‚\X›Âä[ppt/presProps.xmlPK °%‚\Øý¶¶2]ppt/tableStyles.xmlPK °%‚\D >00^ppt/viewProps.xmlPK °%‚\H7ût¯¯!xappt/slideLayouts/slideLayout1.xmlPK °%‚\ÕÑ’ñ77,fdppt/slideLayouts/_rels/slideLayout1.xml.relsPK °%‚\¸‡tiŠŠçeppt/slides/slide1.xmlPK °%‚\¯›3öÎÎ ¤yppt/slides/_rels/slide1.xml.relsPK °%‚\.šš‘°{ppt/notesSlides/notesSlide1.xmlPK °%‚\»:AËË*}‚ppt/notesSlides/_rels/notesSlide1.xml.relsPK °%‚\–©ãžÜ,Ü,„ppt/slides/slide2.xmlPK °%‚\‡2-®ÎÎ Ÿ±ppt/slides/_rels/slide2.xml.relsPK °%‚\ÙÀ·«³ppt/notesSlides/notesSlide2.xmlPK °%‚\xÕ·ËË*xºppt/notesSlides/_rels/notesSlide2.xml.relsPK °%‚\6\(xmxm‹¼ppt/slides/slide3.xmlPK °%‚\ W/ÎÎ 6*ppt/slides/_rels/slide3.xml.relsPK °%‚\K |ÅB,ppt/notesSlides/notesSlide3.xmlPK °%‚\9 YÙËË*3ppt/notesSlides/_rels/notesSlide3.xml.relsPK °%‚\'WÚTÖnÖn"5ppt/slides/slide4.xmlPK °%‚\×`ÎÎ +¤ppt/slides/_rels/slide4.xml.relsPK °%‚\vsœÿ7¦ppt/notesSlides/notesSlide4.xmlPK °%‚\¿JŒ ËË*­ppt/notesSlides/_rels/notesSlide4.xml.relsPK °%‚\Ÿà\±è3è3¯ppt/slides/slide5.xmlPK °%‚\ð5ŸÎÎ 2ãppt/slides/_rels/slide5.xml.relsPK °%‚\ä¸W8>åppt/notesSlides/notesSlide5.xmlPK °%‚\þQeËË* ìppt/notesSlides/_rels/notesSlide5.xml.relsPK °%‚\!oý„Ç/Ç/îppt/slides/slide6.xmlPK °%‚\ج+ÇÎÎ ppt/slides/_rels/slide6.xml.relsPK °%‚\âz«$ ppt/notesSlides/notesSlide6.xmlPK °%‚\=|”ÖËË*ñ&ppt/notesSlides/_rels/notesSlide6.xml.relsPK °%‚\M*“Œšpšp)ppt/slides/slide7.xmlPK °%‚\ÿÉFÎÎ Ñ™ppt/slides/_rels/slide7.xml.relsPK °%‚\)±lÝ›ppt/notesSlides/notesSlide7.xmlPK °%‚\|g¸ËË*ª¢ppt/notesSlides/_rels/notesSlide7.xml.relsPK °%‚\­¿à-/-/½¤ppt/slides/slide8.xmlPK °%‚\6Â¥ÎÎ Ôppt/slides/_rels/slide8.xml.relsPK °%‚\iºÞ)Öppt/notesSlides/notesSlide8.xmlPK °%‚\p÷O¨ËË*öÜppt/notesSlides/_rels/notesSlide8.xml.relsPK °%‚\Kà ——! ßppt/slideMasters/slideMaster1.xmlPK °%‚\NÇ)Ž¾¾,ßýppt/slideMasters/_rels/slideMaster1.xml.relsPK °%‚\6á½TT!çÿppt/notesMasters/notesMaster1.xmlPK °%‚\s® **,zppt/notesMasters/_rels/notesMaster1.xml.relsPKCC²î