Skip to content
brianletort.aiD. Brian Letort, Ph.D.

AI Transformation Library

See how AI changes the work.

105 real-world cases. What changed, what improved, and where people still decide.

CardsTable

105 cases

25–48

CaseIndustryFunctionTypeAI valueBefore → AfterEvidenceOpen
An editorial scene for Sixty-eight English secondary schools in the EEF Teacher Choices trial contrasts creates questions, quizzes, activities, and differentiated resources without generative ai. with uses an implementation guide and prompts the model for questions, quizzes, activity ideas, or tailored materials. in the teacher-controlled chatgpt lesson planning with implementation guide workflow.Sixty-eight English secondary schools in the EEF Teacher Choices trialEducationEnrollmentStep before the decision56.2 minutes in the ChatGPT group, 25.3 minutes or 31% lower.PrepareDecideverified evidenceverified
An editorial scene for Estonian Unemployment Insurance Fund contrasts provides registration information to the estonian unemployment insurance fund. with combines registry events and client changes to refresh employment-prospect assessments regularly. in the ott employment-prospect decision support workflow.Estonian Unemployment Insurance FundGovernment & public servicesCustomer serviceAutonomous + backstopOTT1 achieved over 95% prediction accuracy; OTT2 above 98% remains a 2026 pilot targetManualAutonomousreported evidencereported
An editorial scene for Three Italian endoscopy centers using GI Genius contrasts inspect the live colonoscopy feed and visually identify suspected lesions. with processes the live video and superimposes a green box over a suspected lesion. in the real-time ai computer-aided adenoma detection during colonoscopy workflow.Three Italian endoscopy centers using GI GeniusHealthcareClinical operationsCreator to judge1.07 in the CADe group; incidence-rate ratio 1.46, with no increase in withdrawal timeCreateReviewverified evidenceverified
An editorial scene for GitHub Copilot research experiment contrasts reads the http-server specification and writes javascript without copilot. with offers inline code completions while the developer implements the same server task. in the ai pair-programmed http server workflow.GitHub Copilot research experimentTechnology & softwareSoftware engineeringCreator to judge71.17 minutes treatment; 55.8% faster, P=.0017CreateReviewverified evidenceverified
An editorial scene for Google contrasts predict cooling-energy effects and recommend operating changes. with every five minutes evaluates sensor data and selects energy-minimizing actions that satisfy safety constraints. in the deepmind autonomous cooling control workflow.GoogleTechnology & softwareEngineering designAutonomous + backstopAround 30% average energy savings after nine months, improving from 12% at launchManualAutonomousmixed evidencemixed
An editorial scene for Google Research Flood Hub contrasts uses gauge-calibrated or global physical models to produce river forecasts. with transfer learning from open data to produce river-flow and inundation forecasts up to seven days. in the ai forecasts for ungauged basins workflow.Google Research Flood HubGovernment & public servicesEngineering designContinuous decisioningSimilar or better reliability at up to five daysPeriodicContinuousverified evidenceverified
An editorial scene for Georgia State University contrasts track multiple enrollment requirements and seek help when a task or deadline is unclear. with use university records to send task-specific deadline reminders and offer help. in the pounce summer-melt support workflow.Georgia State UniversityEducationEnrollmentExceptions onlyStudents assigned to Pounce were 3.3 percentage points more likely to enroll on time, equivalent to a 21% reduction in summer melt in the peer-reviewed study.Handle allExceptionsmixed evidencemixed
An editorial scene for IKEA (Ingka Group) contrasts receive routine customer enquiry with resolve routine customer enquiry in the contact-center reskilling workflow.IKEA (Ingka Group)Retail & consumerCustomer serviceCreator to judge47%CreateReviewmixed evidencemixed
An editorial scene for Johns Hopkins Health System contrasts recognizes possible sepsis through ordinary ehr review and clinical presentation. with continuously monitors ehr data and passively displays a clickable alert icon when criteria are met. in the trews provider-confirmed early warning workflow.Johns Hopkins Health SystemHealthcareClinical operationsCreator to judge1.85 hours earlier when confirmed within three hoursCreateReviewverified evidenceverified
An editorial scene for JPMorgan Chase contrasts read commercial-credit agreements and manually identify relevant clauses and attributes. with uses unsupervised machine learning to extract 150 defined attributes from agreements. in the coin agreement extraction workflow.JPMorgan ChaseFinancial servicesCredit and lendingExceptions only150 attributes from 12,000 agreements extracted in secondsHandle allExceptionsreported evidencereported
An editorial scene for Justice Connect contrasts selects a legal category from a fixed list, often choosing the wrong area or 'something else'. with describes the legal problem in ordinary language without selecting an area of law. in the free-text legal-problem classification workflow.Justice ConnectGovernment & public servicesCustomer serviceIntake redesigned9% among users offered the classifierFormsDescribereported evidencereported
An editorial scene for Kaiser Permanente Northern California contrasts identify deterioration through routine observation and manually calculated early-warning scores. with calculate hourly deterioration-risk scores from electronic health-record data. in the advance alert monitor workflow.Kaiser Permanente Northern CaliforniaHealthcareClinical operationsHuman quality loop30-day mortality after alert: 20.4% → 15.8%One passLearning loopverified evidenceverified
An editorial scene for Mayo Clinic primary-care practices contrasts order ecgs for routine clinical reasons and evaluate heart failure using usual-care signals. with analyzes routine ecgs and reports high likelihood of low ejection fraction. in the eagle ai-ecg clinical decision support workflow.Mayo Clinic primary-care practicesHealthcareCustomer serviceContinuous decisioning2.1% with clinician access to AI-ECG results; odds ratio 1.32, P=0.007PeriodicContinuousverified evidenceverified
An editorial scene for NASA JPL contrasts use orbital and rover imagery to define route goals and commanded drive segments. with builds local terrain maps, checks collision/slip risk, and selects paths onboard. in the perseverance enhanced autonomous navigation workflow.NASA JPLScience & researchEngineering designCreator to judgeAbout 90% of 32.1 km used ENavCreateReviewverified evidenceverified
An editorial scene for NHS England stroke networks contrasts acquire ct imaging and wait for local interpretation and sequential specialist consultation. with analyzes ct images and rapidly surfaces features consistent with large-vessel occlusion. in the brainomix e-stroke ai decision support in the thrombectomy pathway workflow.NHS England stroke networksHealthcareCustomer serviceExceptions only79 minutes after deployment, a 62-minute reductionHandle allExceptionsverified evidenceverified
An editorial scene for Octopus Energy contrasts read and draft response to customer email with draft response to customer email (2023) or answer straightforward emails autonomously (2026) in the ai-drafted customer email / arlo assistant workflow.Octopus EnergyEnergy & utilitiesCustomer serviceCreator to judge34%CreateReviewmixed evidencemixed
An editorial scene for City of Pittsburgh and Carnegie Mellon University contrasts create fixed coordinated timing plans for rush periods and simple actuated control otherwise. with recomputes green-time allocation every one to two seconds from incoming flows. in the surtrac decentralized adaptive traffic-signal control workflow.City of Pittsburgh and Carnegie Mellon UniversityTransportation & logisticsEngineering designCoordination compressed25.79% lower travel time and 40.64% lower wait time; projected emissions 21.48% lowerSequentialTogetherverified evidenceverified
An editorial scene for Rwanda national blood service and Zipline contrasts maintain local blood inventory or request an emergency road delivery. with places an on-demand blood order by phone or text. in the on-demand autonomous drone blood delivery workflow.Rwanda national blood service and ZiplineHealthcareClinical operationsAutonomous + backstopMean drone delivery 49.6 minutes, 79 minutes faster than regional-center road estimates; expirations fell 7.1 units per month, a 67% reduction at 12 monthsManualAutonomousverified evidenceverified
An editorial scene for AIA Group contrasts submit physical forms and paper medical documentation with photograph invoices and documents with a phone in the genai-enabled medical claims straight-through processing workflow.AIA GroupFinancial servicesInsuranceStraight-through73% (December 2024)HandoffsStraight throughmixed evidencemixed
An editorial scene for Microsoft Azure contrasts fire alerts inside individual azure services as local symptoms appear. with continue to fire alerts; warden ingests them globally from icm. in the warden automated incident detection in icm workflow.Microsoft AzureTechnology & softwareIT operationsCoordination compressedWarden faster in ~68% of successfully detected incidents; median saving 21.8 NTUs, about 15% of time-to-mitigate for those casesSequentialTogetherverified evidenceverified
An editorial scene for German organised mammography screening programme (PRAIM implementation study; 12 sites) contrasts independently double-read four-view mammograms without ai predictions. with tags confident-normal cases and fires a localization safety net when humans call suspicious cases normal. in the ai-supported double reading with normal triaging and a post-read safety net (vara mg) workflow.German organised mammography screening programme (PRAIM implementation study; 12 sites)HealthcareHealthcare screeningAutonomous + backstop6.70 per 1,000 in the AI-supported group; +17.6% (95% CI +5.7% to +30.8%) after overlap weightingManualAutonomousverified evidenceverified
An editorial scene for Google contrasts submits a change after limited presubmit testing. with submits a change after limited presubmit testing. in the postsubmit speculative cycles with transition prediction workflow.GoogleTechnology & softwareSoftware engineeringContinuous decisioning37 minutes with Speculative Cycles (approximately 65% / 70-minute reduction)PeriodicContinuousverified evidenceverified
An editorial scene for United States Internal Revenue Service contrasts calls and waits for a live assistor for common collection information or payment-plan help. with answers common questions, authenticates eligible users, and creates or modifies permitted payment plans. in the authenticated and unauthenticated collection voice and chat bots workflow.United States Internal Revenue ServiceGovernment & public servicesEngineering designCreator to judgeMore than 13 million taxpayers received voice/chatbot service without waiting.CreateReviewreported evidencereported
An editorial scene for Meta (Facebook) contrasts reports a bug or crash against a specific line of code. with reports a bug against a specific line and forwards its metadata to getafix. in the getafix automated patch suggestion and validation for null-dereference bugs workflow.Meta (Facebook)Technology & softwareSoftware engineeringCreator to judgeGetafix attempted patches for about 60% of the ~2,000 null-method-call bugs; about 90% of attempted patches passed automated validation (compilable and Infer no longer emitted the warning); overall, 1,077 bugs (approximately 53%) were successfully auto-patched.CreateReviewverified evidenceverified
Curated collections →Evidence methodologyPublication history
Transformations | Brian Letort