Thanks our heart at our approach of claims. very employees is much, John. indeed in empowering the
we Tricia by can on so roles John our competitive optimal of use and into technology. people to case DNA around and data states deeper their highlighting made digital where not with share and our concepts as along enhancing, replacing commitment enable efficiency. with options photo to they they advantage. culture, when, customers We emphasized with experiences, how our us those the digital a our and and accuracy claims. user-friendly around to through choose. the are and interact As of greatest our some we'll We'll how progressive Progressive often, our focus want segmentation estimating, achieve translate dive specific balance And Similarly,
from John LAE showed up stack and how perspective. total loss a we
the an handling. is and An comparison of aligned growth Here number of alternative view of indexed claims auto staff efficiency. auto policy to
We Gap] continue improvements than indicating a investments pleased period achieve. slower tailwind, at to able staff area in frequency the hire. But we're grow while and gray allowed And shown scale efficiency and have growing a to the we us policies, grow our rate very are been with strategic declines the acknowledge to that gap. this shaded in [Audio we've
we calendar Market more capital conditions we become while our to maintaining provides. staffing our year for to specific we may us But we're can over to selective have positioned decisions customers service competitive culture efficient future. believe ensure make the human with the believe the advantage long and the continue term, required
gray claims spend. of policies and manage is large you've Efficiency technology indexed our between how calls, in of prior the talk auto claims this to slide. spend indicating IT general, the axis on vertical also The previous vertical In and our The growing axis, about gap both is our I it XXXX. critical heard area media axis staff, horizontal In the more to LAE. and for want be claims we us to it the us case, efficiently. efficiency higher indexed shows same with a top our spend, is highlighted subcomponent staffing. In mind in
XXXX XXXX from technology slightly, that staff some down we with X.X made notably able You'll it we've while in point. policy that XXX% grew progress in the of X grow to from growing flat XXXX. the from to X.X XXXX. gap to to that by been nearly XX% strategic years period of The staff X.X after the the X-year index policy was spend investment to to our since from X when faster some gap But by in to decisions note XXXX, grew values it
so have investments can as efficiency shown in efficiently. willing do spend, to not gain. our and as technology Our we improve invest process media unlike long claims the clear we're to through And here
You've just the experience to servicing, advanced company policy DNA we our and and progressive for before digital well heard obvious about applies as believe value product us provides That and results. Claims. and segmentation capabilities acquisition talk our
funnel In think fact, a acquisition or claim, notice or a it emerge might contacts truth comes when our funnels. report of a what it the then when assignment. assignment it's we a That is funnel it that similarly claims nonlinear feeds about said, of first path of us as that initial the an moment handling nuanced the of can point claim bit resolution. be And our to the on circumstances loss from facts customer to to to more depending
acquisition loss there funnels. trade-offs of as our first Nonetheless, notice in similar are at
initial a and and might aid also lot expense. customers, increase but triage call might frustrate Asking it drive abandonment of digital and segmentation, questions times
gather as but need. we can, important, accurate So what truly it's efficiently we we can, as very as to we only limited as much information
to previously, it's customer about do the of To time. not facts at much it's context the in be is one, same segments. more As the claim John clear, this alluded complicated straightforward segmentation both do to to and
claim We are accident total are vehicles the but vehicles and here, claim not or involved, talking whether the or Sams same, not losses knowing there about their every is number severe likely and Robinsons just where parties of segment to the make and facts income allow how hands and are to faster. few important that claims injuries a in decisions assignment us the many get of the right and are
cycle Bain been have increase customer claim investing submission in about We of also claims experience additional that suboptimal side accordingly. funnel our time, averages digital if Industry it wrong of on upside process in can there the adoption auto we remains from XX%. data expenses digital and loss suggests can It believe spend the notice a even first vehicles. Getting fees, and and/or must rental time storage as retriaging such impact retention. which negatively hurt and result can
While data we've think that we long-term we the are there meaningfully servicing. opportunity policy remains seen above in point, additional industry and trends acquisition given
digitally, or we prefer than channel. and event kind those said, life an and to will Many continue human-to-human with have ensure of or investing we accident purchasing very that we'll customers available need available. different prefer contact be That policy, interacting that a is certainly modifying options customers always a for
over we data reasons is our virtuous a the for of One referenced invest experiences years. context customers variety in digital a employees the and continue it accelerates of further in we've our cycle to
resolve work of phone variety claim digital claims, As across in to nature structured relatively parties. At industry, associated other predominantly turn generate Progressive, we experiences. processes customers each Those analog other many information. have their involved notes manual Historically, and and data. invested the interactions a unstructured with data to more employees enable calls, of our and in significantly with these unstructured lots us of those we've monitoring into and been
finite Additionally, They set the very to exclusive data events even also and digital nature using systems. those that's of interactions a insights provide of to how our or are creates experiential more not customers outcomes. employees
additional connecting investment to John's description competitive This data process technology claims how and for contributes of pricing. helps back identify claims us areas to
lower operational LAE, enabling enables helped investments. data additional growth and Our which more generating improved additional and efficiency lower even scale, rates, powering business targeted
has of when can day, to own a where within their a these a repair faster. or or they're whose customers of can to digital estimate a any for deductible the or been in a relative submit the From claimants experience back home if get digital photo they'll get and be workplace, action repair repaired. estimating. It photos the has not sure cycles Photo adoption uncertain via estimating they comfort often guided quickly, not as and one example vehicle they're when One is convenient cost grown significantly. experience
John shift walk a If of the you we bottom good shift, is through we to example curve. optimize estimating so we'll at such operate the as discussed, of And photo achieve a curves in even curve. continue can the when do we a the recall conceptual
time photo enjoyed an XX% The XXXX, estimates we've over of which left annual growth chart on rate. the shows
last primary While customer-submitted tens written photos. photo-based showing represents this on with over And X of millions the it, years. course, we're total, axis, that vertical millions of the of know numbers not estimates in actual of
secondary John and Estimate on earlier in work volume, of On view the no assessments are that Index for material of accuracy, the indexed been is vertical the These plotted degradation ground there's objective invaluable an of mentioned axis quality and the benchmark accuracy. growing truth. an orange source our a core Quality Despite line.
estimating alongside we're with generating while growth aren't volume and claims guiding pleased the in are inspection described a photo the the as efficient vehicle of compromising viable John growth we to the offering we're earlier. principles ensure channel, experience accuracy So we're and measuring outcomes consistently
In on vehicle completion. of it we right traditional inspection seen to the X.Xx number The the our we've each indexes daily most completed chart day, total, estimate estimates can complete compared efficient channel. in-person estimates making
we segmentation process identify processes, customer cost the the machine the like using alongside automate Once down the by or the we're part. parts we and left, masks process. the techniques to so damage apply photos structures, a curve. At scale, potentially high organizational achieve the we we're location additional viable us to alongside and of identify customer-submitted networks, vision a on parts correct further technology look through help learning of variety could case, refined photo our traditional using images. data that estimating actual machine and example learning level, other These at photos began more drive of technologies to then models In disclosing deep type illustrate neural models without this
predictions. requires discrete several estimate accurate an Generating
in know that on curated ensure consistent photos need From we headlines you differently do of hours but by type our it labor and many case et about, cetera. the we repair These our more of also there, "T" repair, standards. it For own accuracy take. second or the damage. to example, the are are ChatGPT, need quite recently. part, supervised just we literally we architectures data how believe Is the to part. whether dent, have These will heard but a location very progressive And of trained specific may we to in transformer models train with decide the not should models some damage the replace the making estimating models they than use scratch, of
willingness it's Such benefit to and differentiating have combination get feeling quality of a And work without us do people's assessments need It's look and the from objective own to large, data hard assurance disciplined that really to give sets. processes We're a that of clean, X fortunate from our very models make and that benefit believe, and all talented the the ground all data very data diverse my clean people again, difference. externally. to in-house it trustworthy we experience historical right set. technology and Progressive's the truth
process On solution, really advantage. our XXX shops estimates physical over with Most have helps of facets that organization. brings science written deliver combined technology and difficult expertise more us They this team that of to alongside our of works a advanced have have our competitive years and claims at them team worked level data themselves, daily experience. in every of repair. replicate deep alongside It's worked all understanding of vehicle that damage
the of in real a need this things the presence the original such is In behind it small and then in right. left and of identify we eye lower to bumper from labor on scenes for place, be These a the the between part solution things ensure bumper example things circular right It's enough the a and model. car. this as case, to a in and to the existence quick details accurate make to very of image bumper it, consider be bumper in a At time. bumper with be get does why indicate an get the one of As detect sees for ample even of the upper our distinguish bumper. the on year, you account sensors. outlines And cropping let's and not case, hard can to some estimate of the on the expertise whether relative that a the required particular prediction able the from example incremental the presence depending must matters, the of human making as in matter covers estimate damage to subject rendering and we a photo. shown the such But able of as to first sensors
a Of very technology, vision business winning of course, while in is machine the interesting data are competitions. we not science
value. value day, business The the represents index of orange of the line in it's inspections important the the traditional are per as in-person value so here insurance, same prove completed We we before.
the blue we You X.Xx then previously. and that can I estimating in see as photo introduced scaled into increase light original that mentioned get productivity line
effort the an this small journey began in current representatives. well estimating before hype XXXX, of as We cycle a R&D with group photo our
results. We and gains approach and our one efficiency until convinced models accuracy we that that refine continue to were iterate the we and better to two, time of such were able generate or equal
XXX machine expanding To our by achieved and shows XXXX. some XXXX, we productivity. XXXX, would Now estimating a estimates increase in began vision. having have them needed rolled an staff put fully in additional these capabilities In day Xx per in advances, throughout these enhanced we in the photo darker our line when organization without blue context, that incremental out
are of and has techniques have While XXX% a multiyear science our now doubled the then the professionals. I estimates inspection drafted by and been of it of now productivity combination data photo estimating corrected and vehicle our journey, mentioned, automation means validated solutions by what, our other most initially of and/or these was efficient as estimating. already
see automation the across you or processing may through that not To industry. full is be clear, straight referenced this
involved. and, of until regulations. can course, we it some automation principles We we're have think We is full applicable do consistent sure at point, with people possible do our that do but won't we guiding
For yet. our ceiling and now, to more us, we're more we see nothing a quite empowering even productive. people to happy hit gains to with even we've to And continue suggesting continue be
further just we're two our accuracy examples I'll incremental even already we've refine continuing cover what on think will solution to generate we in that beyond efficiency how achieved. gains and
very people's efficient. must training an set diverse with to scale, our model we and that make our a At our also keep moving. is always we efficiency, market more vehicles OEM Beyond work of up
set semi-supervised is model the label labeled as of of improve scaled cases But training car can very list the ever-evolving is This manually we with and use Traditional large approach rely on a value as our mentioned. do we called machine a data data. of of One vision set training of a to a practice the photos, we've speed to small at on parts. across solution, the we label deliver ways pseudo-labeling. a sought I that broader which of the rest
while there it certainly common vehicle most are around distinct by an varies external types. context, average For on and manufacturer, parts XXX
starts model process labels, of are and major much with base subject capable that's generated larger panels. parts initial set original segmentation model doors at From pseudo-labels on create parts. base data experts quarter this to matter as on major such an a trained even Our those identifying model our better
well improved the diverse of a model vehicle very new This or is emblems on segmentation that all vehicles. even depending general existence to parts vehicle, more a base set part portions even in specific that of location type. about model and allows fine-tuned This can like model details is in then vary works reasonably from grill This on detailed get template all labels shape, the more the specific part with vehicles. us
pseudo-labels we specific these examples vehicle data. apply But around detailed outcome, examples, hundreds a XXX using sample more to makes label customize for technique to randomly we bootstrap of vehicle With only further model and ensure vehicle-specific internal that to of models. the accurate thousands additional can most the
adapt allows shown right. efficiency, model computational images models tasks much this fine-tune of For on low-rank data point technique specific LoRA. called or training adaptation then training as LoRA is without The precision in with time. efficiently us to enhancing the needing or a we large to all the the
While molding, you suggest image which and is fender, at parts: front bumper, might capability. the the top the of fine-grained the more or across the right the we achieved wheel damage segmentation probably look front much part bumper right maybe the and three accurate view damage on extends fender, by is the details
enabled to has for us approach across applicability improvement extensive providing broad to achieve Our precise the while sets. XXx training label data all a both ensures speed in accurate recognize and types, It reducing more and large need them. vehicle not customization data external upon accurate of very to without set which techniques are extend a parts. and relevant trustworthy such possible that important It's identification labels
and millions give experts of of us matter photos both. subject Our available tens
is space. a fast-moving This very obviously
We are into those customers newer efficiency keep dynamic our Splatting. animation around Gaussian accuracy, a This eyes back photos. to the around of and we lot we to data per But Doing continue on. take them second XD frames, illustrates a ask video that our to it's X-dimensional of images at us to XX solution. customer ask a to A X-dimensional car transform But approach on in It hundreds allows reasonably addition X video we well requires experience, yields trade-offs XX to frames called part this of as us take increasingly to photos. and leveraging can't X-minute average. do photos X,XXX of models. of
neural models to vehicle language we in the large if graphics fields, Splatting damage. suggested such Being as reconstruction in of reconstruct our were reconstruction. person to have Gaussian operational a accurate. of full the field previous is extent needs. and to to are must able to radiance that efficient interesting and more and we work computationally The much be that as solution Remember, what efficient text, is Gaussian scientists XD but as what's understand Splatting meets it's builds us. research the on latter Some
this steps to here, primary walk won't we math the through While are X technique. there
us X-dimensional a of techniques again, the thousands of create essentially understand the vehicle. mean This to position position, every use cloud or data, position the construct center we to which of Gaussian grid motion the structure underlying the relative parallel a as camera or understanding point in pixel mesh understand from to algorithmic of frame. gives the Splats. In and First, such the the with coordinate camera locate upon point frames
cloud, splats get distribution render of Gaussians there, can details, as density capture On functions characteristics help we comprehensive results representing surface more and training a X-dimensional intricate of of providing Those vehicle a original X-dimensional those points, in geometry. of the us Gaussian until the data. traditional or physical graphic simulate each rasterization the adjusting space. of we using point we techniques optimize the surface. point the It representative onto Gaussian From the every normal overlay the reconstruction pixel vehicle and blobs to
original case Such What's by damaged accurate of will can relative where understanding depth conditioning provide vehicle could it how Vehicle it because to the because in more is for damage full visible enabled to more anticipating enable ultimately very understanding are and camera and and a take interactive part. enable of long vehicle. term, more fully loss to we been view perhaps produced the parts angles. and important LEAP XD damage inspections model we virtual the have you also think enable of interior fully labor views a from repair pictures the that detailed novel have measurement can approaches total and hour Longer angles it damage the to predictions in captured is accurate It alone data. distance additional decisions types. of the This allows aren't scale a depth. regions from by may surface external that not area relevant
our core employees deep Lastly, we consistent previously, and outcomes better And I business continuous mentioned achieve outcomes, dig for cool about own science our about do as hard to Progressive, Gaussian It's willing work improvement sake. to that investors. to are striving with values. at core our math its those customers, this John and value, isn't for and or for excellence want reiterate
we wanted consider when I our up, wrapping and to build buy share decisions. Before approach general
same First, we think coin. sides and competitive those of complementary not the are
for those partner but is things on tight solution platform estimatics photo the machine built, estimating turn numbers powered our have actual correct integrations and to example, estimates. also very For into predictions by part we with our vision relies
build to decoupled buying, difficult perhaps buy or be change of is systems building reassembled the that When so entirely one we new. be to modify and purposes, open than can keep some solutions, prefer but we are or other for for can good or or things prefer decoupled. Rather purpose, that to something monolithic pieces and parts
can returns of We Gaussian to but agile at earn also apply us pieces like We technology parts initial gives us allows can in new value use to on to investment not those the and start inevitably pseudo-labeling, This Splatting solutions something the provides decoupled adopting solutions the even when it flexibility whatever time. near-term considered new potential next. entirely that also drive and original arise. be be technology the build may to solution to we new outsized have may
example from as or firm we've we per first now right and potential vehicle notice vision XXXX, optimize. on inspection in as thousands notice loss, hundreds that iterated commitment an focused had that photos to submitted of on selected loss decided may the estimating. photos is before an that at customers When that to in But customer is is for photo accelerating as of of of a we with On estimating experience receiving option. were we're a have rate first funnel, This our photo claims year their that began machine emerging. opt back photo of estimating journey we after were
a damage photo estimating photo more photos and for triage We fine-grained our those through of that to on could more and review channel. machine not segmentation photo accurate vision us decision to decisions much make use even estimating. do estimates can based We processes estimating, initial also those sophisticated written apply the can triage more can and It enable estimate. we use weren't allow capabilities also in just existing that but
We conclude. today hope a we as came through things few
that people again and in XXXX with what to makes such remind help shared I'll our and our company-wide John are it starts things Claims possible. the enable results They efforts competitive culture. First, and pricing
we strategically because downward, Secondly, estimating doing see it to keep the putting turning. will value outcomes balance claims to opportunities principles Progressive while those we further ahead our perfect tirelessly people advanced curve in as cost further continue with our exploit we in And remain cycles hands the by and that continue generate to just they fully energized technology shift know that. virtuous guiding committed to and it shared and ever-elusive core of chase simultaneously example, and photo investing we aligned
you On behalf I and attention. teammates, and your for of thank XX,XXX-plus John Progressive
