Relevance in Equivi Zm Predictive Cding Technlgy fr Assessment f Dcument Relevance
THE PROBLEM: REDUCING REVIEW COSTS WHILE ENHANCING QUALITY E-discvery is all abut finding relevant dcuments. Legacy prcesses slely dependent n human culling and review are n lnger suitable fr handling the huge amunts f electrnic data. New technlgies are required t minimize the cst f human review, while ensuring that the vast majrity f relevant dcuments are, in fact, discvered. The traditinal culling methd is based n a list f keywrds cmpiled by an attrney, that when fed int a search engine, will retrieve relevant dcuments. Hwever, a series f independent studies has shwn that keywrds typically find nly 20-30% f relevant dcuments. This means that 70-80% f relevant dcuments never see the light f day. Mrever, under the standard keywrds methd, dcument scring is binary i.e., the dcument is either In r Out. The absence f a graduated scale means that it is impssible t priritize review, while rendering the culling prcess clumsy and rigid; fr example, when trying t align the target review set with changes in budgets, strategy, r the size f the ingest cllectin. THE SOLUTION: SYSTEMATIC ORGANIZATION OF DOCUMENT COLLECTIONS BY RELEVANCE Fully integrated within Equivi Zm, Relevance is a predictive cding applicatin that takes the guesswrk ut f the e-discvery prcess. Relevance rganizes a dcument cllectin by respnsiveness, enabling litigatrs t fcus their effrts n the mst imprtant dcuments. As an expert-guided system, Relevance wrks like this: An expert/attrney reviews a sample f dcuments, ranking them as relevant r nt. Based n the results, Equivi learns hw t scre dcuments fr relevance. In an iterative, self-crrecting prcess, Relevance feeds additinal samples t the expert. These statistically generated samples allw Relevance t prgressively imprve the accuracy f its relevance scring. Once a threshld level f accuracy is achieved, "training" is cmplete. Relevance then ranks the entire cllectin autmatically, calculating a graduated relevance scre fr each dcument. 2
The pririty rankings generated by Relevance can be cntrasted with the binary scring f dcuments using manual keywrds. The graduated rankings under Relevance prvide much mre flexibility and cntrl in the culling and review prcesses. 3
SETTING THE STANDARD FOR SMART DISCOVERY Thrugh the use f predictive cding t assess dcument relevance quickly and accurately, Relevance has intrduced a fundamental imprvement in the way that litigatrs perfrm ECA and dcument review: Prven: Relevance is prven in thusands f cases, including engagements with the US Department f Justice. Its effectiveness in retrieving relevant dcuments has als been validated in the TREC prject. Relevance is currently deplyed at several leading law and cnsulting firms such as Baker & McKenzie, Sidley Austin, Squire Sanders and KPMG. Defensible: Relevance is built fr defensibility. The applicatin is built arund a a sund, scientifically valid statistical mdel. The statistical mdel is respnsible fr mnitring the sampling and training prcess, quantifying results and enabling systematic quality assurance and testing. Benchmarks cnducted by Equivi custmers shw that Relevance, used prperly, is as accurate and in sme cases even mre accurate than human review. Defensibility is a guiding principle in all facets f Relevance, including sampling, training, statistics and quality assurance. Setting the standard: As part f the integrated Zm platfrm, Relevance features a decisin supprt envirnment fr review set cnstructin. Mrever, using active learning techniques, Relevance achieves ptimal training efficiency, requiring manual review f nly 1,000-2,000 dcuments befre the training prcess is cmplete. In additin, Relevance is designed t ptimize usability fr supprt f multi-issue training, graduated scres, incremental lads and cllabrative training. KEY BENEFITS Relevance zms in n the mst relevant data t reduce risk and lwer csts: Reduces risk: In early case assessment, allws team t fcus n mst relevant dcuments t make infrmed decisins n case strategy and winnability Discharges discvery bligatins by finding mre f the relevant data Priritizes review effrts by fcusing n the key data Enables systematic QA f review by crss-matching human and sftware tags Lwers cst: Filters nn-relevant dcuments t reduce the review burden 4
Enables stratified review strategies, where high-ptential dcuments are assigned fr review by in-huse attrneys, and lw-ptential dcuments are assigned fr lw-cst cntract review Facilitates prprtinality cnsideratins in determining the size f the review set Eliminates the need fr first pass review in many cases DRIVING VALUE ACROSS THE FULL E- DISCOVERY CYCLE By identifying the mst relevant dcuments, Zm's Relevance applicatin enables infrmed early case assessment, precise data culling, priritized review and systematic QA fr human review. Early case assessment Allws litigatrs t zm in and review the mst imprtant dcuments in the case. This enables rapid, yet infrmed assessment f case winnability. Identifies the vlume f relevant dcuments in the cllectin, enabling a quick assessment f the ptential cst f the review effrt. Enables litigatrs t make well-funded decisins n whether t pursue a fight r flee strategy. Smarter culling Uses statistical tls that vercme the prblems f ver-inclusin (precisin) and under-inclusin (recall) that affect keywrd searching Cnsistently retrieves 80-95% f relevant dcuments in a cllectin, as ppsed t 20-30% recall in typical keywrd search. This means that Relevance finds 3 t 4 times mre relevant dcuments than legacy search techniques. Reduces time wasted reviewing nn-relevant dcuments t a bare minimum. In keywrd search, the vast majrity f dcuments (70-80%) assigned fr review are nt relevant. Relevance dramatically reduces this wasted effrt. Litigatin review Ranking a dcument cllectin by relevance enables a variety f efficient review ptins: Priritized review - start with mst relevant dcuments and wrk back t accelerate case develpment and reduce the risk f missing key material 5
Stratified review - high scring, high ptential dcuments are assigned t high-grade, high cst, in-huse review, while lw scring, lw ptential dcuments are assigned t lw-cst, ffshre and cntract reviewers Privileged review - Zm uses Relevance t identify dcuments with privileged cntent, as well as analyzing metadata t identify dcuments frm privileged cmmunicatin, fr chesive privileged review. Quality Assurance Matches Relevance respnsiveness scres against human review team designatins Enables systematic QA by fcusing n dcuments where there is a high chance f errr - i.e., dcuments where humans and the sftware did nt agree ABOUT EQUIVIO ZOOM Equivi Zm is an integrated platfrm fr e-discvery analytics and predictive cding. Zm brings tgether Equivi's prven technlgies fr e-discvery analytics in a unified web-based platfrm. Cmbining Equivi's best-f-breed near-duplicates, email threads and relevance cmpnents tgether with data imprt, ECA and enriched analytics capabilities, Zm prvides the tls yu need fr easier and smarter e-discvery. ABOUT EQUIVIO Equivi develps text analysis sftware fr e-discvery. Users include the DJ, the FTC, KPMG, Delitte, plus hundreds f law firms and crpratins. Equivi ffers Zm, an integrated web platfrm fr analytics and predictive cding. Zm rganizes cllectins f dcuments in meaningful ways. S yu can zm right in and find ut what s interesting, ntable and unique. Request a dem at inf@equivi.cm r visit us at www.equivi.cm. Zm in. Find ut. Equivi, Equivi Zm, Equivi>NearDuplicates, Equivi>EmailThreads, Equivi>Cmpare, Equivi>Relevance are trademarks f Equivi. Other prduct names mentined in this dcument may be trademarks r registered trademarks f their respective wners. All specificatins in this dcument are subject t change withut prir ntice. Cpyright 2012 Equivi 6