数字が導く洞察：Webパフォーマンスとビジネス指標の相関を探る / Insights Driven by Numbers: Exploring the Correlation Between Web Performance and Business Metrics

*OTJHIUT%SJWFOCZ/VNCFST &YQMPSJOHUIF$PSSFMBUJPO #FUXFFO8FCQFSGPSNBODF BOE#VTJOFTT.FUSJDT 8FC1FSGPSNBODF5FBN 5FDIOPMPHZ&OIBODFNFOU %FQBSUNFOU -:$PSQPSBUJPO 5BLVZB5POBSV

1SPGJMF Name: Takuya Tonaru Departments: - Web performance team -
o11y trace team Histories: Previous: - LTE base station - Air Traffic Surveillance Current: - APM Platform - Alerting Platform

#BDLHSPVOEPG8FC1FSGPSNBODF

Core Web Vitals Google: 2017 （reference） Rakuten 24: 2022（reference） “Core
Web Vitals increased revenue per visitor by 53.37% and conversion rate by 33.13%”

LCP CLS INP Web Performance Metrics ＊

CVR Bounce Exit Business Metrics Buy

"HFOEB *OTJHIUTGSPNPVSBDUVBMEBUB 4ZTUFNGPS7JTVBMJ[BUJPOPG8FC 1FSGPSNBODFBOE#VTJOFTT.FUSJDT 5JQTGPS*NQSPWJOH8FC1FSGPSNBODF $PODMVTJPO/FYUDIBMMFOHF

*OTJHIUTGSPNPVSBDUVBMEBUB

1500ms 2000ms 3000ms LCP CVR • The distribution appears to
follow a log-normal pattern. • While 2.5s is the CWV threshold, aiming for 1.5s or faster can lead to stronger conversion impact. -$1 $PSSFMBUJPOXJUI$73 *OTJHIUTGSPN0VS"DUVBM%BUB 1000ms 2500ms LCP under 2.5s = good (Core Web Vitals)

-$1 $PSSFMBUJPOXJUI$73 *OTJHIUTGSPN0VS"DUVBM%BUB LCP CVR 2000ms 3000ms 1000ms • No
clear correlation with conversions • Pages that are not important in the user journey • The sample size is insufficient

-$1 $PSSFMBUJPOXJUI#PVODF&YJU *OTJHIUTGSPN0VS"DUVBM%BUB LCP Bounce/Exit 2000ms 3000ms 1000ms • Performance
may not correlate with conversions, but often does with bounce or exit rates.

$-4*/1 $PSSFMBUJPOXJUI$73 *OTJHIUTGSPN0VS"DUVBM%BUB CLS CVR INP CVR • CLS and
INP show weaker correlation compared to rendering metrics. • In our case, CLS is usually stable, and moderate INP improvement is enough—what matters more is rendering after user input. 1000ms 500ms 0.2 0.1

4ZTUFNGPS7JTVBMJ[BUJPOPG 8FC1FSGPSNBODFBOE#VTJOFTT .FUSJDT

*OTUBMMBHFOU XJUINJOJNBMDPEF OQN JOTUBMMSVNBHFOU JNQPSUSVNGSPNSVN BHFOU SVNDXW "OBMZUJDT%BTICPBSE 7JTVBMJ[FBOEBOBMZ[F
UIFDPSSFMBUJPO CFUXFFOXFC QFSGPSNBODFBOE CVTJOFTTNFUSJDT 4VQQPSU 8FCQFSGPSNBODF UFBNBTTJTUTXJUIUIF QSPWJTJPOPGBO BOBMZTJTFOWJSPONFOU BOEUFDIOJDBM DPOTVMUJOH $PNQBOZ8JEF&OBCMFNFOU4USVDUVSF 4ZTUFNGPS7JTVBMJ[BUJPOPG8FC1FSGPSNBODFBOE#VTJOFTT.FUSJDT

browser datalake Trino Tableau "OBMZUJDT"SDIJUFDUVSF 4ZTUFNGPS7JTVBMJ[BUJPOPG8FC1FSGPSNBODFBOE#VTJOFTT.FUSJDT field data

$PSSFMBUJPO 'JOEQBHFTXJUIIJHI DPSSFMBUJPO &MFNFOU4FMFDUPS *EFOUJGZFMFNFOUT XJUITJHOJGJDBOU JNQBDUPO JNQSPWFNFOU 7BMJEBUJPO 7BMJEBUFUIFSFTVMUT
PGJNQSPWFNFOUT SFDPNNFOE"# UFTU %BTICPBSE%FUBJMT 4ZTUFNGPS7JTVBMJ[BUJPOPG8FC1FSGPSNBODFBOE#VTJOFTT.FUSJDT

5JQTGPS*NQSPWJOH8FC1FSGPSNBODF

6OEFSTUBOEUIFDSJUJDBMSFOEFSJOHQBUI • -FWFSBHFCSPXTFSEFWFMPQFSUPPMT • "OBMZ[JOHPUIFSTFSWJDFTDBOQSPWJEFWBMVBCMFJOTJHIUTUPP "WPJECMPDLJOH)5.-QBSTJOH • %FGFSSFTPVSDFTOPUOFFEFEGPSJOJUJBMSFOEFSJOH • 6TFUFDIOJRVFTMJLFQSFMPBE
FBSMZIJOUT BOEBTZODEFGFS 'SPOUFOEJNQSPWFNFOU 5JQTGPS*NQSPWJOH8FC1FSGPSNBODF

• 3FNPWFVOVTFETUZMFTBOE +BWB4DSJQUUPSFEVDFMPBETJ[F • 4QMJU$44VTJOHUFDIOJRVFTMJLF$44 .PEVMFT • %FGFSOPODSJUJDBM+4UPTQFFEVQ JOJUJBMSFOEFSJOH "WPJE#MPDLJOH
5JQTGPS*NQSPWJOH8FC1FSGPSNBODF

1SJPSJUJ[FGBTUSFTQPOTFUJNFT • -FWFSBHF&EHF$%/JOGSBTUSVDUVSF • 'BTUJOGSBTUSVDUVSFJTLFZUPXFCQFSGPSNBODF JODMVEJOHUIFQSJWBUFDMPVE 5VOFTFSWFSTJEFQFSGPSNBODF • 'VOEBNFOUBMMZ SFEVDFDPNQVUBUJPOBMMPBE
FH • 6TF$16QSPpMJOHUPJEFOUJGZIFBWZDPNQVUBUJPOT • *GSFEVDUJPOJTOPUQPTTJCMF QBSBMMFMJ[FPSEFGFSUIFN • #BDLFOEQFSGPSNBODFJNQSPWFNFOUTBMTPSFEVDFDPNQVUJOHSFTPVSDFVTBHF #BDLFOEJNQSPWFNFOU 5JQTGPS*NQSPWJOH8FC1FSGPSNBODF

• 7JTVBMJ[FXIFSFBOEIPXNVDI$16 UJNFJTTQFOUJOUIFQSPHSBN • 1SPGJMJOHUIFXPSLMPBEPGUFOTIPXT UJNFCFJOHTQFOUJOVOFYQFDUFE QMBDFT $161SPGJMJOH 5JQTGPS*NQSPWJOH8FC1FSGPSNBODF

3FOEFSJOHTQFFETUJMMNBUUFSTNPTU • -$1"HPPETUBSUJOHQPJOU • 4QFFE*OEFY6TFGVMGPSGJOFUVOJOHSFOEFSJOHBSFBBOETQFFE /PQFSGPSNBODFXJUIPVUGBTUSFTQPOTF • 55'#"LFZCFODINBSLGPSCBDLFOESFTQPOTFTQFFE )FBMUIPGUIFDSJUJDBMSFOEFSJOHQBUI •
EPN$PNQMFUF"CFODINBSLGPSEPDVNFOUMPBEDPNQMFUJPO 5BSHFU.FUSJDT 5JQTGPS*NQSPWJOH8FC1FSGPSNBODF

$PODMVTJPOT/FYUDIBMMFOHFT

8FCQFSGPSNBODFJNQBDUTCVTJOFTTPVUDPNFT ,FFQJUFSBUJOHUISPVHINFBTVSFNFOU BOBMZTJT BOEPQUJNJ[BUJPO $PODMVTJPOT $PODMVTJPOT/FYUDIBMMFOHFT

&WPMVUJPOPGUIF"OBMZUJDT%BTICPBSET • &BTJMZDPNQBSFDPSSFMBUJPOTBDSPTTQBHFT • 1SFEJDUUIFJNQBDUPGJNQSPWFNFOUT • &BTJMZSVOTUBUJTUJDBMUFTUTPO"#UFTUSFTVMUT 1SBDUJDFJO0VS4FSWJDFT • 8FC1FSGPSNBODFBTB.BOBHFE2VBMJUZ*OEJDBUPS
/FYUDIBMMFOHFT $PODMVTJPOT/FYUDIBMMFOHFT

/FYUDIBMMFOHFT $PODMVTJPOT/FYUDIBMMFOHFT Field data Generative AI Distributed Traces AI-Based Performance
Optimization Proposals Source Code

5IBOLZPV

数字が導く洞察：Webパフォーマンスとビジネス指標の相関を探る / Insights Driv...

数字が導く洞察：Webパフォーマンスとビジネス指標の相関を探る / Insights Driven by Numbers: Exploring the Correlation Between Web Performance and Business Metrics

LINEヤフーTech (LY Corporation Tech) PRO

Transcript

*OTJHIUT%SJWFOCZ/VNCFST &YQMPSJOHUIF$PSSFMBUJPO #FUXFFO8FCQFSGPSNBODF BOE#VTJOFTT.FUSJDT 8FC1FSGPSNBODF5FBN 5FDIOPMPHZ&OIBODFNFOU %FQBSUNFOU -:$PSQPSBUJPO 5BLVZB5POBSV

1SPGJMF Name: Takuya Tonaru Departments: - Web performance team -

#BDLHSPVOEPG8FC1FSGPSNBODF

Core Web Vitals Google: 2017 （reference） Rakuten 24: 2022（reference） “Core

LCP CLS INP Web Performance Metrics ＊

CVR Bounce Exit Business Metrics Buy

"HFOEB OTJHIUTGSPNPVSBDUVBMEBUB 4ZTUFNGPS7JTVBMJ[BUJPOPG8FC 1FSGPSNBODFBOE#VTJOFTT.FUSJDT 5JQTGPSNQSPWJOH8FC1FSGPSNBODF $PODMVTJPO/FYUDIBMMFOHF

*OTJHIUTGSPNPVSBDUVBMEBUB

1500ms 2000ms 3000ms LCP CVR • The distribution appears to

-$1 $PSSFMBUJPOXJUI$73 *OTJHIUTGSPN0VS"DUVBM%BUB LCP CVR 2000ms 3000ms 1000ms • No

-$1 $PSSFMBUJPOXJUI#PVODF&YJU *OTJHIUTGSPN0VS"DUVBM%BUB LCP Bounce/Exit 2000ms 3000ms 1000ms • Performance

$-4/1 $PSSFMBUJPOXJUI$73 OTJHIUTGSPN0VS"DUVBM%BUB CLS CVR INP CVR • CLS and

4ZTUFNGPS7JTVBMJ[BUJPOPG 8FC1FSGPSNBODFBOE#VTJOFTT .FUSJDT

*OTUBMMBHFOU XJUINJOJNBMDPEF OQN JOTUBMMSVNBHFOU JNQPSUSVNGSPNSVN BHFOU SVNDXW "OBMZUJDT%BTICPBSE 7JTVBMJ[FBOEBOBMZ[F

browser datalake Trino Tableau "OBMZUJDT"SDIJUFDUVSF 4ZTUFNGPS7JTVBMJ[BUJPOPG8FC1FSGPSNBODFBOE#VTJOFTT.FUSJDT field data

$PSSFMBUJPO 'JOEQBHFTXJUIIJHI DPSSFMBUJPO &MFNFOU4FMFDUPS *EFOUJGZFMFNFOUT XJUITJHOJGJDBOU JNQBDUPO JNQSPWFNFOU 7BMJEBUJPO 7BMJEBUFUIFSFTVMUT

5JQTGPS*NQSPWJOH8FC1FSGPSNBODF

6OEFSTUBOEUIFDSJUJDBMSFOEFSJOHQBUI • -FWFSBHFCSPXTFSEFWFMPQFSUPPMT • "OBMZ[JOHPUIFSTFSWJDFTDBOQSPWJEFWBMVBCMFJOTJHIUTUPP "WPJECMPDLJOH)5.-QBSTJOH • %FGFSSFTPVSDFTOPUOFFEFEGPSJOJUJBMSFOEFSJOH • 6TFUFDIOJRVFTMJLFQSFMPBE

• 3FNPWFVOVTFETUZMFTBOE +BWB4DSJQUUPSFEVDFMPBETJ[F • 4QMJU$44VTJOHUFDIOJRVFTMJLF$44 .PEVMFT • %FGFSOPODSJUJDBM+4UPTQFFEVQ JOJUJBMSFOEFSJOH "WPJE#MPDLJOH

1SJPSJUJ[FGBTUSFTQPOTFUJNFT • -FWFSBHF&EHF$%/JOGSBTUSVDUVSF • 'BTUJOGSBTUSVDUVSFJTLFZUPXFCQFSGPSNBODF JODMVEJOHUIFQSJWBUFDMPVE 5VOFTFSWFSTJEFQFSGPSNBODF • 'VOEBNFOUBMMZ SFEVDFDPNQVUBUJPOBMMPBE

• 7JTVBMJ[FXIFSFBOEIPXNVDI$16 UJNFJTTQFOUJOUIFQSPHSBN • 1SPGJMJOHUIFXPSLMPBEPGUFOTIPXT UJNFCFJOHTQFOUJOVOFYQFDUFE QMBDFT $161SPGJMJOH 5JQTGPS*NQSPWJOH8FC1FSGPSNBODF

3FOEFSJOHTQFFETUJMMNBUUFSTNPTU • -$1"HPPETUBSUJOHQPJOU • 4QFFE*OEFY6TFGVMGPSGJOFUVOJOHSFOEFSJOHBSFBBOETQFFE /PQFSGPSNBODFXJUIPVUGBTUSFTQPOTF • 55'#"LFZCFODINBSLGPSCBDLFOESFTQPOTFTQFFE )FBMUIPGUIFDSJUJDBMSFOEFSJOHQBUI •

$PODMVTJPOT/FYUDIBMMFOHFT

8FCQFSGPSNBODFJNQBDUTCVTJOFTTPVUDPNFT ,FFQJUFSBUJOHUISPVHINFBTVSFNFOU BOBMZTJT BOEPQUJNJ[BUJPO $PODMVTJPOT $PODMVTJPOT/FYUDIBMMFOHFT

&WPMVUJPOPGUIF"OBMZUJDT%BTICPBSET • &BTJMZDPNQBSFDPSSFMBUJPOTBDSPTTQBHFT • 1SFEJDUUIFJNQBDUPGJNQSPWFNFOUT • &BTJMZSVOTUBUJTUJDBMUFTUTPO"#UFTUSFTVMUT 1SBDUJDFJO0VS4FSWJDFT • 8FC1FSGPSNBODFBTB.BOBHFE2VBMJUZ*OEJDBUPS

/FYUDIBMMFOHFT $PODMVTJPOT/FYUDIBMMFOHFT Field data Generative AI Distributed Traces AI-Based Performance

5IBOLZPV

More Decks by LINEヤフーTech (LY Corporation Tech)

Other Decks in Technology

Featured