{"id":8688,"date":"2026-06-30T15:51:07","date_gmt":"2026-06-30T13:51:07","guid":{"rendered":"https:\/\/igloonet.cz\/blog\/?p=8688"},"modified":"2026-07-01T07:29:16","modified_gmt":"2026-07-01T05:29:16","slug":"agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor","status":"publish","type":"post","link":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/","title":{"rendered":"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor"},"content":{"rendered":"
<\/div>\n

Agent v datech? Co ud\u011blat pro to, aby se v nich neztratil a nep\u00e1lil zbyte\u010dn\u011b jak moc sv\u00fdch token\u016f, tak computing datov\u00e9ho skladu? Tedy: aby n\u00e1s nebolel \u00fa\u010det za n\u00e1\u0161 BigQuery \u00fa\u010det?<\/p>\n

Na posledn\u00edm tu\u010d\u0148\u00e1\u010d\u00edm webin\u00e1\u0159i<\/i><\/a> jsme prob\u00edrali jak doporu\u010den\u00ed<\/i> Anthropicu pro pr\u00e1ci s agentickou analytikou<\/i><\/a>, tak na\u0161e lok\u00e1ln\u00ed zku\u0161enosti, kter\u00e9 jsou vzhledem k mo\u017enostem m\u00edstn\u00edm firm\u00e1m bli\u017e\u0161\u00ed. Agentick\u00e1 analytika m\u00e1 dnes dv\u011b nejd\u016fle\u017eit\u011bj\u0161\u00ed \u010d\u00e1sti: konverza\u010dn\u00ed analytiku a vlastn\u00ed tvorbu p\u0159ehled\u016f, co\u017e beru jako nadstavbu nad konverza\u010dn\u00ed analytikou.<\/p>\n

<\/p>\n

Konverza\u010dn\u00ed analytika umo\u017e\u0148uje pt\u00e1t se na data b\u011b\u017enou \u0159e\u010d\u00ed, tedy bez pou\u017e\u00edv\u00e1n\u00ed SQL a dost\u00e1vat odpov\u011bdi op\u011bt v b\u011b\u017en\u00e9 \u0159e\u010di.<\/i><\/p><\/blockquote>\n

V teorii to zn\u00ed skv\u011ble, ale jak p\u00ed\u0161e s\u00e1m Anthropic \u201cjen napojit Claude na datov\u00fd sklad m\u016f\u017ee vytv\u00e1\u0159et fale\u0161n\u00fd pocit p\u0159esnosti\u201d. V moment\u011b, kdy to \u0159\u00edkaj\u00ed sami tv\u016frci Claude, bych to nebral na lehkou v\u00e1hu.<\/p>\n

Celkov\u011b je toti\u017e pot\u0159eba vy\u0159e\u0161it t\u0159i probl\u00e9my<\/h2>\n

1) Nejednozna\u010dnost metrik<\/b>: u\u017eivatel chce zjistit, jak\u00e9 byly tr\u017eby za minul\u00fd m\u011bs\u00edc, ale v datov\u00e9m skladu jsou jak tr\u017eby namodelovan\u00e9 pro mana\u017eersk\u00fd report, tak pro finan\u010dn\u00ed. Nav\u00edc nen\u00ed jasn\u00e9, jestli bude u\u017eivatel cht\u00edt tr\u017eby s DPH, bez DPH, p\u0159\u00edpadn\u011b jak pracovat s vratkami.<\/p>\n

2) Zastar\u00e1v\u00e1n\u00ed dat<\/b>: je super, kdy\u017e postav\u00edme spr\u00e1vn\u00e9 datov\u00e9 z\u00e1klady a v\u0161echno vylad\u00edme, ale firma \u017eije d\u00e1l, a proto je pot\u0159eba jak assety, tak agentn\u00ed \u201eznalosti\u201c st\u00e1le aktualizovat.<\/p>\n

3) Nedohled\u00e1n\u00ed informace<\/b>: opak k nejednozna\u010dnosti metrik, v tomto p\u0159\u00edpad\u011b nen\u00ed probl\u00e9m v tom, \u017ee by agent neum\u011bl vybrat mezi v\u00edce r\u016fzn\u00fdmi metrikami, ale v tom, \u017ee nen\u00ed schopen informaci v datech v\u016fbec naj\u00edt.<\/p>\n

Jak ty probl\u00e9my \u0159e\u0161it<\/h2>\n

Samotn\u00fd datov\u00fd sklad<\/h3>\n

Dobr\u00e1 zpr\u00e1va je, \u017ee t\u00e9m\u011b\u0159 v\u0161echny body byly best practices pro budov\u00e1n\u00ed datov\u00e9ho skladu u\u017e dlouho p\u0159ed p\u0159\u00edchodem analytick\u00fdch agent\u016f. Koncept data lake \/ data swamp na\u0161t\u011bst\u00ed po z\u00e1sluze bloud\u00ed po skladi\u0161ti slep\u00fdch uli\u010dek a vr\u00e1tili jsme se k po\u0159\u00e1dn\u011b popsan\u00fdm a dostate\u010dn\u011b srozumiteln\u00fdm datov\u00fdm sklad\u016fm, ve kter\u00fdch se l\u00e9pe vyznaj\u00ed jak lid\u00e9, tak agenti.<\/p>\n

Data pot\u0159ebujeme namodelovat nejen efektivn\u011b, abychom minimalizovali budouc\u00ed n\u00e1klady spojen\u00e9 se \u010dten\u00edm t\u011bchto dat, ale hlavn\u011b p\u0159ehledn\u011b. C\u00edlem jsou \u010dist\u00e9 L3, ve kter\u00fdch se vyzn\u00e1 jak analytik, tak agent.<\/p>\n

V r\u00e1mci datov\u00e9ho skladu tak\u00e9 \u0159e\u0161\u00edme role, maskov\u00e1n\u00ed dat a nastavov\u00e1n\u00ed limit\u016f. Tedy omezen\u00ed dat, ke kter\u00fdm by agent m\u011bl a nem\u011bl m\u00edt p\u0159\u00edstup, a\u0165 u\u017e z hlediska leg\u00e1lnosti (osobn\u00ed identifika\u010dn\u00ed \u00fadaje), proto\u017ee n\u00e1m p\u0159ijdou p\u0159\u00edli\u0161 citliv\u00e9, nebo proto, \u017ee je\u0161t\u011b nejsou p\u0159ipraven\u00e1 a nechceme, aby je zbyte\u010dn\u011b prol\u00e9zal a p\u00e1lil na nich nejen tokeny. \u010c\u00edm men\u0161\u00ed hrac\u00ed pole pro agenta vyty\u010d\u00edme, t\u00edm lep\u0161\u00ed a levn\u011bj\u0161\u00ed odpov\u011bdi budeme dost\u00e1vat.<\/p>\n

Datov\u00e9 sklady t\u00e9m\u011b\u0159 v\u017edycky stav\u00edme v BigQuery. D\u016fvod je jednoduch\u00fd: jedin\u00e9 p\u0159\u00edm\u00e9 napojen\u00ed na Google Analytics. Pokud firma u\u017e n\u011bjak\u00fd datov\u00fd sklad m\u00e1 jinde (Microsoft \/ Amazon stack, tedy Fabric, nebo Redshift), rozv\u00edj\u00edme ten existuj\u00edc\u00ed, ale minim\u00e1ln\u011b streamu z BigQuery se nevyhneme, pokud nechceme jen omezen\u00e1 data p\u0159es GA API.<\/p>\n

Na modelov\u00e1n\u00ed dat je pak industry standard dbt. V moment\u011b, kdy pracujeme s BigQuery, tak v\u011bt\u0161inou pou\u017e\u00edv\u00e1me Dataform, kter\u00fd je sou\u010d\u00e1st\u00ed Google Cloud Platform. Ka\u017edop\u00e1dn\u011b n\u00e1m p\u0159ijde dobr\u00e9 stav\u011bt v\u0161e modul\u00e1rn\u011b, abychom kdykoliv mohli \u010d\u00e1st infrastruktury vym\u011bnit.<\/p>\n

\"modelov\u00e1n\u00ed<\/a><\/p>\n

S\u00e9mantick\u00e1 vrstva<\/h3>\n

Jeden z m\u00e1la overhyped p\u0159\u00edstup\u016f k pr\u00e1ci s daty z posledn\u00edch let, kter\u00fd je v agentn\u00ed dob\u011b st\u00e1le aktu\u00e1ln\u00ed. S\u00e9mantick\u00e1 vrstva je ve sv\u00e9 podstat\u011b hrozn\u011b jednoduch\u00fd koncept. V podstat\u011b se jedn\u00e1 o v\u00fdkladov\u00fd slovn\u00edk (pozor nepl\u00e9st s data dictionary, kter\u00fd m\u00e1 se s\u00e9mantickou vrstvou velk\u00fd p\u0159ekryv, ale p\u0159eci jen je n\u011bco trochu jin\u00e9ho). Na jednom m\u00edst\u011b jsou namapovan\u00e9 analyticky pojmenovan\u00e9 metriky na re\u00e1ln\u00e9 byznysov\u00e9 metriky, vysv\u011btlen\u00fd jejich v\u00fdpo\u010det, definovan\u00fd povolen\u00fd rozpad (m\u016f\u017eeme marketingov\u00e9 n\u00e1klady rozpadat pouze na zdroj \/ kampa\u0148, nebo je m\u00e1me namodelovan\u00e9 i na \u00farove\u0148 produktu \u010di z\u00e1kazn\u00edka) a tak\u00e9 informace o tom, s \u010d\u00edm je m\u016f\u017eeme spojovat, aby to v na\u0161em modelu st\u00e1le d\u00e1valo smysl.<\/p>\n

 <\/p>\n

S\u00e9mantickou vrstvu v\u011bt\u0161inou zpracov\u00e1v\u00e1me v cube.js<\/a>. I tady op\u011bt plat\u00ed, \u017ee se sna\u017e\u00edme preferovat open source projekty, a hlavn\u00ed za n\u00e1s je, aby bylo mo\u017en\u00e9 jednotliv\u00e9 komponenty modul\u00e1rn\u011b vym\u011bnit. Vyhnout se nap\u0159\u00edklad tomu, abychom byli zam\u010den\u00ed ve vizualiza\u010dn\u00edch n\u00e1stroj\u00edch, kter\u00e9 se \u010dasto sna\u017e\u00ed, aby si firmy budovaly s\u00e9mantickou vrstvu p\u0159\u00edmo v jejich prost\u0159ed\u00ed.<\/p>\n

Datov\u00fd katalog<\/h3>\n

T\u00edm ukon\u010dujeme svatou trojici p\u0159\u00edpravy datov\u00e9ho skladu na agentickou analytiku. Dobr\u00fd datov\u00fd sklad by m\u011bl minim\u00e1ln\u011b obsahovat informace o tom, kdo je za metriku zodpov\u011bdn\u00fd, jestli obsahuje osobn\u00ed identifika\u010dn\u00ed \u00fadaje, jak \u010dasto se aktualizuje a jak vypad\u00e1 jej\u00ed lineage (odkud se bere, jak se po\u010d\u00edt\u00e1). M\u016f\u017ee to vypadat, \u017ee datov\u00fd sklad je z t\u00e9 trojice nejm\u00e9n\u011b d\u016fle\u017eit\u00fd, ale i tak je ned\u00edlnou sou\u010d\u00e1st\u00ed.<\/p>\n

V jednom z nejv\u011bt\u0161\u00edch e-commerce projekt\u016f v \u010cesk\u00e9 republice te\u010f testuj\u00ed agentickou analytiku bez s\u00e9mantick\u00e9 vrstvy a datov\u00e9ho katalogu (\u010dist\u011b s velmi solidn\u011b p\u0159ipraven\u00fdm datov\u00fdm skladem). Na dotaz, jak jsou spokojen\u00ed s procentu\u00e1ln\u00ed kvalitou odpov\u011bd\u00ed, sd\u011blili, \u017ee p\u0159i t\u011bch p\u00e1r testovac\u00edch pokusech super. Jen p\u0159i jednom dotazu agent nezjistil, \u017ee d\u00e1v\u00e1 data za celou skupinu, a ne jen za \u010deskou \u010d\u00e1st. Co\u017e je na\u0161t\u011bst\u00ed dostate\u010dn\u011b velk\u00e1 chyba, aby si j\u00ed v\u0161iml t\u00e9m\u011b\u0159 ka\u017ed\u00fd. Tak jako tak jde p\u0159esn\u011b o ten typ chyby, kter\u00fd by m\u011bla lineage odstranit.<\/p>\n

Na datov\u00fd katalog pou\u017e\u00edv\u00e1me Knowledge catalog v r\u00e1mci GCP, p\u0159\u00edpadn\u011b z open source \u0159e\u0161en\u00ed DataHub. Mo\u017enost\u00ed je ale daleko v\u00edc.<\/p>\n

\"DataHub\"<\/a><\/p>\n

A co je c\u00edlem? Certifikovan\u00fd datov\u00fd produkt. Popsan\u00e1 metrika, jak se po\u010d\u00edt\u00e1, jak jde rozpadnout, s jak\u00fdmi dimenzemi zobrazovat, z\u00e1kladn\u00ed pravidla a business kontext, informace o tom, kdo za ni zodpov\u00edd\u00e1, jak \u010dasto se aktualizuje, zda obsahuje osobn\u00ed identifika\u010dn\u00ed \u00fadaje, jak vypad\u00e1 lineage a kde se vlastn\u011b nach\u00e1z\u00ed.<\/p>\n

\"Certifikovan\u00fd<\/a><\/p>\n

U spousty v\u011bc\u00ed sv\u00e1d\u00ed nechat si je rovnou vygenerovat Claude Code, tady ale Anthropic celkem d\u016frazn\u011b varuje, \u017ee to tak\u00e9 zkou\u0161eli a v\u00fdsledky byly dost \u0161patn\u00e9. Minim\u00e1ln\u011b zat\u00edm to vypad\u00e1, \u017ee ide\u00e1ln\u00ed postup je nechat si lidmi (klidn\u011b ve spolupr\u00e1ci s AI) p\u0159ipravit datov\u00e9 z\u00e1klady a teprve potom pozvat na upraven\u00e9 h\u0159i\u0161t\u011b agenty.<\/p>\n

A co skills?<\/h2>\n

Toto je up\u0159\u00edmn\u011b \u010d\u00e1st, kter\u00e9 jsme nep\u0159ikl\u00e1dali takovou d\u016fle\u017eitost. Minim\u00e1ln\u011b ne tak velkou, jak nazna\u010duje Anthropic. Ten tvrd\u00ed, \u017ee p\u0159ed zapojen\u00edm skills do agentn\u00ed analytiky nebyli schopni se dostat p\u0159es 21% sk\u00f3re na sv\u00fdch evals (klasifik\u00e1torech agentn\u00ed odpov\u011bdi). S postupn\u00fdm zaveden\u00edm skills se ale dostali a\u017e na 95 %. To je v\u0161ak minim\u00e1ln\u011b trochu zav\u00e1d\u011bj\u00edc\u00ed. Anthropic ve sv\u00fdch evals m\u00e1 jako jeden z hlavn\u00edch bod\u016f, zda se agent jako prvn\u00ed dopt\u00e1v\u00e1 s\u00e9mantick\u00e9 vrstvy a a\u017e pot\u00e9 jde p\u0159\u00edpadn\u011b jinam. Pr\u00e1v\u011b to je jedno z chov\u00e1n\u00ed, kter\u00e9 se bez skills neda\u0159ilo konzistentn\u011b vynucovat.<\/p>\n

Samoz\u0159ejm\u011b, \u017ee ka\u017ed\u00fd, kdo si s Claude Code hr\u00e1l d\u00e9le ne\u017e jedno odpoledne v\u00ed, \u017ee skilly nejsou samosp\u00e1sn\u00e9. Ob\u010das se agent rozhodne jinak bez ohledu na to, jak jasn\u011b m\u00e1 popsan\u00fd doporu\u010den\u00fd postup.<\/p>\n

Pozor, bez udr\u017eov\u00e1n\u00ed skill\u016f up to day u\u017e za m\u011bs\u00edc poklesla p\u0159esnost u Anthropicu z 95 % na 65 %.<\/p>\n

---\nname: [warehouse-skill]\nversion: [x.y.z]\ndescription: \"IF the user asks to query [the company]'s data warehouse for any\n  [list of business domains] question \u2014 THEN invoke this skill. DO NOT invoke\n  for [adjacent engineering tasks] or questions with no data-warehouse component.\"\n---\n\n# [Warehouse] Skill Instructions\n\n## Description\nThe single source of truth for safe and effective [warehouse] querying.\nReferenced by other skills [listed] for query execution guidance.\n\nAct as a Data Analyst, providing strategic insights and data-driven\nrecommendations but seek guidance along the way.\n\n**Out-of-scope decisions**: [product areas, etc.] \u2192 surface data only,\nstate \"decision is [owning team]'s call\", do NOT take a position or author\ncode fixes.\n\n## Executing queries\nPriority:\n1. **[Managed connection]** (if available): [query tool] \/ [schema tool]\n2. **[CLI fallback]** (if installed): [default project, fallback project]\n3. **Neither** \u2014 ask the user to authenticate, then stop\n\n# Semantic Layer (REQUIRED first step)\n\nThe governed semantic layer is the **mandatory default path** for every data\nquestion \u2014 same numbers as [the BI tool], joins\/grain\/filters baked in. Raw SQL\nvia the reference docs below is the **fallback**, used only after the\nsemantic-layer path is shown not to cover the ask.\n\n## Required workflow\n1. **Load** \u2014 [how to load the semantic layer in each runtime, with fallbacks]\n2. **Discover** \u2014 search measures\/dimensions by keyword; **always check\n   segments** (the named canonical population filters \u2014 hand-rolled WHERE\n   clauses for these are the dominant wrong-answer mode)\n3. **Compile + run** \u2014 build the spec \u2192 compile to SQL \u2192 execute\n4. **Fallback** \u2014 only if discovery finds no relevant metric or compile fails\n   \u2192 raw SQL via `references\/*.md` (PART 3 below)\n\n> **Don't bail early.** Do NOT fall back to raw SQL on these grounds:\n> - \"[custom date filtering \/ cohorts]\" \u2192 [covered by time-dimension specs]\n> - \"[needs a join]\" \u2192 [the metric layer already encapsulates its joins]\n> - [3\u20134 more pre-rebutted excuses agents use to skip the semantic layer]\n\n### Date windows & timezone \u2014 decide before you query\n- **As-of date vs trailing-N days**: [convention for each]\n- **\"Last week\/month\"** \u2192 the last *complete* calendar week\/month, not trailing-7\/30\n- **Timezone default**: [TZ]; [exception for certain reporting rollups]\n- **Freshness lag**: [some] tables settle late \u2014 anchor on MAX(date), not \"yesterday\"\n\n<\/pre>\n
Uk\u00e1zka skillu<\/em><\/p>\n
Co je \u00fapln\u011b k ni\u010demu<\/h2>\n
Jedna z best practices p\u0159\u00edpravy na agentickou analytiku je query corpus. Uk\u00e1zky spr\u00e1vn\u00fdch SQL s v\u00fdsledky, notebooky, historick\u00e9 anal\u00fdzy. Tady jsme my osobn\u011b \u017e\u00e1dn\u00e9 testov\u00e1n\u00ed vlivu query corpusu na spr\u00e1vnost v\u00fdsledk\u016f ned\u011blali, ka\u017edop\u00e1dn\u011b Anthropic toto testoval \u00fadajn\u011b dostate\u010dn\u011b d\u016fkladn\u011b a zjistil, \u017ee na spr\u00e1vnost odpov\u011bdi to nem\u00e1 v podstat\u011b \u017e\u00e1dn\u00fd vliv. Agent zkr\u00e1tka v\u011bt\u0161inou nebyl schopn\u00fd nap\u00e1rovat nov\u00fd dotaz na ji\u017e existuj\u00edc\u00ed query \u010di anal\u00fdzu. Nav\u00edc analytick\u00e9 best practices jsou sou\u010d\u00e1st\u00ed skill\u016f a jako takov\u00e9 jsou pro agenta jednodu\u0161eji uchopiteln\u00e9 ne\u017e destilace spr\u00e1vn\u00e9ho postupu z query corpusu.<\/p>\n
95 % a nen\u00ed to m\u00e1lo, Antone Pavlovi\u010di?<\/h2>\n
P\u0159esnost, na kterou se Anthropic dostal, zn\u00ed dost skv\u011ble. Je pot\u0159eba si ale \u0159\u00edct, \u017ee to je:<\/p>\n
1) opravdu po\u0159\u00e1dn\u011b p\u0159ipraven\u00fd datov\u00fd sklad, s\u00e9mantick\u00e1 vrstva, datov\u00fd katalog i vypimpen\u00e9 skills,
\n2) Anthropic nejsp\u00ed\u0161 \u00fapln\u011b ne\u0161et\u0159\u00ed na intern\u00edm pou\u017e\u00edv\u00e1n\u00ed Claude, jeho limitech a modelu - situace, kter\u00e1 nejsp\u00ed\u0161 nebude ve v\u0161ech firm\u00e1ch stejn\u00e1,
\n3) i tak je ka\u017ed\u00e1 dvac\u00e1t\u00e1 odpov\u011b\u010f \u0161patn\u00e1, zav\u00e1d\u011bj\u00edc\u00ed.<\/p>\n
Anthropic si nejsp\u00ed\u0161 s\u00e1m tyto limitace uv\u011bdomuje, proto uv\u00e1d\u00ed, \u017ee agentickou analytiku pou\u017e\u00edvaj\u00ed na 95 % business dotaz\u016f. Na ty strategicky nejd\u016fle\u017eit\u011bj\u0161\u00ed ot\u00e1zky tedy nejsp\u00ed\u0161 agenty nepou\u017e\u00edvaj\u00ed. To je podle m\u011b spr\u00e1vn\u00fd p\u0159\u00edstup. Tam, kde to pro m\u011b nen\u00ed naprosto kritick\u00e9, m\u016f\u017eu agenta pou\u017e\u00edt a na odpov\u011b\u010f se do n\u011bjak\u00e9 m\u00edry spolehnout. Na ty z\u00e1sadn\u00ed dotazy v\u0161ak pot\u0159ebuji minim\u00e1ln\u011b n\u00e1slednou kontrolu.<\/p>\n
\"Adam<\/a><\/p>\n
Co ty validace<\/h2>\n
Asi jste pochopili, \u017ee pou\u0161t\u011bt agenty do sebel\u00e9pe p\u0159ipraven\u00fdch dat bez prvotn\u00edho testov\u00e1n\u00ed p\u0159esnosti a pr\u016fb\u011b\u017en\u00e9 validace je celkem blb\u00fd n\u00e1pad. Bohu\u017eel to ale b\u00fdv\u00e1 nej\u010dast\u011bj\u0161\u00ed postup. Pro firmy v\u011bt\u0161inou p\u0159\u00edprava na agenty v datech kon\u010d\u00ed \u00fapravami datov\u00e9ho skladu, postaven\u00edm s\u00e9mantick\u00e9 vrstvy a spu\u0161t\u011bn\u00edm datov\u00e9ho katalogu.<\/p>\n
\u00dapln\u00e9 minimum by m\u011blo b\u00fdt z\u00e1t\u011b\u017eov\u00e9 testov\u00e1n\u00ed p\u0159ed spu\u0161t\u011bn\u00edm, p\u0159i zm\u011bn\u011b modelu nebo roz\u0161\u00ed\u0159en\u00edm oblasti, ke kter\u00e9 m\u00e1 agent p\u0159\u00edstup (nap\u0159\u00edklad budu cht\u00edt, aby agent nov\u011b analyzoval i skladov\u00e1 data).<\/p>\n
A jak by takov\u00e1 offline validace m\u011bla vypadat?<\/p>\n
Kombinace jednodu\u0161e vyhodnotiteln\u00fdch ot\u00e1zek z minulosti (m\u00e1me na n\u011b p\u0159ipraven\u00e9 odpov\u011bdi) a referen\u010dn\u00edch ot\u00e1zek, p\u0159i kter\u00fdch ov\u011b\u0159ujeme, jak agent rozum\u00ed pojm\u016fm jako \u201cnejhor\u0161\u00ed kampa\u0148\u201d, nebo \u201cminul\u00fd t\u00fdden\u201d. Krom\u011b toho tak\u00e9 sledujeme, jestli n\u00e1sleduje doporu\u010den\u00fd postup (s\u00e9mantick\u00e1 vrstva -> odpov\u00eddaj\u00edc\u00ed L3, pot\u0159ebn\u00e1 kontrola v datov\u00e9m katalogu), p\u0159\u00edpadn\u011b analytick\u00e9 best practices, kter\u00e9 by dle skill\u016f m\u011bl (pokud testujete extern\u00edho agenta a ne built in, jako je t\u0159eba Gemini in BigQuery) dodr\u017eovat.<\/p>\n
Stejn\u011b by se m\u011bly sledovat nejen p\u0159ipraven\u00e9 dotazy v r\u00e1mci offline validace, ale i pr\u016fb\u011b\u017en\u00e9 agentn\u00ed v\u00fdstupy. Teprve ty mohou odhalit slep\u00e9 skvrny, kter\u00e9 by n\u00e1m jinak unikly. To je ale daleko slo\u017eit\u011bj\u0161\u00ed, proto\u017ee je pot\u0159eba p\u0159ipravit celou infrastrukturu k logov\u00e1n\u00ed vstup\u016f, agentova postupu, fin\u00e1ln\u00edho v\u00fdstupu, p\u0159\u00edpadn\u00e9 u\u017eivatelsk\u00e9 zp\u011btn\u00e9 vazby, ale i hodnocen\u00ed efektivity dotaz\u016f na datab\u00e1zi a hled\u00e1n\u00ed vzorc\u016f u nep\u0159esn\u00fdch v\u00fdstup\u016f.<\/p>\n
Na webin\u00e1\u0159i jsme se dotkli ot\u00e1zky cloud-based a self-hosted model\u016f, bezpe\u010dnosti a celkov\u011b ot\u00e1zky vendor locku. Jakmile z\u00e1znam uprav\u00edme, r\u00e1di bychom ho za p\u00e1r m\u011bs\u00edc\u016f zve\u0159ejnili. P\u0159\u00edpadn\u011b pokud si jej chcete pustit je\u0161t\u011b v syrov\u00e9 form\u011b, dejte n\u00e1m v\u011bd\u011bt a po\u0161leme v\u00e1m odkaz. Ka\u017edop\u00e1dn\u011b stay tuned na dal\u0161\u00ed tu\u010d\u0148\u00e1\u010d\u00ed webin\u00e1\u0159e a akce. Nezpomalujeme.<\/p>\n","protected":false},"excerpt":{"rendered":"
Agent v datech? Co ud\u011blat pro to, aby se v nich neztratil a nep\u00e1lil zbyte\u010dn\u011b jak moc sv\u00fdch token\u016f, tak computing datov\u00e9ho skladu? Tedy: aby n\u00e1s nebolel \u00fa\u010det za n\u00e1\u0161 BigQuery \u00fa\u010det?<\/p>\n
Na posledn\u00edm tu\u010d\u0148\u00e1\u010d\u00edm webin\u00e1\u0159i jsme prob\u00edrali jak doporu\u010den\u00ed Anthropicu pro pr\u00e1ci s agentickou analytikou, tak na\u0161e lok\u00e1ln\u00ed zku\u0161enosti, kter\u00e9 jsou vzhledem k mo\u017enostem m\u00edstn\u00edm firm\u00e1m bli\u017e\u0161\u00ed. Agentick\u00e1 analytika m\u00e1 dnes dv\u011b nejd\u016fle\u017eit\u011bj\u0161\u00ed \u010d\u00e1sti: konverza\u010dn\u00ed analytiku a vlastn\u00ed tvorbu p\u0159ehled\u016f, co\u017e beru jako nadstavbu nad konverza\u010dn\u00ed analytikou.<\/p>\n","protected":false},"author":7,"featured_media":8695,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"wp_typography_post_enhancements_disabled":false,"footnotes":""},"categories":[370,70],"tags":[169,340,432],"class_list":["post-8688","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-analytika","category-marketing","tag-automatizace","tag-data","tag-data-driven"],"yoast_head":"\nAgenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor | igloonet blog<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/\" \/>\n<meta property=\"og:locale\" content=\"cs_CZ\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor | igloonet blog\" \/>\n<meta property=\"og:description\" content=\"Agent v datech? Co ud\u011blat pro to, aby se v nich neztratil a nep\u00e1lil zbyte\u010dn\u011b jak moc sv\u00fdch token\u016f, tak computing datov\u00e9ho skladu? Tedy: aby n\u00e1s nebolel \u00fa\u010det za n\u00e1\u0161 BigQuery \u00fa\u010det? Na posledn\u00edm tu\u010d\u0148\u00e1\u010d\u00edm webin\u00e1\u0159i jsme prob\u00edrali jak doporu\u010den\u00ed Anthropicu pro pr\u00e1ci s agentickou analytikou, tak na\u0161e lok\u00e1ln\u00ed zku\u0161enosti, kter\u00e9 jsou vzhledem k mo\u017enostem m\u00edstn\u00edm firm\u00e1m bli\u017e\u0161\u00ed. Agentick\u00e1 analytika m\u00e1 dnes dv\u011b nejd\u016fle\u017eit\u011bj\u0161\u00ed \u010d\u00e1sti: konverza\u010dn\u00ed analytiku a vlastn\u00ed tvorbu p\u0159ehled\u016f, co\u017e beru jako nadstavbu nad konverza\u010dn\u00ed analytikou.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/\" \/>\n<meta property=\"og:site_name\" content=\"igloonet blog\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/igloonet\" \/>\n<meta property=\"article:published_time\" content=\"2026-06-30T13:51:07+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-07-01T05:29:16+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/igloonet.cz\/blog\/wp-content\/uploads\/2026\/06\/adam-ai-data.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1200\" \/>\n\t<meta property=\"og:image:height\" content=\"628\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Adam \u0160ilhan\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@NEmarketak\" \/>\n<meta name=\"twitter:site\" content=\"@igloonet\" \/>\n<meta name=\"twitter:label1\" content=\"Napsal(a)\" \/>\n\t<meta name=\"twitter:data1\" content=\"Adam \u0160ilhan\" \/>\n\t<meta name=\"twitter:label2\" content=\"Odhadovan\u00e1 doba \u010dten\u00ed\" \/>\n\t<meta name=\"twitter:data2\" content=\"10 minut\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/\"},\"author\":{\"name\":\"Adam \u0160ilhan\",\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/#\\\/schema\\\/person\\\/a1aada43c3b360cc63f8e2c01809d429\"},\"headline\":\"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor\",\"datePublished\":\"2026-06-30T13:51:07+00:00\",\"dateModified\":\"2026-07-01T05:29:16+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/\"},\"wordCount\":2076,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/06\\\/adam-ai-data.png\",\"keywords\":[\"automatizace\",\"data\",\"data-driven\"],\"articleSection\":[\"Analytika\",\"Marketing\"],\"inLanguage\":\"cs\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/\",\"url\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/\",\"name\":\"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor | igloonet blog\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/06\\\/adam-ai-data.png\",\"datePublished\":\"2026-06-30T13:51:07+00:00\",\"dateModified\":\"2026-07-01T05:29:16+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/#breadcrumb\"},\"inLanguage\":\"cs\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"cs\",\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/#primaryimage\",\"url\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/06\\\/adam-ai-data.png\",\"contentUrl\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/06\\\/adam-ai-data.png\",\"width\":1200,\"height\":628,\"caption\":\"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Analytika\",\"item\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/rubriky\\\/analytika\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/#website\",\"url\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/\",\"name\":\"igloonet blog\",\"description\":\"\",\"publisher\":{\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"cs\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/#organization\",\"name\":\"igloonet\",\"url\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"cs\",\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/wp-content\\\/uploads\\\/2015\\\/07\\\/RGB_LVPC_igloonet.png\",\"contentUrl\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/wp-content\\\/uploads\\\/2015\\\/07\\\/RGB_LVPC_igloonet.png\",\"width\":650,\"height\":465,\"caption\":\"igloonet\"},\"image\":{\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/www.facebook.com\\\/igloonet\",\"https:\\\/\\\/x.com\\\/igloonet\",\"https:\\\/\\\/instagram.com\\\/igloonet\\\/\",\"https:\\\/\\\/www.linkedin.com\\\/company\\\/igloonet\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/#\\\/schema\\\/person\\\/a1aada43c3b360cc63f8e2c01809d429\",\"name\":\"Adam \u0160ilhan\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"cs\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/aa03ce5f493a168de15b09b9835eb6f92c752b564ac1f18cf0ff09c3863bcba2?s=96&d=retro&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/aa03ce5f493a168de15b09b9835eb6f92c752b564ac1f18cf0ff09c3863bcba2?s=96&d=retro&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/aa03ce5f493a168de15b09b9835eb6f92c752b564ac1f18cf0ff09c3863bcba2?s=96&d=retro&r=g\",\"caption\":\"Adam \u0160ilhan\"},\"description\":\"Vedu igloo t\u00fdm analytik\u016f a s klienty \u0159e\u0161\u00edm tu z\u00e1bavnou \u010d\u00e1st na pomez\u00ed dat a marketingu. V posledn\u00ed dob\u011b m\u011b chytnul freediving, tak m\u011b ob\u010das m\u016f\u017eete vid\u011bt topit.\",\"sameAs\":[\"https:\\\/\\\/igloonet.cz\\\/adam\",\"https:\\\/\\\/www.linkedin.com\\\/in\\\/adamsilhan\\\/\",\"https:\\\/\\\/x.com\\\/NEmarketak\"],\"url\":\"https:\\\/\\\/igloonet.cz\\\/blog\\\/author\\\/adam\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor | igloonet blog","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/","og_locale":"cs_CZ","og_type":"article","og_title":"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor | igloonet blog","og_description":"Agent v datech? Co ud\u011blat pro to, aby se v nich neztratil a nep\u00e1lil zbyte\u010dn\u011b jak moc sv\u00fdch token\u016f, tak computing datov\u00e9ho skladu? Tedy: aby n\u00e1s nebolel \u00fa\u010det za n\u00e1\u0161 BigQuery \u00fa\u010det? Na posledn\u00edm tu\u010d\u0148\u00e1\u010d\u00edm webin\u00e1\u0159i jsme prob\u00edrali jak doporu\u010den\u00ed Anthropicu pro pr\u00e1ci s agentickou analytikou, tak na\u0161e lok\u00e1ln\u00ed zku\u0161enosti, kter\u00e9 jsou vzhledem k mo\u017enostem m\u00edstn\u00edm firm\u00e1m bli\u017e\u0161\u00ed. Agentick\u00e1 analytika m\u00e1 dnes dv\u011b nejd\u016fle\u017eit\u011bj\u0161\u00ed \u010d\u00e1sti: konverza\u010dn\u00ed analytiku a vlastn\u00ed tvorbu p\u0159ehled\u016f, co\u017e beru jako nadstavbu nad konverza\u010dn\u00ed analytikou.","og_url":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/","og_site_name":"igloonet blog","article_publisher":"https:\/\/www.facebook.com\/igloonet","article_published_time":"2026-06-30T13:51:07+00:00","article_modified_time":"2026-07-01T05:29:16+00:00","og_image":[{"width":1200,"height":628,"url":"https:\/\/igloonet.cz\/blog\/wp-content\/uploads\/2026\/06\/adam-ai-data.png","type":"image\/png"}],"author":"Adam \u0160ilhan","twitter_card":"summary_large_image","twitter_creator":"@NEmarketak","twitter_site":"@igloonet","twitter_misc":{"Napsal(a)":"Adam \u0160ilhan","Odhadovan\u00e1 doba \u010dten\u00ed":"10 minut"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/#article","isPartOf":{"@id":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/"},"author":{"name":"Adam \u0160ilhan","@id":"https:\/\/igloonet.cz\/blog\/#\/schema\/person\/a1aada43c3b360cc63f8e2c01809d429"},"headline":"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor","datePublished":"2026-06-30T13:51:07+00:00","dateModified":"2026-07-01T05:29:16+00:00","mainEntityOfPage":{"@id":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/"},"wordCount":2076,"commentCount":0,"publisher":{"@id":"https:\/\/igloonet.cz\/blog\/#organization"},"image":{"@id":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/#primaryimage"},"thumbnailUrl":"https:\/\/igloonet.cz\/blog\/wp-content\/uploads\/2026\/06\/adam-ai-data.png","keywords":["automatizace","data","data-driven"],"articleSection":["Analytika","Marketing"],"inLanguage":"cs","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/","url":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/","name":"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor | igloonet blog","isPartOf":{"@id":"https:\/\/igloonet.cz\/blog\/#website"},"primaryImageOfPage":{"@id":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/#primaryimage"},"image":{"@id":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/#primaryimage"},"thumbnailUrl":"https:\/\/igloonet.cz\/blog\/wp-content\/uploads\/2026\/06\/adam-ai-data.png","datePublished":"2026-06-30T13:51:07+00:00","dateModified":"2026-07-01T05:29:16+00:00","breadcrumb":{"@id":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/#breadcrumb"},"inLanguage":"cs","potentialAction":[{"@type":"ReadAction","target":["https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/"]}]},{"@type":"ImageObject","inLanguage":"cs","@id":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/#primaryimage","url":"https:\/\/igloonet.cz\/blog\/wp-content\/uploads\/2026\/06\/adam-ai-data.png","contentUrl":"https:\/\/igloonet.cz\/blog\/wp-content\/uploads\/2026\/06\/adam-ai-data.png","width":1200,"height":628,"caption":"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor"},{"@type":"BreadcrumbList","@id":"https:\/\/igloonet.cz\/blog\/agenti-v-datech-nejen-konverzacni-analytika-ale-i-na-co-si-dat-pozor\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Analytika","item":"https:\/\/igloonet.cz\/blog\/rubriky\/analytika\/"},{"@type":"ListItem","position":2,"name":"Agenti v datech: nejen konverza\u010dn\u00ed analytika, ale i na co si d\u00e1t pozor"}]},{"@type":"WebSite","@id":"https:\/\/igloonet.cz\/blog\/#website","url":"https:\/\/igloonet.cz\/blog\/","name":"igloonet blog","description":"","publisher":{"@id":"https:\/\/igloonet.cz\/blog\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/igloonet.cz\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"cs"},{"@type":"Organization","@id":"https:\/\/igloonet.cz\/blog\/#organization","name":"igloonet","url":"https:\/\/igloonet.cz\/blog\/","logo":{"@type":"ImageObject","inLanguage":"cs","@id":"https:\/\/igloonet.cz\/blog\/#\/schema\/logo\/image\/","url":"https:\/\/igloonet.cz\/blog\/wp-content\/uploads\/2015\/07\/RGB_LVPC_igloonet.png","contentUrl":"https:\/\/igloonet.cz\/blog\/wp-content\/uploads\/2015\/07\/RGB_LVPC_igloonet.png","width":650,"height":465,"caption":"igloonet"},"image":{"@id":"https:\/\/igloonet.cz\/blog\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/igloonet","https:\/\/x.com\/igloonet","https:\/\/instagram.com\/igloonet\/","https:\/\/www.linkedin.com\/company\/igloonet"]},{"@type":"Person","@id":"https:\/\/igloonet.cz\/blog\/#\/schema\/person\/a1aada43c3b360cc63f8e2c01809d429","name":"Adam \u0160ilhan","image":{"@type":"ImageObject","inLanguage":"cs","@id":"https:\/\/secure.gravatar.com\/avatar\/aa03ce5f493a168de15b09b9835eb6f92c752b564ac1f18cf0ff09c3863bcba2?s=96&d=retro&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/aa03ce5f493a168de15b09b9835eb6f92c752b564ac1f18cf0ff09c3863bcba2?s=96&d=retro&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/aa03ce5f493a168de15b09b9835eb6f92c752b564ac1f18cf0ff09c3863bcba2?s=96&d=retro&r=g","caption":"Adam \u0160ilhan"},"description":"Vedu igloo t\u00fdm analytik\u016f a s klienty \u0159e\u0161\u00edm tu z\u00e1bavnou \u010d\u00e1st na pomez\u00ed dat a marketingu. V posledn\u00ed dob\u011b m\u011b chytnul freediving, tak m\u011b ob\u010das m\u016f\u017eete vid\u011bt topit.","sameAs":["https:\/\/igloonet.cz\/adam","https:\/\/www.linkedin.com\/in\/adamsilhan\/","https:\/\/x.com\/NEmarketak"],"url":"https:\/\/igloonet.cz\/blog\/author\/adam\/"}]}},"amp_enabled":true,"_links":{"self":[{"href":"https:\/\/igloonet.cz\/blog\/wp-json\/wp\/v2\/posts\/8688","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/igloonet.cz\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/igloonet.cz\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/igloonet.cz\/blog\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/igloonet.cz\/blog\/wp-json\/wp\/v2\/comments?post=8688"}],"version-history":[{"count":3,"href":"https:\/\/igloonet.cz\/blog\/wp-json\/wp\/v2\/posts\/8688\/revisions"}],"predecessor-version":[{"id":8701,"href":"https:\/\/igloonet.cz\/blog\/wp-json\/wp\/v2\/posts\/8688\/revisions\/8701"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/igloonet.cz\/blog\/wp-json\/wp\/v2\/media\/8695"}],"wp:attachment":[{"href":"https:\/\/igloonet.cz\/blog\/wp-json\/wp\/v2\/media?parent=8688"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/igloonet.cz\/blog\/wp-json\/wp\/v2\/categories?post=8688"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/igloonet.cz\/blog\/wp-json\/wp\/v2\/tags?post=8688"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}