[{"data":1,"prerenderedAt":542},["ShallowReactive",2],{"site-chrome-en-id":3,"site-chrome-en":-1,"site-chrome-th":-1,"glossary-term-en-id":287,"blog-categories-en-id":300,"article-glossary-en-id-\u002Fglossary\u002Frecommendation-confidence":320,"og-copy-en-id":478,"og-posts-en-id-none":520,"glossary-related-en-id-recommendation-confidence":521},{"id":4,"extension":5,"footer":6,"header":116,"indexFacets":202,"meta":243,"notFound":244,"stem":250,"ui":251,"__hash__":286},"chrome_en\u002Fen\u002Fchrome.yml","yml",{"columns":7,"legal":83,"quote":84,"askAi":87,"socials":103},[8,29,47,65],{"label":9,"children":10},"Product",[11,14,17,20,23,26],{"label":12,"to":13},"Ads optimization","\u002Fen-id\u002Fblog\u002Fads-optimization",{"label":15,"to":16},"Unit economics","\u002Fen-id\u002Fblog\u002Fprofit-margin",{"label":18,"to":19},"Agentic KAM","https:\u002F\u002Flearn.dataglasslabs.com\u002Fdocs\u002Fai-and-api\u002Fassistants",{"label":21,"to":22},"One-click deploy","https:\u002F\u002Flearn.dataglasslabs.com\u002Fdocs\u002Fgrowth\u002Factions\u002Faction-flow",{"label":24,"to":25},"Integrations","https:\u002F\u002Flearn.dataglasslabs.com\u002Fdocs\u002Foverview\u002Fget-started\u002Fsupported-platforms",{"label":27,"to":28},"Pricing","\u002Fen-id\u002Fpricing",{"label":30,"children":31},"Resources",[32,35,38,41,44],{"label":33,"to":34},"Blog","\u002Fen-id\u002Fblog",{"label":36,"to":37},"Customer stories","\u002Fen-id\u002Fcustomers",{"label":39,"to":40},"Research","\u002Fen-id\u002Fresearch",{"label":42,"to":43},"Glossary","\u002Fen-id\u002Fglossary",{"label":45,"to":46},"Docs","https:\u002F\u002Flearn.dataglasslabs.com\u002Fdocs",{"label":48,"children":49},"Company",[50,53,56,59,62],{"label":51,"to":52},"About","\u002Fen-id\u002Fcompany",{"label":54,"to":55},"Contact","\u002Fen-id\u002Fcontact",{"label":57,"to":58},"Book a demo","\u002Fen-id\u002Fbook-a-demo",{"label":60,"to":61},"Privacy","\u002Fen-id\u002Flegal\u002Fprivacy",{"label":63,"to":64},"Terms","\u002Fen-id\u002Flegal\u002Fterms",{"label":66,"children":67},"DataGlass for",[68,71,74,77,80],{"label":69,"to":70},"Agencies","\u002Fen-id\u002Ffor\u002Fagencies",{"label":72,"to":73},"Brands","\u002Fen-id\u002Ffor\u002Fbrands",{"label":75,"to":76},"Planners and optimizers","\u002Fen-id\u002Ffor\u002Fplanners",{"label":78,"to":79},"Managers and analysts","\u002Fen-id\u002Ffor\u002Fmanagers",{"label":81,"to":82},"Sellers and shop owners","\u002Fen-id\u002Ffor\u002Fsellers","© 2026 DataGlass Labs. All rights reserved.",{"text":85,"caption":86},"Measure what you keep, and then grow what you measure.","The DataGlass team",{"label":88,"prompt":89,"assistants":90},"Ask AI about DataGlass","I want to understand DataGlass, a decision engine for Shopee, Lazada and TikTok Shop sellers that optimizes ads, pricing and promotions for true profit. Use web search to read their website at https:\u002F\u002Fwww.dataglasslabs.com (with an agent-readable summary at https:\u002F\u002Fwww.dataglasslabs.com\u002Fllms.txt), then explain what DataGlass does, who it is for and how it differs from an analytics dashboard.",[91,95,99],{"label":92,"icon":93,"url":94},"ChatGPT","chatgpt","https:\u002F\u002Fchatgpt.com\u002F?prompt={prompt}",{"label":96,"icon":97,"url":98},"Claude","claude","https:\u002F\u002Fclaude.ai\u002Fnew?q={prompt}",{"label":100,"icon":101,"url":102},"Perplexity","perplexity","https:\u002F\u002Fwww.perplexity.ai\u002Fsearch?q={prompt}",[104,108,112],{"label":105,"icon":106,"to":107},"LinkedIn","i-lucide-linkedin","https:\u002F\u002Fwww.linkedin.com\u002Fcompany\u002Fdataglass-labs",{"label":109,"icon":110,"to":111},"Facebook","i-lucide-facebook","https:\u002F\u002Fwww.facebook.com\u002Fdataglasslabsth",{"label":113,"icon":114,"to":115},"Instagram","i-lucide-instagram","https:\u002F\u002Fwww.instagram.com\u002Fdataglasslabs\u002F",{"links":117,"cta":196,"secondaryCta":199},[118,152,191,193,194,195],{"label":9,"to":119,"menu":120},"\u002Fen-id",{"layout":121,"heading":122,"cards":123,"groups":135,"footer":142},"cards","What DataGlass does",[124,128,132],{"label":125,"description":126,"to":13,"ui":127},"True profit ads optimization","Budgets, pacing and Target ROAS tuned for profit.","budget-tomorrow",{"label":129,"description":130,"to":16,"ui":131},"Unit economics breakdown","What you actually earn from every sale.","unit-economics",{"label":18,"description":133,"to":19,"motif":134},"An AI agent that knows your shops.",true,[136],{"label":137,"size":138,"links":139},"More","secondary",[140,141],{"label":21,"to":22},{"label":24,"to":25},[143,148],{"lead":144,"label":145,"icon":146,"to":147},"DataGlass v.2.6.","See what's new","i-lucide-box","https:\u002F\u002Flearn.dataglasslabs.com\u002Freleases",{"lead":149,"label":150,"icon":151,"to":37},"Read our","customer stories","i-lucide-book-open",{"label":30,"to":34,"menu":153},{"layout":154,"groups":155},"directory",[156,164,170,182],{"label":66,"size":157,"links":158},"primary",[159,160,161,162,163],{"label":69,"to":70},{"label":72,"to":73},{"label":75,"to":76},{"label":78,"to":79},{"label":81,"to":82},{"label":165,"size":157,"links":166},"Insights",[167,168,169],{"label":33,"to":34},{"label":36,"to":37},{"label":39,"to":40},{"label":171,"size":157,"links":172},"Learn",[173,174,177,180],{"label":45,"to":46},{"label":175,"to":176},"Use cases","https:\u002F\u002Flearn.dataglasslabs.com\u002Fuse-cases",{"label":178,"to":179},"Training","https:\u002F\u002Flearn.dataglasslabs.com\u002Ftraining",{"label":181,"to":147},"Changelog",{"label":183,"size":138,"links":184},"Support",[185,186,189],{"label":42,"to":43},{"label":187,"to":188},"Brand kit","\u002Fen-id\u002Fbrand",{"label":190,"to":55},"Contact us",{"label":192,"to":37},"Customers",{"label":27,"to":28},{"label":48,"to":52},{"label":54,"to":55},{"label":197,"to":198},"Start for free","https:\u002F\u002Fapp.dataglasslabs.com\u002Fsign-up",{"label":200,"to":201},"Sign in","https:\u002F\u002Fapp.dataglasslabs.com\u002Fsign-in",{"clear":203,"groups":204},"Clear filters",[205,218],{"key":206,"label":207,"options":208},"platforms","Platform",[209,212,215],{"value":210,"label":211},"shopee","Shopee",{"value":213,"label":214},"lazada","Lazada",{"value":216,"label":217},"tiktok","TikTok Shop",{"key":219,"label":220,"options":221},"format","Format",[222,225,228,231,234,237,240],{"value":223,"label":224},"guide","Guide",{"value":226,"label":227},"article","Article",{"value":229,"label":230},"comparison","Comparison",{"value":232,"label":233},"calculator","Calculator",{"value":235,"label":236},"paper","Paper",{"value":238,"label":239},"survey","Survey",{"value":241,"label":242},"dataset","Dataset",{},{"title":245,"description":246,"cta":247,"errorTitle":249},"Page not found","The page may have moved, or it never existed.",{"label":248,"to":119},"Back to home","Something went wrong","en\u002Fchrome",{"previous":252,"next":253,"pause":254,"play":255,"goTo":256,"expand":257,"close":258,"readMore":259,"learnMore":260,"seeAll":261,"menu":262,"back":263,"share":264,"shareOn":265,"copyLink":266,"copied":267,"onThisPage":268,"minRead":269,"sources":270,"viewMore":271,"searchPosts":272,"allPosts":273,"all":274,"filters":275,"search":276,"searchClose":277,"searchPlaceholder":278,"searchSuggested":279,"searchResults":280,"searchNoResults":281,"searchLoading":282,"relatedTitle":283,"mainNav":284,"footerNav":285},"Previous","Next","Pause","Play","Go to item","Expand","Close","Read more","Learn more","See all","Menu","Back","Share","Share on","Copy link","Link copied","On this page","{n} min read","Sources","View more","Search posts","All posts","All","Filters","Search","Close search","Search DataGlass","Suggested","Results","No results for","Loading","More to discover","Main","Footer","HZQSEulcpFGY0nWnUTNYhG2n06nO97XnUBc_pEVOFxo",{"id":288,"cta":289,"extension":5,"glossary":42,"meta":295,"related":296,"stem":297,"whatIs":298,"__hash__":299},"glossaryTerm_en\u002Fen\u002Fglossary-term.yml",{"headline":290,"description":291,"primaryCta":292,"secondaryCta":293},"Put this number to work on your own shops","See it computed from your orders, fees and ads, with a demo on your data.",{"label":57,"to":58},{"label":294,"to":28},"See pricing",{},"Related terms","en\u002Fglossary-term","What is {term}?","lxyPsyClqTqVxZhmfDihDsuAz7Al3t22gFHKeu2Lza8",[301,304,306,309,312,315,318],{"slug":302,"label":303},"profit","Profit",{"slug":305,"label":27},"pricing",{"slug":307,"label":308},"ads","Ads",{"slug":310,"label":311},"operations","Operations",{"slug":313,"label":314},"market","Market",{"slug":316,"label":317},"data-science","Data Science",{"slug":319,"label":48},"company",{"entry":321,"document":355,"toc":469},{"id":322,"title":323,"aliases":324,"category":316,"cta":325,"date":328,"description":329,"draft":330,"draftReason":324,"extension":331,"meta":332,"navigation":324,"ogImage":324,"path":333,"readMinutes":334,"related":335,"robots":324,"schemaOrg":324,"seo":341,"short":327,"sitemap":343,"sources":344,"stem":353,"updatedAt":324,"__hash__":354},"glossary_en\u002Fglossary\u002Frecommendation-confidence.md","Recommendation confidence",null,{"title":326,"body":327},"See it on your own shop, free.","How often a proposed change beats doing nothing. A comparison against the status quo, not a score for the campaign.","2026-10-23","Recommendation confidence is the probability that an action beats the status quo, measured on paired scenarios. What it is not, and why the pairing matters.",false,"md",{},"\u002Fglossary\u002Frecommendation-confidence",3,[336,337,338,339,340],"probability-of-loss","cvar","bayesian-inference","incrementality","thompson-sampling",{"title":342,"description":329},"What is Recommendation Confidence? | DataGlass",{"loc":333},[345,349],{"title":346,"url":347,"note":348},"Russo, Van Roy, Kazerouni, Osband & Wen — A Tutorial on Thompson Sampling (arXiv:1707.02038)","https:\u002F\u002Farxiv.org\u002Fabs\u002F1707.02038","Background on posterior-based sequential decision making and on comparing candidate actions across sampled scenarios rather than against point estimates.",{"title":350,"url":351,"note":352},"Expected shortfall","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FExpected_shortfall","The magnitude-aware complement that covers the acknowledged blind spot of a probability-based decision gate.","glossary\u002Frecommendation-confidence","mawzaeCoKRB99D7ezVdrmRHqlZOkVhKcEPBBMjTk5Us",{"frontmatter":356,"meta":363,"nodes":392},{"title":323,"description":329,"seo":357,"short":327,"category":316,"related":358,"date":328,"sources":359,"cta":362,"readMinutes":334},{"title":342,"description":329},[336,337,338,339,340],[360,361],{"title":346,"url":347,"note":348},{"title":350,"url":351,"note":352},{"title":326,"body":327},{"toc":364},{"title":365,"searchDepth":366,"depth":366,"links":367},"",2,[368,371,374,377,380,383,386,389],{"id":369,"depth":366,"text":370},"formula","Formula",{"id":372,"depth":366,"text":373},"example","Example",{"id":375,"depth":366,"text":376},"why-must-the-comparison-be-paired","Why must the comparison be paired?",{"id":378,"depth":366,"text":379},"how-is-it-different-from-probability-of-loss","How is it different from probability of loss?",{"id":381,"depth":366,"text":382},"what-does-confidence-deliberately-not-tell-you","What does confidence deliberately not tell you?",{"id":384,"depth":366,"text":385},"why-it-matters","Why it matters",{"id":387,"depth":366,"text":388},"in-dataglass","In DataGlass",{"id":390,"depth":366,"text":391},"where-this-shows-up","Where this shows up",[393,397,400,409,411,416,418,421,423,426,428,431,433,436,441,443,446,448],[394,395,396],"p",{},"Recommendation confidence is the probability that taking a recommended action leaves the business better off than doing nothing. It is a relative measure, anchored to the current setting rather than to zero, and it is a probability rather than an amount: it answers how often the action wins, not how much it wins by. A confidence of 0.8 on a budget increase means that across the plausible futures the model considers, the higher budget produces more profit than the current one in four cases out of five. A confidence near 0.5 means the change is a coin flip and the honest recommendation is to leave it alone.",[398,399,370],"h2",{"id":369},[401,402,404],"pre",{"language":403},"text",[405,406,408],"code",{"class":407},"language-text","confidence = P( profit(recommended) ≥ profit(current) )\n\nEvaluated on paired scenarios — the same simulated demand applied to\nboth settings — so the comparison isolates the change:\n\n  Δ_i = profit_i(recommended) − profit_i(current)\n  confidence = share of scenarios where Δ_i ≥ 0",[398,410,373],{"id":372},[401,412,413],{"language":403},[405,414,415],{"class":407},"A campaign flagged for a spend cut.\n\n  probability of loss at current spend   0.65\n  confidence that cutting spend helps    0.017\n\nThe campaign loses money on most days, and cutting it beats keeping it\nin under 2% of scenarios. Both numbers are true. Only the second one\nis about the decision being made.",[398,417,376],{"id":375},[394,419,420],{},"Because otherwise the noise swamps the signal. If the current setting and the proposed setting are each simulated with their own independent random demand, the difference between them carries the randomness of both, and a small genuine improvement disappears into simulation jitter. Pairing means applying the identical simulated demand to both settings — the same quiet Tuesday, the same campaign-day surge — so that whatever is common to both cancels and what remains is the effect of the change itself. This is the common random numbers technique, and without it a confidence figure is mostly measuring the simulator.",[398,422,379],{"id":378},[394,424,425],{},"Different reference point, different question, and they routinely disagree. Probability of loss asks whether profit at the current setting is below zero. Confidence asks whether the proposed setting beats the current one. A lumpy but profitable campaign scores badly on the first and, correctly, protects itself on the second. Because they are anchored differently, one minus the probability of loss is not the confidence, and treating them as complements is the single most common error made with these two metrics.",[398,427,382],{"id":381},[394,429,430],{},"How much is at stake. Being a probability, it is blind to magnitude in the same way probability of loss is: an action that wins 55% of the time by a tiny amount and loses 45% of the time catastrophically scores as a modest positive. That is the acknowledged blind spot, and it is covered by pairing confidence with a magnitude-aware tail measure such as conditional value at risk. The reason confidence is still the right primary gate for most advertising decisions is that it is robust — a probability does not get dragged around by a single enormous simulated outcome, whereas an average does.",[398,432,385],{"id":384},[394,434,435],{},"Most automated recommendations are generated by comparing a metric against a threshold, which asks whether a campaign looks bad rather than whether the proposed fix is an improvement. Those are different questions and they disagree most sharply on exactly the campaigns worth getting right: the profitable-but-lumpy ones, which look bad on any absolute metric and should not be touched. Confidence puts the comparison where the decision is.",[437,438,440],"blog-callout",{"title":439},"Common mistake","Treating confidence as a quality score for the campaign. It scores the action, not the campaign. A very low confidence on a proposed pause is a strong statement that the campaign should keep running — it is not a statement that the campaign is performing badly.",[398,442,388],{"id":387},[394,444,445],{},"DataGlass evaluates every ads recommendation as a paired comparison against the current setting on the same simulated demand, and surfaces the resulting confidence on the recommendation itself. Actions whose confidence does not clear the bar are not shown as recommendations, which is why a campaign that looks alarming on absolute metrics may correctly produce no suggested change.",[398,447,391],{"id":390},[449,450,451,457,463],"ul",{},[452,453,454],"li",{},[455,456,12],"a",{"href":13},[452,458,459],{},[455,460,462],{"href":461},"\u002Fen-id\u002Fblog\u002Fhow-ecommerce-decision-engines-work","E-commerce Decision Engine: How Marketplace Sellers Turn Data Into Profit Recommendations",[452,464,465],{},[455,466,468],{"href":467},"\u002Fen-id\u002Fblog\u002Fcross-platform-ad-budget-allocation","Cross-platform ad budget allocation for SEA marketplace sellers",[470,471,472,473,474,475,476,477],{"id":369,"depth":366,"text":370},{"id":372,"depth":366,"text":373},{"id":375,"depth":366,"text":376},{"id":378,"depth":366,"text":379},{"id":381,"depth":366,"text":382},{"id":384,"depth":366,"text":385},{"id":387,"depth":366,"text":388},{"id":390,"depth":366,"text":391},{"id":479,"categories":480,"domain":487,"extension":5,"fallback":488,"home":491,"industry":500,"meta":505,"pricing":506,"stem":518,"__hash__":519},"og_en\u002Fen\u002Fog.yml",[481,482,483,484,485,486],{"slug":302,"label":303},{"slug":305,"label":27},{"slug":307,"label":308},{"slug":310,"label":311},{"slug":316,"label":317},{"slug":319,"label":48},"dataglasslabs.com",{"title":489,"description":490},"DataGlass","The decision engine for Shopee, Lazada and TikTok Shop sellers.",{"chart":492,"metrics":493},"Net profit, last 30 days",[494,497],{"label":495,"value":496},"Net profit","฿1.24M",{"label":498,"value":499},"Profit ROAS","3.4x",{"axis":501,"point":502},"Ad spend",{"label":503,"value":504},"Next baht still pays","+฿0.42",{},{"plans":507},[508,512,515],{"name":509,"price":510,"period":511},"Free","฿0","\u002Fmonth",{"name":513,"price":514,"period":511},"Standard","฿1,499",{"name":516,"price":517,"period":511},"Pro","฿2,499","en\u002Fog","63L8y4Lv8I2q3NvR26fMXU1sE9Wd7TZaz2l_tV6oZuo",[],[522,526,530,534,538],{"path":523,"title":524,"short":525,"draft":330},"\u002Fglossary\u002Fprobability-of-loss","Probability of loss","How often a campaign has a losing day. A frequency, not an amount — and on sparse campaigns it measures the wrong thing.",{"path":527,"title":528,"short":529,"draft":330},"\u002Fglossary\u002Fcvar","CVaR","Conditional Value at Risk — the average outcome in the worst cases. Not how often it goes wrong, but how bad it gets when it does.",{"path":531,"title":532,"short":533,"draft":330},"\u002Fglossary\u002Fbayesian-inference","Bayesian inference","Estimating with a range instead of a single number, so a decision can account for how little it actually knows.",{"path":535,"title":536,"short":537,"draft":330},"\u002Fglossary\u002Fincrementality","Incrementality","The sales an ad actually caused, as opposed to the sales it was credited with. Usually a much smaller number.",{"path":539,"title":540,"short":541,"draft":330},"\u002Fglossary\u002Fthompson-sampling","Thompson Sampling","A way to keep learning about uncertain campaigns while still funding the proven ones — exploration without a fixed test budget.",1792785067134]