Machine Learning會唔會令Performance Marketer冇工做?
Machine learning而家automate咗bid optimization同targeting,但佢答唔到最重要嘅strategic問題:你到底應唔應該做廣告,同埋佢有冇帶嚟incremental revenue?
開始之前,一啲context會幫你明白我點解有呢個問題。我career入面花咗significant時間喺performance marketing。我仲記得我啱入行嗰陣,Google Adwords同而家完全唔同。例如,以前有一條rule話如果你嘅keyword click-through rate低過0.5%,你嘅keyword就會變成「inactive」,要recover一個inactive keyword就extremely difficult(或者impossible)。
Facebook同一年成立,所以嗰陣仲未有Facebook advertising :D
快進到2022年,machine learning係Google同Meta advertising platforms(同其他好似Amazon嘅platform)嘅核心。兩間公司都advocate account setup simplification(畀machine更多data去work with)、diverse set嘅creatives(formats同concepts兩方面)、同當然site tagging/conversion API畀machine佢需要嘅outcome signal(即conversion)。
以前嗰啲要set up granular search campaign structure或display campaign structure去customize each audience segment配relevant creatives嘅日子已經過去。而家我哋被建議set up just one campaign(喺Performance Max for Google嘅case),machine會自動搵到啱嘅inventory source(search、youtube、Gmail等)同serve最好嘅creatives畀audience(via responsive ad format)。Bid optimization會透過每個platform offer嘅off-the-shelf bid strategies自動進行。Campaigns之間嘅budget optimization亦可以semi-automatically進行。
咁我哋成日做啲咩?:D run excel reports? :P
Machine learning只係一個tool
係,佢係一個potent tool但歸根結底係一個tool。呢個代表machine唔知咩嘢對你嘅business好。(停一停諗下。)
Machine喺達到你set嘅outcome(conversion或ROI)at the right efficiency level方面incredible。但佢唔知達到嗰個goal係唔係suit你嘅business。
佢唔知你應唔應該run Google Ads或Meta ads,或者根本應唔應該做advertising。
Machine有好多嘢唔知。
- 佢唔知你嘅brands或你嘅potential customers。
- 佢唔知點解你嘅potential customers揀你嘅brand而唔係競爭對手。
- 冇大量training data佢做唔到engaging嘅messaging或landing page experiences畀你嘅potential customers。
- 佢唔知做advertising有冇帶嚟incremental revenue畀你嘅business。
- Incremental revenue係如果你唔做ads就唔會有嘅revenue。
- Machine可能知道single channel嘅incremental revenue或conversion但唔係你business嘅overall level。
Walled gardens limit咗machine。
佢可以喺Google ecosystem、Meta ecosystem、Amazon嘅或Tiktok嘅入面做incredible嘅optimization⋯⋯但唔係across佢哋。呢個fact短期內唔太可能改變,鑑於consumer privacy嘅focus。
呢個代表human決定喺邊度run ads同喺每個walled garden入面花幾多錢。
未來3-5年
所以至少目前,我唔擔心我嘅工作會被machine replace,唔會喺未來3-5年。但我應該concern點樣continue提供更多value畀business。例如,我應該繼續worry/學更多關於:
-
了解brand同佢嘅potential customers。
-
了解公司有或者可能冇嘅power。
-
做advertising到底make唔make sense?
-
如果需要advertising,我哋點樣evaluate advertising對business嘅incremental business impact?關鍵字係incremental。
- 對business嘅impact可以喺短期(三個月內)或長期(幾年),所以我哋需要唔同嘅measurement solutions。
-
點樣set up machine for success做performance marketing?
- 鑑於全球唔同嘅privacy laws同regulations,咩framework可以確保我哋respect user privacy、follow the law、同畀到machine需要嘅signals去succeed?
- 以前performance marketers just collect as much data as possible然後send返去ad platform嘅日子已經過去。
- 而家我哋需要purposeful地了解點解certain data係重要或者allowed被收集/使用。同埋點樣以正確嘅方式send返去machine。
- Creative messaging同overall user experience:formats、concepts、imagery、representation等嘅diversity似乎係好多platforms advocate嘅message。
- 鑑於全球唔同嘅privacy laws同regulations,咩framework可以確保我哋respect user privacy、follow the law、同畀到machine需要嘅signals去succeed?
-
點樣set up machine for success做其他non-advertising activities。
你點諗?你有冇見到machine learning改變你day-to-day嘅performance marketing工作,如果有,你double down咗邊啲skills嚟stay ahead?
祝好,
Chandler


