Comment by Gustomaximus | Hacker News Reader

Comment by Gustomaximus | Hacker News Reader

Parent Full thread

This article is correct in the lack of transparency is concerning, but the simplify eneed view is incorrect IMO.

"I think Google Ads should be simpler to use (dumb it down), should be more transparent with its costs and bids, and should offer more options to control ad appearance"

Firstly, more options won't simplify most likely.

Secondly google has been 'simplifying' for a long time, to their advantage. Increasingly Google push people down the performance/CPA path saying our smart systems will get you the best results 'believe us we're Google' but it's mostly not true. The best results are often from a manual as possible campaign and they make this an ever harder option to take. There needs to be far better regulation here.

Marketing is one of those skills where art meets science. If you let the algorithms take over they'll often look good at a base metric but as soon as you dig around you find you're sacrificing wider goals or mostly getting sales you would have anyway, or paying crazy amounts to finish off the budget type etc.

Google is a great advertising platform, but adding your money, limited controls and trusting google systems to work it out is a horrible way, and something I see all too often. I suspect it's largely because you can spend an < an hour setting the easy way, or spend a day or more setting up a well researched and manual way, but on both client/agency side people don't want to spend the time/money getting foundations right, or simply have the time or aptitude.

I say the best thing is for some regulator body to go over the system changes over the years and ask why features plus analytics were removed, hidden or changed and put rules about these controls and ongoing changes to live up to the level playing field tenent of good capitalism. Maybe it is happening somewhere but I haven't seen it, and it is long overdue.

Would respectfully disagree. Matching a consumers query to an ad, and optimizing that outcome at a given return on investment threshold is a task best left to a computer to optimize. It is a perfect use case for reinforcement learning, and a computer can work 24 hours a day and you need to sleep. In my experience, 100/100 times the machine beats the confident manual optimizer. We (humans) don't have a great track record in beating computers at these kinds of problems, certainly not a game as simple as optimizing for a high score (driving results from an ad).

If the goals for all parties are the same, yes. Google is optimizing for their own fill rates and ePM, _then_ the advertiser's CPA/goals. If there are 100 equally valued conversions for $10 and 500 for $40, you would want all $10 conversions first. This is possible with manual bidding on exact match terms you know convert. With broader match types or many automated bidding strategies, you get a mix of everything (Epsilon greedy/multi-armed bandit strategy) and the CPA will almost always trend towards/exceed the $40 point.

The challenge with that approach is the variation in both (a) queries and (b) other players. (a) There are a small % of queries that are knowable for you to target on exact, or you are required to update and prune infinitely over time. [https://twitter.com/Google/status/1493681643290300425?lang=e...]. (b) A mix is better than a fixed CPA at a set price, up to to the player to pick the acceptable range. At a fixed price and many players competing for the same 100 10$ conversions, the price would quickly rise for those same conversions anyways. You're replicating the work the computer is doing, but doing so with the illusion of control.

It does require manual intervention, and the campaign strategy is typically called "alpha beta" where the beta is a broader match type, with the goal of a lower IS (via budget-- not always AP). As you state, there is in fact more competiton for these terms (for a reason) in efficient markets (both from other manual bidders and automated bidding), but it still works (many times you end up with competitors who hit their budget caps due to their other misaligned spend). In situations where the bids cause CPA to exceed target, you simply choose to lose IS due to bids and shift resources. With automated bidding, you don't always have this option. The key concept that people miss is that automated strategies optimize fill rate for Google-- so you end up with tons of longtail spend that will likely never convert and is hard to identify. Automated strategies will also abstain from auctions if they can't hit specific goals, or will have severely limited volume.

Auctions may vary-- most of my managed spend (mid 8 figures, so not huge) has been in high CPC (think $5-25) verticals like home services, movers, insurance, mortgage and mattresses (insane period from 2015-2018). I have set things up manually, let Google AEs setup the "recommended" bid strategies (and even run longer than their suggested timeframe) and utilized known agencies, and the manual bidding (always accompanied by aggressive negatives, high AP/IS goals) always won by large (30%+ lower CPA with higher volume) amounts. Performance may be drastically different in ecommerce niches where everyone is using automated strategies, as well as other factors like number of auction participants, participant budgets, sophistication and even objectives (executive spite is a valid reason-- hence Google's target outrank share strategy). Anyone who has done PLAs/shopping ads will be incredibly cynical due to their targeting methodology of being all inclusive with the onus of exclusions on the advertiser when the majority of errant spend is in the longtail-- good luck talking to 99% of marketers about tokenization, stemming and lemmatization for identifying negative matching strategies.

So yes, automated bidding can work better than just YOLO'ing some broad match campaigns, but most people shouting their praises, that I have encountered, are parroting their ad rep's talking points. Any competent performance marketer (not many) will be able to outperform automated bidding strategies, but things like scale (why bother saving 20% on 5k spend?), principal/agent problem, or workload (or being lazy, relying on salespeople to do their job) will also have an impact.

Reply on news.ycombinator.com