menu close
menu close
back to Insight Page

What Performance Marketers Need to Know About Their Algorithms

Written by: Cheryl Morris, VP Marketing

If you’re spending millions of dollars across social and/or mobile, there’s a good chance you’re working with an agency or a SaaS platform like Nanigans. While this should be a positive and profitable experience, one question that never seems to get answered for the “customer” is “so how does this tool and these algorithms work?”

The teams behind these algorithms are usually data scientists. Data scientists are smart. Most, (ours) have PhDs from the top universities around the world including MIT and Harvard. So when we talk to them about how to approach or explain performance marketing from an algorithmic perspective we expect that they’ll have sound theories. Unfortunately, however, we also expect to not understand one word that comes out of their mouths. As marketers, we don’t often use phrases in our every day lives such as…

  • Non-Parametric imputation
  • Regularized statistical model
  • Hierarchical pooling
  • Intra-day yield optimization

First and foremost, all of that sounds like Latin and let’s be honest, we (marketers) have other things to think about such as our offline efforts across TV, Print or Radio. We have branding campaigns and creative meetings. We’re aligning the sales and marketing departments and worried about product inventory and our company roadmap.

Performance marketing is one component in our overall marketing mix. Yet, when it comes to performance marketing across desktop and mobile, we’re expected to select/work with the best vendor, with the best data science team and product, and then understand in detail how their algorithms work? Oh and let’s not forget that at some point your boss is going to ask for an explanation.

Nope. Too hard.

So what then as a marketer, should we know about in regards to performance marketing algorithms? In addition to “knowing,” what should we truly understand and go deep on?  Before we even talk algorithms, there are a few things you should know (whether you’re already working with a solution like Nanigans or not).

What You Should Know:

1.) Black Box Doesn’t Exist: If you can’t see detailed and fully (100%) transparent reports about how your budget is being spent, or would be spent during a potential campaign, then stop and demand the data because it’s 2013 and it’s readily available. If you aren’t being given this level of honesty, then it’s time for a change. Let me repeat that. If you can’t see detailed and fully (100%) transparent reports about how your budget is being spent, or would be spent during a potential campaign, then stop and demand the data because it’s 2013 and is readily available. If you aren’t being given this level of honesty, then it’s time for a change. 

2.) Everything Can Be Explained: As a marketer, you should be able to ask your vendor all about your strategy from a high level. You should ask about creative, targeting and optimization. Creative should consist of aesthetics, language and relevancy. Targeting options should consist of more combinatorial options than you know what to do with (Nanigans has over a googol targeting combinations). Optimization should consist of cohort analysis, affinity models and maturity curves.

3.) Positive ROI isn’t a “Celebratory Achievement ”: This is performance marketing, which means positive return on ad spend. It’s a given that within your marketing mix, performance marketing has a positive ROI, and when it comes to performance marketing the focus should be on ROI and lifetime value, NOT any other metrics/measurables such as:

  • Engagement
  • Impressions
  • Clicks
  • Leads
  • CPC
  • CPM
  • CPL
  • CPA

Performance marketing is when you get what you want according to what’s reasonable for YOUR business.

So then what about going a little bit more in depth and understanding a few key elements to performance marketing algorithms? You don’t need your PhD in statistics in order to be lead a successful performance marketing practice, but there are a few things that you should and can understand.

What You Can Understand:

Algorithms are simply bidding rules put in place to execute on a desired outcome, for example an advertiser may want to acquire net new customers that look like “X,” cost, “Y” and generate a return on ad spend of “Z.” With any performance marketing based algorithm though, there are a few things that you’ll want to make sure you understand…and that you can understand.

Performance Marketing Algorithms Should Incorporate Flexible Value Modeling: At Nanigans, we work with over 250 advertisers across ecommerce, travel, financial services and gaming, so as you can imagine, “value” is defined differently across all of our clients. The important part of any performance marketing algorithm is that is can be flexible to your exact needs. Certain advertisers are able to place pixels directly into downstream transactions such as purchases, however, others are not which will impact the campaign optimization methodology.

If an advertiser can pass back direct transactions and purchase revenue, a specific algorithm will optimize to real time, however, if an advertiser would rather provide purchase information at the end of each day or week then it’s understandable that the algorithm would need to change. In addition, various advertisers and verticals see different conversion windows. There’s often a direct correlation between click and the inevitable purchase, however, it often differs by advertiser depending on conversion windows, attribution types or the advertiser/vertical. The takeaway here is that if your algorithms aren’t customized specifically for what you define as value (which can change), then they’re underperforming.

2.) Performance Marketing Algorithms Should be Based on Portfolio Optimization: What is portfolio optimization? It means taking into account numerous factors when optimizing. Just like in the venture capital world, VC firms try to have a balanced portfolio at all times and invest in areas where they’re weak but could be realizing value, or in emerging areas of growth. Portfolio based optimization is similar, in the sense that your optimization algorithms shouldn’t be based on just one thing but rather numerous factors that are always updating and changing in real time.

For instance, if you were to run an ad copy test over a two-week period, achieve statistical significance and then use the “winning” ads moving forward you would be operating with a false sense of confidence.  Below are just a few examples of what a portfolio based optimization strategy would include outside of simply CTR or even purchase rates:

  • Ad copy updates based on location, current news, events, or purchase patterns/trends
  • Inventory tracking to only be showing ads for products that are currently in stock
  • Conversion window tracking and channel attribution consideration

With so many ad campaigns and initiatives live at the same time, there are numerous factors that impact performance, and they change daily. What worked yesterday may not work today, and what worked last month will certainly need to be reconfigured this month.  With each day that a campaign is live, there’s uncertainty as you go to market and that uncertainty needs focus as well as the ability to be universally optimized across numerous contributing factors at any time. Ensure that all aspects of the algorithms working for you have the level of sophistication that you need, and can regularize in real time, the technology is there. So next time you’re talking algorithms with the “experts” be sure to ask about flexible value modeling as well as portfolio based optimization.

As a marketer working with an external vendor focused on performance marketing, you should always be generating positive ROI for your business, know how things work and what the strategy is both today and long-term. As a bonus, if you can understand how algorithms work at least at a high level (which clearly you can), you’ll be much more informed to make the right decisions in regards to your marketing efforts and ultimately your profitability.

If you’d like to learn more about performance marketing on Facebook and/or our concept of predictive lifetime value, feel free to contact us.

next post

But wait, there's more

Join Our Newsletter