Photo: Getty Images/Laurence Dutton
New marketing analytics for a new COVID-19 reality
Through a special arrangement, presented here for discussion is an excerpt of a current article from the Joel Rubinson on Marketing Research Consulting blog.
It certainly isn’t business as usual for marketers navigating COVID-19, and it can’t be business as usual for marketing analytics teams either.
Here are three changes that analytics teams need to focus on:
- Reduce reliance on marketing mix models. No one can (or should) trust any updates to marketing mix models for at least a year. They model historical relationships that then rely on the principle that “the past is prologue” to apply their findings to future actions. But nothing could be further from the truth — not this year anyway. Marketers need to hit the pause button on new marketing mix modeling until 2021.
- Increase reliance on user level attribution modeling and experiments. On the other hand, marketers need more help when the past is not prologue. Fresh guidance for the next 12 months, especially as marketers open up advertising again, should come from multi-touch attribution (MTA) and controlled experiments. Both use data that are forward looking, not backward historical. MTA analyzes conversions vs. ad serving as data are unfolding going forward during campaigns and flights of advertising. The experiment is designed and executed in the future, not the past. The fresh results can be focused on the big questions, such as, “Should I start advertising again?” and “What mix of brand and performance marketing works best to balance brand needs and the need to show tangible results?”
- Proving and maximizing the value of targeting. “Reducing wasted ad impressions” or “addressing advertising to consumer targets producing 2X (or more) return” are two sides of the same coin. One is the sickness, the other the cure. Choosing the right segments can make your advertising self-funding and build your brand at the same time. Choose the wrong segment and you are back to ad waste. Either way, there will be financial pressure on marketing to prove that advertising is adding to the bottom line.
Coming out of COVID-19 lockdowns, as we reboot marketing, we must reboot marketing analytics. Don’t just do what you were doing before COVID-19 hit. Change practices in order to provide the right guidance.
Discussion Questions
DISCUSSION QUESTIONS: How will marketing analytics have to be reassessed for COVID-19? Would you add any other suggestions to those presented in the article?
Poll
Sorry, there are no polls available at the moment.
BrainTrust
"Now is the time for experiments rather than models that, almost by definition, won’t work for a while."
Self-servingly, I’m with Joel. Now is the time for experiments rather than models that, almost by definition, won’t work for a while.
The majority of the general public no longer has disposable income to spend on items that are not food or services essential to them. I think that will probably last for a while. The question is how do you repurpose what you have (probably with a reduced budget) to get your message out? It can’t be about “buy this, buy this, buy this.” I will be interested to see how marketers revise their approach for what looks like the long term.
Hypothesis driven experimentation, testing and refining should be the process. Macro indicators on consumer behavior, winning categories and behaviors are just that – they are at macro level. Experimenting and testing will tell what exactly works for each organization.
Anything that relies just on a macro level insight is bound to give sub-optimal results and can cause severe setback to the competitive position, not to mention wasted ad spend.
There are a ton of learnings that will emerge from this crisis. Marketing analytics is one area where true machine learning technologies can help decision making. The more data you feed into your tools, the better perspectives you will get. So… since we’re not quite outta the woods with this COVID-19 stuff, I’d wait to see what “hyperlocal” data inputs you get over the next few months prior to making wholesale changes in your strategy. Each zip code across the country will have different recovery patterns. Get as local as you can with the tools available today.
I’d totally disagree, Ralph — whatever a machine learns from past data is out the window for a while and whatever it learns from recent data is subject to change on a monthly basis. Machine learning is no better than any other form of market analytics at the moment.
Thanks for the reply, Stephen! Having worked with retailers on AI/ML since its inception, I would push back just a bit on throwing past data “out the window,” partly due to the fact that anomalous spikes and dips have occurred in the past and will continue to occur over time. Throwing out these drastic variances simply reduces eventual accuracy, as ML capabilities help better match demand forecasting trend lines versus actual demand lines, just as one example.
For marketing analytics in general, keeping a watchful eye on all data builds a much clearer picture of the market, taking into account all variances, rather than just those most recent trends. We all know business is cyclical in many ways, and this is just one more way we can leverage the past to see where we will most likely emerge from the current dip (for some retailers) and spike (for others).
Great conversation, THANKS!
If I may add: Applying pre-covid models on recent data to get recommendations is surely a bad idea.
Treating the past 3 months as an anomaly and ignoring is equally bad.
What I am seeing is the need reevaluate the assumptions and recalibrate the models. Review each product category and look for the bounce back potential and timeline for each category. Look at the new customers acquired during this period, see if you are attracting new demographics. Looks for customers who dropped off based on Time Between Purchases, that would have otherwise bought by this time.
A lot of data science and business intuition to create hypothesis, experiment is small sets, and refine. Not an easy task, but the worse that organizations could do is ignore the need.
A digital nuance during this time of COVID-19 is the plethora of exact same thoughtless retargeted ads showing up constantly everywhere I go online. Retargeted ads for products I never had an interest in in the first place. While this is nothing new, it seems the folks behind the retargeted ads have thrown up their hands in despair. In particular, a large retailer on the chopping block sending the same product over and over again … did I say over again? I did some investigating out of curiosity; the product is not only out of stock but discontinued.
COVID-19 or no COVID-19, there is not enough thinking going on in the marketing department, rather a reliance by marketers to let machines run wild to do the work and thinking of human marketers. This is not a result of COVID-19, COVID-19 only exposes the folly of misused, misunderstood marketing analytics by humans fearful to think. Have marketing analytics reduced human thinking to the lowest common denominator?
Spot on Cynthia.
So much has changed and from our (customer marketing) perspective that starts with customer insights. Customers have switched brands, defected and gone inactive in staggering numbers since the start of the pandemic. Given that 35% of consumers have tried new brands, it seems one place to start is by profiling new customers and understanding how they compare to existing ones.
These shifts also underscore (no pun intended) existing definitions of HVCs (high value customers) so re-evaluating customer values now might have real implications for loyalty strategies and prioritizing customers. Whatever the new or next normal is, these rapid shifts in customer composition point directly to the value of updating customer marketing analytics.
Very true, Phil. Brands are attracting new customers — new demographic segments, new personas. At the same time they are losing existing ones too. Re-baselining the customer base (should be an ongoing effort, ideally) is a much needed exercise that would inform the marketing strategy.
This is a very important discussion — not only for advertising and promotion decision makers, but also for any professional who is trying to map a path forward in merchandising, category management, fulfillment, even store operations.
There is no choice but to hit the reset button on most of the analytical tools we depend upon in the retail-consumer products industry. The past has been disconnected from the present by the pandemic, and the future has yet to be defined.
Planning for assortment, space, pricing, promotion, and replenishment must be re-started from a blank slate. A worrisome prospect, to be sure, but it also means an opportunity to test scenarios and build a data foundation that will support more accurate decision making over the coming months and beyond.
I agree that the latency of insights and top-down view of traditional media mix models does not support the types of insights needed by marketeers in this level of market disruption. Instead, marketing organizations need more real time “bottom-up” insights which are possible with predictive models based on attribution modeling, Bayesian approaches, and experimentation. In fact, advances in AI/ML now enable a family of ensemble models to forecast ROI and uplift with high rates of accuracy across hundreds of tactics with stability and at lower levels of granularity at brand/week/store, even during the current pandemic event. This level of granularity enables marketing teams to leverage both “art” and “science” to better understand hyper-localized execution, consumption pattern shifts locally, and an integrated view of next best action of dollars spent by forecasting similar events going forward.
I must say, this is one of the most important conversations I’ve seen in relation to the COVID conversation. I just shared it on LinkedIn and challenged retailers to schedule a team deep dive on the topic and really dig into the many compelling points you all are making.
While I live in the world of store/venue analytics, I’m tracking every point you raise while asking myself how the store behavior lens of fit into this larger view. Further, how new metrics play a role–we have anonymous data on temp scans, results, occupancy levels and more. We can tie them to key experience points that are using new technologies and CX experiments. This can tie to anonymous sentiment analytics as well as very general demographics. All of this matters as well!
As Jamie shares, we’re really looking at a reboot here. As Cynthia so cogently stated, it’s going to require marketers to really think and quit hiding behind old models and machines. Joel, I’d LOVE to see you all get on a webcast and dive into this further.