Friday, September 22, 2006

Can Analytics Influence Direct Marketing Creative Process?

Direct Marketing Association (DMA's) Northern California chapter hosted its first independent meeting yesterday (Thursday Sep 21st) at Intuit's campus in Sunnyvale. This summer, the national DMA "abandoned" formal support for all its local chapters asking them all to go on their own. It's good to see that Northern California DMA has managed to make this independent start, hopefully they'll get adequate local support to thrive.

The main speaker was Bill Mirbach, VP of Direct marketing and direct sales at Intuit -- presenting a talk titled " Owner's Manual for the Creative Process". Now, I'll be the first admit that professionally direct marketing creative is the last thing we deal with -- at our work, we leverage direct marketing data for predictive analytics, so while we can measure and predict if creative version A will perform better than version B for a target audience, that is very different from the "creative process" itself. So, I was intrigued with the topic.

Bill has been around the silicon valley high tech industry. The highlight of the talk for me was a story he shared going back to 1984 when he helped the founder of a fledgling software company called Intuit with their direct ads. The talk was mostly about how companies should chose vendors and vice versa for the creative process, not the artistic aspect itself -- so, during Q&A, I asked Bill how does knowledge of your audience impact the direct marketing creative process? Can one leverage their direct marketing data, knowledge of their customer segments, etc. to make the creative process more effective? What's been his experience?

Here's what I heard:

Yes, a better knowledge of one's audience and their likes and dislikes about an organization's products and services certainly helps the creative person to craft their message more effectively? However, while this sounds logical, this is not what normally happens. The creative process is more the product of the discipline and idiosyncracies of the "creator", rather than driven by data-derived intelligence.

Bill mentioned a particular test where they wanted to measure the performance of a scare-tactic message ("if you don't use our product, you'll be sorry") versus a benefits-focused message ("and you can get X, Y, and Z at the click of a button") -- where the creative person just didn't believe in scare-tactic and came up with a very tepid "scary" message (which obviously didn't perform well). Whereas he used a different creative person who specialized in scare-tactic message (scary thought, pardon the pun) -- which turned out to be very effective.

What does this tell us?

I don't know how much of Bill's story is the norm or exception, but he certainly has been around direct marketing creative people a lot more than I have -- so I must respect his POV. Still, it seems like rather than asking creative people to use marketing data driven intelligence to fine tune their message -- it's probably better the other way around -- i.e. leverage the analytics to find out what type of messages you want to be send out to different segments -- THEN find the right creative team to craft those messages.

What do you think?


adelinoyuan said...

Sandeep: this is some great insight into a common problem in DM. I've expanded on your post in my blog:

Adelino de Almeida

Sandeep Giri said...

TrackBack URL for this entry:

This is exactly what I'd hoped to get out of blogging -- conversations! Thanks for elaborating on my post.

In our line of work (predictive analytics side of the house), we often run predictive models to rank customers/prospects by their propensity to either respond to direct marketing, or to spend, or to even attrite. The result is a prioritzed list of customers/prospects in terms of who should you focus on contacting first.

In other words, this segmentation tells you who to contact, not necessarily how to contact.

Now, with adequate past campaign data (both send and response), you can build models that will predict which particular tactics will work best, i.e. performance of different offers, media (direct mail vs email vs telephone), message frequency (propensity to respond on subsequent contacts), etc. So, within a segment, you can find further breakdowns of tactical segments.

IMHO, this is where the creative process needs to be aligned with the analytics. Predictive models can generate a prioritized list of customers/prospects for a particular marketing goal, and even break it down into tactical segments based on past marketing data.

With this, the "creative" team should have a much better insight to its audience, their preferences, and of course, the overall goal of the marketing campaign -- which should help crafting the most compelling message to the audience.

Sandeep Giri said...

We had a couple of other comments from some colleagues at work on this topic that are worth sharing:

I like his definition/analysis of “segmentation”

This multifaceted aspect of “segmentation” is why I get uncomfortable with prospects who want us to segment their data without a clear description of business purpose the results will be used for.

His point on being in a high churn group not translating into a message is spot on. IOW, being AtRisk identifies likely candidates for retention marketing but says nothing about how to communicate with them. > is well aware of this. The messaging in the Retention trial is primarily age/gender driven with some tweaking of message based on usage and/or tenure.


Yup, as I’ve said before “segmentation” = grouping, nothing more, and nothing more useful without additional intelligence around it. Picking a handful of universal segmentation schemes that are useful and possible with only (transactions and behavioral) data is difficult, but something I’d like to see baked with more thought towards what has worked in the past and the marketing implications for future applications. Every segmentation we provide should point clearly to marketing actions and carry these implications as part of our offering, rather than being a group that exists solely because we can slice it in the data.

Some quick examples of making segmentations actionable and attractive to a prospective customer:

* At-risk groups should be limited to three groups – high risk, low risk, neutral
* FM segments (frequency and monetary spend to date) should be aggregated in a different way – probably a matrix of 2 frequency levels vs. 2 monetary levels for:
-- HighF – HighM -> Reward and keep them happy
-- HighF – LowM -> Moderate up-sell effort
-- LowF – HighM -> Continue to market, cross-sell
-- LowF – LowM -> Drop from marketing after recency falls off

Le_Plan said...
This comment has been removed by the author.
Le_Plan said...

I have an blog entry regarding segmentation. I thought of sharing here.