The most advanced online & offline Attri­bu­tion for Retail, in Adobe Insight

My friend & col­league Derek Tan­gren recently wrote about whether you need an advanced attri­bu­tion model. As he pro­poses, some of the impor­tant con­sid­er­a­tions are What are you doing with your mar­ket­ing attri­bu­tion data and What will you do with an advanced model?

In addi­tion, the recent Adobe Dig­i­tal Index Report “How social stacks up: Why mar­keters aren’t giv­ing social the credit it deserves” made a very strong case for mov­ing beyond the long-used, long-maligned “last click” attri­bu­tion, illus­trat­ing the gap in value attrib­uted to social sim­ply between first and last click models.

I have spent much of the last few years of my career work­ing with advanced attri­bu­tion mod­el­ing with our retail & travel clients, as attri­bu­tion is one of the main rea­sons that most new clients are adding Adobe Insight to their analy­sis tool­box. As a result, I have some fairly strong opin­ions based in solid expe­ri­ence – and we’ve devel­oped some solid ser­vices and method­olo­gies around help­ing you under­stand and use your attri­bu­tion data. Many of those ideas were fur­ther val­i­dated by feed­back and direct con­ver­sa­tions with many dig­i­tal mar­keters as a result of sev­eral attribution-related pre­sen­ta­tions I was able to give at the recent Adobe Dig­i­tal Mar­ket­ing Sum­mit in Salt Lake City.

To under­stand the con­cept of attri­bu­tion, con­sider a typ­i­cal path to pur­chase. Let’s imag­ine a cus­tomer who sees a dis­play ad for Adobe Pho­to­shop, per­forms a search for “Pho­to­shop” a cou­ple of days later and clicks on anat­ural search result, per­forms a search for “Buy Pho­to­shop” the next day and clicks on a paid search result, and then the next day expe­ri­ences the Adobe Pho­to­shop page on Face­book, clicks through, and pur­chases Photoshop:

Attri­bu­tion seeks to pro­vide answers to which mar­ket­ing efforts had what share of influ­ence on the con­ver­sion. Many dig­i­tal mar­keters in retail are very famil­iar with attri­bu­tion of online clicks to online con­ver­sions, and most exclu­sively use “last click” attri­bu­tion, giv­ing full con­ver­sion credit to the very last click.

How about going beyond the click? And how about going beyond just the online conversion?

A key dif­fer­ence about Adobe Insight, as it stands among the rest of the Adobe Dig­i­tal Mar­ket­ing Suite, is its abil­ity to inte­grate event-level log data from prac­ti­cally any data source. Because of this, Insight is fre­quently used to com­bine not only Web traf­fic data (from Adobe Site­Cat­a­lyst, Adobe Insight Sen­sor, or other com­mer­cial or pro­pri­etary log­ging sys­tems), but also event-level data from store sales (POS) sys­tems and event-level data from mar­ket­ing ven­dors like ad servers (DART, Atlas, etc.) or email ser­vice providers (Respon­sys, Exact­Tar­get, Chee­tah­mail, etc.)

All Events in a Cus­tomer View in Insight

Unified Customer ProcessVia Adobe Gen­e­sis, many retail and travel mar­keters already inte­grate aggregate-level mar­ket­ing data with their dig­i­tal data in Site­Cat­a­lyst and the rest of the suite. But Insight takes this a step fur­ther, allow­ing you to cor­re­late pre­cise events within a sin­gle view of the cus­tomer and under­stand how each dis­play ad view, email send, direct mail piece – in short, every address­able mar­ket­ing touch point – influ­ences conversions.

In Insight, all of these indi­vid­ual events can be asso­ci­ated and grouped by Cus­tomer, then orga­nized by expe­ri­ence (some expe­ri­ences are Web vis­its; oth­ers are store pur­chases; oth­ers are call cen­ter ses­sions; oth­ers are mar­ket­ing ses­sions). All events are ordered by time, pre­sent­ing fast, multi-dimensional analy­sis on any level of the data.

These con­ver­sions don’t have to just be cart check­outs. Attri­bu­tion can be made to mul­ti­ple con­ver­sion points, both online and offline: retail store counter pur­chases, hotel check-ins, air­port kiosk events, rental car check-outs, B2B lead form sub­mis­sions & Webi­nar atten­dance. You name it. If it’s an address­able event, your attri­bu­tion can be con­fig­ured to it in Insight.

This is impor­tant when you con­sider that “con­ver­sion” for most types of busi­nesses — retail and oth­er­wise — hap­pens in both online and offline channels.

Offline & Online Conversion Funnel

Base­line Insight Attri­bu­tion Solution

Because we’ve found that many retail and travel mar­keters seek first to know what’s pos­si­ble, then to under­stand how it applies to their busi­ness, our phi­los­o­phy around attri­bu­tion mod­el­ing has “grown” to a point where we ini­ti­ate an Insight attri­bu­tion engage­ment by turn­ing on a set of pre-configured mod­els. We’ve devel­oped our best prac­tices around con­fig­ur­ing these few key mod­els, then spend­ing time with your data in them:

First Touch — Pretty sim­ple: The first mar­ket­ing touch gets the full credit for the conversion.

Last Touch — Also pretty straight­for­ward: The last mar­ket­ing touch gets the full credit for the conversion.

Even (Lin­ear) — Every mar­ket­ing touch within a defined period of time receives an equal share of credit for the conversion.

Starter/Player/Closer — The “Starter” (ini­ti­at­ing) touch receives a set per­cent of credit for the con­ver­sion, the “Closer” (last) touch receives a set per­cent of credit for the con­ver­sion, and the remain­ing “Player” touches each receive an equal share of the remain­ing credit for the conversion.

Latency Score — Every mar­ket­ing touch is scored with a numeric value reflect­ing how many days it fell prior to the next conversion.

Pathing — Lever­age sev­eral advanced visu­al­iza­tions in Adobe Insight, once all mar­ket­ing touches and con­ver­sions are orga­nized for analy­sis, to under­stand both the direct and indi­rect paths between cus­tomers’ expe­ri­ences with var­i­ous touch points.

Engag­ing: “Touches”, time frames, con­ver­sions & more

Note that I used the term “touches” a lot above, not nec­es­sar­ily clicks. That’s because, with impression-level, or other touch (email sends, email opens, etc.) data inte­grated in your Insight dataset, you can go beyond stan­dard Web “click only” attri­bu­tion and also give credit to all views/impressions.

When we kick off an attri­bu­tion engage­ment in Insight, we first ensure you’ve orga­nized all of the event data you want in a customer-centric dataset.

We then dis­cuss a few key questions:

  1. What to con­sider a mar­ket­ing touch (what logic to use to flag each of your event rows that rep­re­sent mar­ket­ing touches in the data)
  2. What to con­sider a con­ver­sion (is it an online order? a store pur­chase? a lead sub­mis­sion form? other major or micro conversions?)
  3. How far back to look for attribut­ing value to the mar­ket­ing touches (5 days? 15 days? 30 days?)
  4. Any excep­tions to con­sider (ignore branded search? ignore ban­ner impres­sions if they’re imme­di­ately fol­lowed by a click?)

Next, we turn on the base­line mod­els above.

Finally, we go ini­ti­ate a series of work­shops with you and your ana­lysts. Insight-focused busi­ness con­sul­tants and predictive/statistical con­sul­tants from Adobe Con­sult­ing work with you to under­stand the intri­ca­cies of your data, pro­vide ini­tial results report­ing and rec­om­men­da­tions based on the ini­tial base­line mod­els, and pro­vide rec­om­men­da­tions for fur­ther cus­tomiza­tion of the mod­els, before turn­ing the solu­tion over to you and your team to use.

Then what?

Once these mod­els are in place, you’re empow­ered as a mar­keter to have much deeper analy­sis on the impact of your dig­i­tal mar­ket­ing on your online & offline con­ver­sions. You can:

  • Gain a stronger under­stand­ing of how cer­tain mar­ket­ing chan­nels are “closers”, con­tribut­ing directly to con­ver­sion and other mar­ket­ing chan­nels are fur­ther up the mar­ket­ing influ­ence funnel
  • Test and under­stand the results of shifts in your dig­i­tal mar­ket­ing efforts
  • Per­form more informed test­ing of var­i­ous cre­atives in spe­cific place­ments, seek­ing to boost their effectiveness
  • Build upon these mod­els with fur­ther cus­tom mod­els in Insight, includ­ing mod­els informed by sta­tis­ti­cal and user engage­ment inputs

The real results are evi­dent in the poten­tial for more effi­cient mar­ket­ing efforts, avoid­ance of mar­ket­ing can­ni­bal­iza­tion, and direct impact on online & store con­ver­sions. And what mod­ern retail mar­keter wouldn’t want the poten­tial for those results?

Next week, my Insight peer Jeremy King will be shar­ing some more detail around how these mod­els are built and con­fig­ured in Insight for the exist­ing retail Insight archi­tec­ture geeks who might be read­ing. Stay tuned!