Mar­keters use attri­bu­tion mod­els to fig­ure out what fac­tors cause a sale. Mar­keters want to know how much of a pur­chase is attrib­ut­able to a cer­tain action. Attri­bu­tion mod­els build off of rel­e­vant curiosity—i.e., what actions led up to a sale. For mar­keters, it pays to know which tac­tics are push­ing sales and which are stag­nant. It goes back to the marketer’s com­mit­ment never to waste a customer’s time. If mar­keters can fig­ure out which schemes are push­ing sales, they can focus their resources there. Thus, they can save time, effort, and money for them­selves and their cus­tomers. A good mar­keter should know that the mar­ket is not sta­tic and needs dynamic adjust­ment. That’s why it’s impor­tant to crit­i­cally ana­lyze attri­bu­tion mod­els in order to seek per­sis­tent improvement.

Attri­bu­tion Models

There are many attri­bu­tion mod­els; a lot of them rely on first touch and its vari­ants. The first-touch attri­bu­tion model attrib­utes a sale to the first action a con­sumer takes along the road to pur­chase. So, if a con­sumer wants to buy a lap­top and the first thing he or she does is a Google search for a spe­cific model, this search will count as the first touch. In a stan­dard first-touch attri­bu­tion model, the mar­keter attrib­utes the entire sales process to this Google search.

Obvi­ously, there may be an issue with a stan­dard first-touch model. It doesn’t make sense to attribute an entire pur­chase to the first touch, espe­cially if it’s a long-term pur­chase. When con­sumers spend weeks or months search­ing for, research­ing, and com­par­ing prod­ucts, the first touch is nearly irrel­e­vant at the time of pur­chase. Thus, mar­keters devel­oped the NFS-30 attri­bu­tion model. Instead of a stan­dard first-touch model, NFS-30 attrib­utes the pur­chase to the first touch within the 30 days imme­di­ately prior to the sale. This model relies more on recent devel­op­ments than orig­i­nal inter­ac­tions. Mar­keters use it for long-term, big-ticket purchases.

The Prob­lem

Even with the adjusted NFS-30 attri­bu­tion model, there is still a prob­lem. Attri­bu­tion mod­els only give mar­keters one side of the story. They posit that it is rea­son­able to attribute an entire sale to one action. Attri­bu­tion mod­els are some­what use­ful, but they don’t tell the entire sales story.

To uti­lize infor­ma­tion most effec­tively, mar­keters can­not gen­er­al­ize and attribute sales to sin­gle actions. They need to gather infor­ma­tion about the jour­ney through­out the sales pipeline, not just at one point. The prob­lem with attri­bu­tion mod­els is that they do not take into account all avail­able con­sumer infor­ma­tion. They are waste­ful, and mar­keters should never waste resources.

Mar­keters who gather sales infor­ma­tion in attri­bu­tion mod­els are much like casual bas­ket­ball fans who look at the stan­dard stats in the box score after an NBA game is over. These fans can see who scored the most points, who grabbed the most rebounds, who blocked the most shots, etc. and can there­fore make decent con­jec­tures about which play­ers were the most valuable.

There is a prob­lem with such casual rea­son­ing. Although fans will be able to see the sta­tis­tics lead­ers and make okay pro­jec­tions about who con­tributed most to the game, they can’t know how much each player con­tributed just by look­ing at stan­dard sta­tis­tics. This approach sees only the sta­tis­tics that get marked on the board.

The Solu­tion

Stan­dard sta­tis­tics only pro­vide part of the story about player con­tri­bu­tions. One would think that it’d be nec­es­sary to watch the game to real­ize who the biggest con­trib­u­tors were. But there’s a secret tool for under­stand­ing who con­tributed most to the game. The plus-minus tool doesn’t only give point totals; it shows the point dif­fer­en­tial for each player dur­ing his time on the floor. So if Lebron James and the Heat went on a 10 to 2 run dur­ing his time on the floor, Lebron’s plus-minus would be +8. An oppos­ing player who spent the same exact time on the floor would have a –8 plus-minus.

Plus-minus shows the fan what each player con­tributed beyond basic sta­tis­tics. It takes defense into account with­out rely­ing merely on attrib­ut­able actions like blocks and steals. It offers a fuller pic­ture of how each player affects the game. Plus-minus uses a lot of the avail­able infor­ma­tion to offer a clearer pic­ture of each player’s effec­tive­ness and game con­tri­bu­tions. The fan who ana­lyzes the game with plus-minus is like the mar­keter who seeks con­tri­bu­tion mod­els instead of attri­bu­tion models.

Con­tri­bu­tion Models

Instead of attribut­ing pur­chases to sin­gle actions, con­tri­bu­tion mod­els hope to uncover how each action con­tributes to a sin­gle sale. A con­tri­bu­tion model tries to fig­ure out, what do par­tic­u­lar mar­ket­ing activ­i­ties con­tribute to a cer­tain sale? These mod­els weigh every action simul­ta­ne­ously, to fig­ure out how each action con­tributes to the ulti­mate sale. The most effec­tive con­tri­bu­tion mod­els can break down a sale by con­tri­bu­tion. Maybe ban­ners accounted for 20% of the sale, organic searches for 40%, paid search for 30%, and direct web­site vis­its for 10%.

Con­tri­bu­tion mod­els look at all the leads, ana­lyze pat­terns, and try to fig­ure out which com­bi­na­tions and pat­terns most con­tribute to sales. Instead of fig­ur­ing out which par­tic­u­lar action leads to a sale (as attri­bu­tion mod­els do), con­tri­bu­tion mod­els fig­ure out which com­bi­na­tion of actions best leads to a sale. Con­tri­bu­tion mod­els put the pieces of the sales puz­zle together. Attri­bu­tion mod­els try to fig­ure out what the puz­zle is with­out first com­bin­ing its pieces.

You notice a lot about con­sumers’ pur­chas­ing process with con­tri­bu­tion mod­els. You can see that the deci­sions to make pur­chases are not single-faceted, but that mar­ket­ing schemes in tan­dem con­tribute to sales. The attri­bu­tion model asks the ques­tion, what action led to this pur­chase? But that’s the wrong ques­tion. The con­tri­bu­tion model takes into account other fac­tors by ask­ing the right ques­tion: what actions con­tributed to this purchase?

Effec­tive Utility

Con­tri­bu­tion mod­els effec­tively use more of the avail­able infor­ma­tion than attri­bu­tion mod­els. That’s the point. Mar­keters should use all of the infor­ma­tion they have so that they can effec­tively mar­ket prod­ucts to the right peo­ple at the right time. Attri­bu­tion is not com­pletely inef­fec­tive, but con­tri­bu­tion is more effec­tive. On a small scale, attri­bu­tion and con­tri­bu­tion are nearly iden­ti­cal. Once mar­keters view more sig­nif­i­cant pur­chases, the con­tri­bu­tion model proves its worth.

Peo­ple often choose the attri­bu­tion model because it’s eas­ier. But the con­tri­bu­tion model offers deeper ana­lyt­i­cal thought, bet­ter use of infor­ma­tion, and a fuller pic­ture of the con­sumer. If you want to use your (and the con­sumers’) resources most effec­tively, then ask the right ques­tions and use con­tri­bu­tion mod­els to max­i­mize the use of pur­chase information.