Optimising FSI Customer Experiences, Part 3: Machine Learning

Today’s cus­tomers are hyper­con­nect­ed. They expect seam­less jour­neys across every dig­i­tal and phys­i­cal touch­point, tai­lored around their indi­vid­ual needs and goals. This means mar­keters need to lever­age huge vol­umes of data to deliv­er increas­ing­ly per­son­alised experiences.

Over the course of this series on the cus­tomer expe­ri­ence in the finan­cial ser­vices indus­try (FSI), we’ve been explor­ing the com­po­nents of deliv­er­ing great expe­ri­ences to our cus­tomers. In the first arti­cle, we spoke about the explo­sion of cus­tomer and trans­ac­tion­al data. In the sec­ond arti­cle, we exam­ined the need to stream­line and deliv­er more con­tent than ever before.

As we saw in those ear­li­er pieces, it’s impos­si­ble to man­u­al­ly man­age the amount of con­tent nec­es­sary for all those expe­ri­ences. The only way you’ll be able to deliv­er prop­er con­tent veloc­i­ty, at scale, with indi­vid­ual mes­sag­ing, is with the help of machine learning.

In a recent Adobe sur­vey ask­ing FSI mar­keters which areas seem to offer the most excit­ing prospects for 2020, the major­i­ty agreed that the num­ber-one most excit­ing prospect is using arti­fi­cial intel­li­gence (AI) to dri­ve cam­paigns and expe­ri­ences. IDC research found that Euro­pean busi­ness spend­ing on AI has increased by 40 per­cent since 2016, while even more orga­ni­za­tions are cur­rent­ly plan­ning to increase their spend­ing in this area, with the goal of improv­ing cus­tomer engagement.

This is because AI is no longer just an oppor­tu­ni­ty. It’s a require­ment for any organ­i­sa­tion that aims to stay con­nect­ed with its cus­tomers in the com­ing decade. But this doesn’t mean the marketer’s role is becom­ing obso­lete. On the con­trary, machine learn­ing is designed to han­dle the “heavy lift­ing” of rote, repet­i­tive tasks (such as check­ing and tag­ging thou­sands of images), leav­ing the mar­keter free to pro­vide more of the unique human touch­es that make a cam­paign brilliant.

There are many areas that AI and ML can help and sup­port a mar­keter in deliv­er­ing the opti­mal expe­ri­ence. In this blog, I am going to focus on only a few areas such as anom­aly detec­tion, con­tri­bu­tion analy­sis and con­tent per­son­al­i­sa­tion. Oth­er areas like sub­ject line opti­miza­tion, media mix mod­el­ling and rec­om­men­da­tions can be just as effective.

So, how do you use the pre­dic­tive capa­bil­i­ty of AI and ML to deliv­er bet­ter cus­tomer expe­ri­ences? It all starts when you look beyond your basic ana­lyt­ics and begin to search for action­able anom­alies in the data.

A predictive marketing workflow

The first step in an AI-cen­tric mar­ket­ing approach is to under­stand why your ana­lyt­ics fluc­tu­ate at spe­cif­ic times. For exam­ple, you might know your aver­age con­ver­sion rate and aver­age num­ber of dai­ly vis­i­tors to your web­site, which makes it very easy to get an aver­age and upper and low­er bound, but you might not know when or why those rates shift.

This is where anom­aly detec­tion comes into play. You can use machine learn­ing to look at the his­tor­i­cal data and recog­nise the trends, such as upward or down­ward fluc­tu­a­tions accord­ing to the day of the week or time of day. Once you’ve estab­lished those trends, you can use anom­aly detec­tion to tell you when some­thing out of the ordi­nary has happened.

The detec­tion of the anom­aly leads to the sec­ond com­po­nent of this workflow—contribution analy­sis. Say, for exam­ple, you’ve seen a spike in traf­fic and you want to know what caused it. This task would ordi­nar­i­ly require your ana­lysts to scour hun­dreds of reports. It may take them days to pin­point the cause of the spike. But with AI and machine learn­ing, you can search through many fac­tors that may cor­re­late with that spike, and quick­ly zero in on the most like­ly root cause.

Proactive content personalisation

Now that you have the abil­i­ty to detect traf­fic anom­alies and iden­ti­fy their caus­es, you’re well-equipped to per­form the next lay­er of analy­sis to under­stand why cer­tain audi­ence seg­ments are drawn to cer­tain touch­points at cer­tain times.

This is where propen­si­ty mod­el­ling comes in. This machine learn­ing tech­nique auto­mat­i­cal­ly fig­ures out which fac­tors engage or con­vert cer­tain audi­ence seg­ments at cer­tain times. These fac­tors may be spe­cif­ic types of con­tent, spe­cif­ic mes­sages or calls to action (CTAs), or oth­er fac­tors that vary through­out your A/B tests.

The final part of the pre­dic­tive mar­ket work­flow is acti­vat­ing these new intel­li­gent seg­ments with the right con­tent. If you’ve got 10 dif­fer­ent tem­plates for your home page ban­ner, 1,000 dif­fer­ent back­ground images, and 15 poten­tial mes­sages, that adds up to an enor­mous num­ber of poten­tial com­bi­na­tions of image, tem­plate, and mes­sage. There’s no way your design team could man­u­al­ly assem­ble all those final pieces for each indi­vid­ual customer—certainly not in real time. Nor could your Ana­lyt­ics team iden­ti­fy every poten­tial audi­ence seg­ment – espe­cial­ly when it is a seg­ment of one!

But AI can han­dle all that heavy lift­ing for you. Now that you know exact­ly what each audi­ence seg­ment responds to, you can use AI to gen­er­ate thou­sands of per­son­alised com­bi­na­tions of image, mes­sage, and CTA based on quan­ti­ta­tive analy­ses of all your pre­vi­ous inter­ac­tions with that par­tic­u­lar seg­ment. You’ll always deliv­er just the right con­tent and mes­sage for each spe­cif­ic visitor.

In fact, this con­tent per­son­al­i­sa­tion brings you back full cir­cle. You’ve gone from anom­aly detec­tion to con­tri­bu­tion analy­sis to propen­si­ty mod­el­ling to acti­vat­ing the audi­ence, and now, as you serve per­son­alised con­tent to each audi­ence seg­ment, you can con­tin­ue to gath­er data on their reac­tions, and use that data to fur­ther sharp­en your con­tent for even stronger engage­ment in the future. The next step is to deliv­er those flu­id expe­ri­ences across all chan­nels. That’s what we’ll be dis­cussing in the next instal­ment of this series. See you there!