Artificial Intelligence can be an extremely useful tool for organizations looking to improve the effectiveness of their sales and marketing activities, and this can be seen in use cases from some of the big hitters like Visa, and Google.
At Elavon, we’re in the process of implementing AI to achieve these goals, and in this article, I want to demystify AI for sales enablement. I’ll share with you how we’ve done it, the results we’ve seen thus far, our learnings from the process, plus the insights we’ve gleaned and will work on in the future.
At Elavon, we're currently working through our sales enablement strategy. We've got a lot of work to be done - we've undertaken a journey and we're at the start of that journey.
However, there are many component parts in terms of being able to deliver productivity and productivity gains for our salespeople that I do want to be able to highlight because we're undertaking that journey.
I may be a bit of a dark horse in the context of sales enablement because I actually come from a data and analytics background. What I'm going to be talking about is our artificial intelligence program in the context of augmenting sales and marketing activities, and how we've deployed that at Elavon.
There are three key simple parts to my article:
- Firstly, why did we elect to go with an AI program? Why should anybody elect to go with an AI program?
- Secondly, where are we in terms of that journey?
- Thirdly, so what? If you implement a program of this nature, are there any real business results that you get from it?
- As a data and analytics person, I am going to finish this article with some numbers, to give a bit of a heads up.
A bit about me
My title is Head of Sales Enablement and Productivity at Elavon Europe. I've held that title for about eight or nine months - that’s not to say that we haven't been doing enablement activities at Elavon, I think the role itself is an opportunity to bring together a number of those existing activities and then to also start about some of the gaps that we have.
Whether it's content management or learning management, working in a lot more of an aligned and metro-sized way with marketing and L&D, bringing all of that together, there's a clear opportunity for us to do that a lot better at Elavon, and that's great.
In terms of my background, I've got my role divided into two parts. So principally from a technology or technical aspect part, I look after CRM and marketing automation. Then I also have within my team a commercial analytics division, they look after deal and business case development, as well as sales performance reporting, and of course, lastly, and most importantly, the AIPs in terms of our delivery.
I have a corporate career almost 20 years in length of delivering capabilities within financial service organizations, a good chunk of that at American Express, and then, over the last five years at Elavon.
I'd say the one key strand that's common across both those financial service companies is that both of those companies are typically more expensive in the marketplace - if you're holding an American Express card, you know that you can't actually spend with it anywhere.
The one key thing that I've learned over my 20 years is how do you effectively sell on value? Because you can't sell on price. The companies that I've worked for, we don't sell on price. So how do you effectively sell on value? On that basis, how do you set up the right sales operations, the right sales enablement, and of course the right data and analytics capability to help those organizations?
The bit that I really love and that really makes me tick is the right here right now. Increasingly, there is a convergence of data analytic capabilities against business operational data and I think the most representative example for me on that front is the world of CRM had been in many respects over the last few years dealing with a divorce to some extent from business intelligence.
That's coming together and coming together in a much more meaningful way and I think that's really exciting because what it's doing is it's bringing a number of those techniques into the business operational data space, and really beginning to leverage the power of data analytics for improved decision making.
What I think is really interesting in this space is it's not the business intelligence function that's drawing operational data to it, it's the other way around. So Salesforce's acquisition of Tableau is a really key move in that space of saying, "You know what, we've got all of this operational data, if you want to get better at making smart business decisions without operational data, at Salesforce we're going to classically buy in a few other companies that help our clients get better at doing that type of activity".
That's what makes me tick, now I want to talk a little bit about Elavon.
A bit about Elavon
We're a leader in global payments, we do payments processing. We have offices in 11 countries across the globe, though we're licensed to provide payment processing across 36 countries.
What that means is we provide other businesses with the ability to accept credit and debit card payments or any electronic type of payments, both in-store and online.
In the UK, our key competitors are WorldPay and Barclaycard, you naturally will have heard of them - amongst them they own about 75% of the processed volumes in UK market share.
At Elavon, we're ranked about number five in the UK, and we have about 8% of that market. So again, a lot of growth opportunity for us. Our capabilities in terms of the businesses that we serve, they extend from small to medium-sized businesses all the way through to multinational enterprises, although our footprint has traditionally been in the SMB segment.
The reason that's the case is because we've built up a lot of our operations and the footprint of our operations through partner organizations. These are external partners, other banking organizations, other financial institutions, etc, as you can on the image we've got 550 odd partners across the globe in terms of financial institutions that help us get our footprint out into market.
So we don't have much brand recognition, we're not a WorldPay, we're not a Barclaycard, and therefore, we rely on partner networks to be able to bring us into market. That's a key point and one I'll come back to at the end of the article, to give you a little bit of an additional flavor around that.
Why has Elavon invested in an AI program?
The deployment of this actually began in late 2018 although the story predates that by a couple of years. The story at its heart is really simple. Its a story of a company that had not been invested in for some time, got a rapid degree of investment through the course of 2016, increased its capacity, its sales organization almost overnight and in the process shot itself in the foot because it was somewhat wholly unprepared for that level of growth.
That's exactly what happened at Elavon, that's the story of Elavon from 2016 to 2017.
What happened at Elavon
Rapidly changing industry
It's a rapidly changing industry, it's consumers shifting their consumption patterns away from cash over to electronic payments - that's great, that's wonderful. Up until 2016, Elavon had been driving almost near double-digit growth year over year.
Historic strong growth attracted increased investments
Our parent company is US Bank and they wanted to capitalize on this, particularly in Europe, they invested in us heavily in 2016 and our sales force grew by nearly 25%, almost overnight.
What we didn't do was peripherally and proportionately invest in other parts of the organization so they could grow at the same pace, and what we didn't do is have a decent sales enablement strategy in place for us to be able to be smart about the way that we were growing as opposed to being reactive and probably our sales enablement strategy has been and continues to be a little bit reactive.
That's exactly what we did, we grew at a very fast clip overnight, got very excited about it, had a huge amount of capacity increase in terms of our sales organization, but in the process, lost track and lost sight of what was important to us. That was the main challenge.
Expectation of accelerated returns
Add in the additional pressure of targets, when you get an accelerated amount of investment, naturally along with that comes an accelerated expectation of returns. Put that in and that's your quarter pressure, if you've got that pressure coming through, what happens?
People lose focus, they're not supported, they're not provisioned, they're not empowered and two things came out the back end of this.
Slower sales productivity and higher customer attrition
We lost productivity on a bi-person ratio from a sales perspective and we also lost focus on some of the areas where we had been strong, which is retention of existing customers existing levels of customer service, and that led to a greater than normal degree of churn. Leaving 2017 our story was not a pretty picture.
Our total portfolio had been in contraction by about 3%, not a totally dire picture, because that was the book of customers had reduced by 3%. Revenue stayed flat against 2016 - not a totally dire picture, but certainly a picture that warranted a rethink in terms of where we were at and what we needed to do.
Investment followed by reverse performance - that's never a good story.
So do you need to have a cliffhanger moment of that nature, to be able to put forward a business case that allows for investment in productivity activities, and in this instance, in terms of an artificial intelligence program for improved productivity? I'd say no.
There are a number of household brand names that are utilizing artificial intelligence as one of the component parts within that technology stack to be able to improve productivity and to be able to drive that overall sales enablement journey.
What is sales enablement?
For me sales enablement is a case of, how do we get the appropriately skilled person into the appropriate interaction set? So the right person with the right message, having the right conversations against the right target at the right point in time, how do we accelerate that? How do we make that happen? And hopefully that resonates across everybody.
Maybe it's an overly simplified definition, it's probably not as pithy as some of the definitions that you've seen in the past, but actually that's what we're trying to do. The good thing about this ethos is that it translates across in person, digital, virtual - all of these different channels.
The benefit of AI
Actually, when you think about all of those different channels and the complexity that comes across a multi-channel environment, it's natural that capabilities such as artificial intelligence provide a really strong value-added.
Ultimately, all AI is doing is helping support improved decision making. It's just taking away some of the factors and some of the challenges and some of the decisions that we potentially have to make. It's just improving some of that.
What have we elected to do with AI at Elavon?
If you take a look at the diagram, what we’ve opted to do is more towards the bottom. How do we utilize AI to really help intelligent segmentation and in the process of driving intelligent segmentation, get the right deals in front of the right salespeople?
What we're trying to do here is really link back to a really simple Pareto principle. If I've got my reps out there undertaking sales or account management activities, how do I make sure that 80% of their effort goes against the 20% of opportunities that are the genuine sweet spot?
Don't equitably spend your energy against all hundred percent of deals that you've got your vision over - don't do that. Try and identify the 20% where you know you're going to get a higher conversion ratio, you're going to get more money, and actually you're going to hit that sweet spot and that bang for buck at a much more accelerated rate.
How do we do that effectively? At Elavon that's the job of AI is to be able to provide a view of that 20% and how do we get to that?
AI is founded on data
I want to go through a generalized simplistic model of artificial intelligence, and demystify AI. In terms of a generalistic model, AI is obviously founded on data - there has to be a data input into your artificial intelligence model. Something needs to happen there with information going in.
The process of that information, and actually right at the start of the workflow, that information needs to be organized, ideally into some type of desirable versus undesirable outcome. The more binary in many respects, the better.
AI makes predictions
The reason being that once that information then goes into stage two, which is being plugged into an artificial intelligence environment, what starts to happen there is the environment gets to learn about the information and starts to then, in a very predictable way put out what's a positive outcome versus what's a negative outcome.
It's those outcomes that help to then start generate predictive frameworks that can be utilized on an ongoing basis.
Utilize the predictions
The third part here is taking current data and operationalizing that data through the AI environment, getting a predicted outcome and then utilizing that predicted outcome because that outcome's gonna say, "Here's the good stuff and here's the not so good stuff, do something useful with the good stuff".
When you feed that back in, and you get an automation of the feed, and you get some adjustments to the way the system learns about the feedback of the data and starts to organize that data in a way that's meaningful for the feedback loop, and again, self-adjusting feedback loops, it's really machine learning.
When you get into the space where the system starts to adjust its own algorithms, almost like a human neural brain pathway, that's deep learning. That, of course, is your kind of Arnold Schwarzenegger Terminator end of the spectrum, which is fantastic because, at the end of the day, we still know where the on-off switch is.
All of this is great, and we're not utilizing it to any extent of that level of sophistication. But you can see it of course across a number of day-to-day applications whether it's being able to observe your consumption habits and in the process serve up recommendations for you, whether it's speech analytics and again being able to understand and recognize whether a certain word or a certain vocabulary is hitting the absolute expression or not, that whole process of information being treated in a slightly binary manner to be able to generate predictive outcomes so that the system can learn from that.
AI in action: Google
Of course, Google's or AlphaGo's recent venture into deep learning and the very decisive beating of a Go World Champion is again a very compelling case in terms of how deep learning artificial intelligence has become much more prevalent.
AI in action: Visa
In the B2B space, Visa are using AI for fraud checking which is great and again, this is another example of a company that's utilizing artificial intelligence and utilizing that as a mechanism by which to drive value to its partner network.
Visa has obviously got all of the information and the transactions flowing through its scheme's platform, but they're utilizing that information as a mechanism by which to go back to the issuing banks and give them a view around how they can better handle fraudulent transactions.
AI in action: Salesforce
Of course, our beloved Salesforce have Einstein embedded into all of their products, and what we're trying to do here with a number of the capabilities, I think Salesforce is certainly moving forward at a pace in terms of how do you build that in and get that wonderful view of operational data and making that very meaningful.
AI in action: Elavon
At Elavon what we're doing, we've got historic sales and retention data, we've identified wins, the good stuff versus losses, the not so good stuff, we've pushed that through the system, the system understands, we've then gone right let's get our current list of customers, for example, our existing book of customers, let's take our SMB portfolio, we're doing this once a month, pass it through the system so that the system can say to us, "This is the good stuff and this is the not so good stuff, focus your effort and intention on the good stuff and then take some campaign based activity".
I would love to say that we're at that end of the model, we're not, we're at the point where anything that comes through we then feedback and at least we're doing some feedback in terms of understanding value from the overall program.
Build it or buy it?
Did we build it or did we buy it? We actually went with a company called Lattice.
Forrester, in terms of their most recent wave report on customer data platforms, puts them right at the top in terms of the leading space. They have got a huge platform and a huge amount of data. They've recently been purchased by Dun and Bradstreet, so they have access to a significant data set.
Lattice get their intent signals from sixth sense, they're able to really get into that space of web scraping and begin to identify where certain businesses are taking certain actions, provide that intent data over to us so that we can again start to react against that, and it's one of the capabilities we haven't explored yet, but we will do.
You'll see another household name right at the bottom of the image above, Dell. The reason I call this one out is, in many respects, that's exactly what we're trying to do. The success that Dell has had with Lattice in terms of reducing the number of deals that have to go out to salespeople, but then improving the productivity per deal basis, that's really what we're trying to do as well.
So it's fantastic that Lattice have achieved that, they're a really good partner, they've been really engaged with us. We're probably a smaller company against the roster of clients they have and yet they take a very well-proven Customer Success management model and that's what they're working on with us which is great.
What I will say in terms of the build it or buy it is a couple of things.
Build it or buy it: cost
For us, outsourcing to an external partner such as Lattice is probably about a third cheaper than trying to build it. If you've got data engineers, an Amazon Web service platform, data scientists, etc, even a small team, it's actually quite expensive in terms of being able to resource that out.
That's one of the first things and that's a pro for us in terms of us being a smaller organization and needing the flexibility and the dynamism and being naturally constrained by cost.
Build it or buy it: awareness
The con is that we don't necessarily really know what's happening. It's their black box, we feed it with data, it gives us some output, it gives us a lot of statistical information around that output. So we've got a high level of confidence in terms of we're getting good stuff from it.
We don't necessarily know what's happening under the bonnet, and that's the challenge.
What I will say and one of the keys learnings that I'll share is I do think, having been to a number of data and analytic conferences, having talked to a number of peers, I do think for us as a company, Elavon is a mid-sized company, it's the right strategy.
Let's start with an outsourced partner, let's start small, let's prove the use case and actually if we ever become a Citibank in years to come, wonderful, we'll then go and hire a whole carter of data scientists and what have you. But we're not there, we're not that, so an outsourced partner is what we need for the moment.
AI at Elavon
The below image comes from marketing and its a complex picture but I do think what this expresses, and the key point that I wanted to bring out on this, is in the action in the workflow, Lattice and AI are at the bottom of the workflow in terms of that technology stack and where it sits on the streaming of effort.
The bottom of this, the downstream part that it plays, is deliberate. Again, that's appropriate for us for now, we could have Lattice upstream, right at the top where we've got different types of lead generation activities, yes, absolutely Lattice and in fact, any type of AI program can play a very valuable role there.
In time, we'll get it to play that role, but right now, Lattice plays a downstream role where it actually looks at the information that we've got, let's push it through, let's score it, let's understand it and then let's make deliberate actions on the basis of that.
Crawl, walk, run
We're trying not to bite off more than we can chew. That was clearly one of the learnings from trying to accelerate our sales organization and I'm allowed to, on a quite regular basis use the expression, 'crawl, walk, run', and nobody laughs at me because that's exactly what we're trying to do.
We're trying to crawl at this stage and when we get to a point where we feel more comfortable with our competence in this technology we'll start to accelerate some of it.
Conversion results: new business pilot programme
l'm going to close off the article, as promised, by talking through the numbers. The below image looks at our prospective sales, we utilized Lattice in Q1 to be able to run effectively a prospective sales program.
What we did was we took about 1000 recycled leads - these were lost deals that were sitting in Salesforce, and then recycled them back into Salesforce having scored them through Lattice. Lattice, as an AI environment, is output-oriented, so tries to simplify the way that you see this output.
High fit vs low fit
Lattice has a very simple scoring spectrum going from A to D, where A is a high probability of good stuff happening, and D is low probability of good stuff happening. So in this context, I've simplified it, high fit means those deals that Lattice have told us would be at the strongest probability of having a high conversion rate. Low fit is those deals that would have a low probability of an accelerated conversion rate.
AKA, the high fit stuff is good for us.
We deliberately hid the scores from our sales agents when we put them into the system, the reason being that we wanted to do a blind test on the AI environment.
That was really important for us because we wanted to isolate the causality of AI, as opposed to it being blended into - it's AI plus it's sales skills, plus it's time, plus it's all of this good stuff. We just wanted to isolate AI and the impact of AI.
Skew leads to motivate
We took a deliberate decision to also skew the mix of leads in favor of high fit. The reason we did that is that we wanted sales agents who knew that they were participating in a pilot program to feel that they were getting more positive interactions on a more regular basis.
They didn't know if it was a high fit or a low fit, so they had to obviously equitably put their energy against each deal and every single interaction. But we did want them to feel like they were making progress and it was a positive pilot program to be supporting.
This happened in Q1, as everybody knows Q1 is a busy time of year, so asking our sales agents to then do this on top of all the Q1 stuff, it needed to have a little bit of a motivating factor to it.
There's also a really compelling aspect relating to data quality. What Lattice had classed as high fit - and the benchmark there being low fit - turned out to be genuinely a much better conversation.
Those leads were being qualified and in the Salesforce vernacular 'conversions over to opportunities' - there was a much higher percentage of conversions over to opportunities from leads that were in the high fit range, as opposed to those that were in the low fit range, in fact, two and a half times more, which is certainly very compelling.
In terms of closed one, again, less impact here, because in this space, you're really talking about the sales skill set. So an AI program is going to do less here, within the context that we're using in until you get into coaching tools and mechanisms and all of that good stuff.
But still, some value because we're clearly passing through deals that are qualifying and accelerating and moving through the pipeline at a much more accelerated rate.
So fantastic that we're getting a conversion rate coming through there as well. I certainly felt very comfortable and honestly very, very pleased with this. You've got to remember as well that this is fundamentally a phone-based cold call campaign against customers and deals that we had spoken to previously that had previously said, "No thanks, Elavon, you're too expensive".
The fact that we're going back to them and having a conversation, and the fact that AI's told us, "Spend your time there", and that coming through in terms of proof of concept, was really valuable.
Conversion results: pre-emptive retention
The second use case that we used, and we've been doing much more of this actually in recently, is using artificial intelligence and Lattice for pre-emptive conversation. What we're trying to do here is get ahead of a potential customer being at risk of cancellation.
Can the AI environment, once it's been trained, give us a call list that allows us to have a conversation with a customer potentially two to three months before they're actually going to cancel?
This is certainly an interesting use case, and again, all of the stuff that's happening down in saved cases, this really is ultimately the skill set of the customer account agent that's on the phone, because this is an elegant conversation.
You're getting on the phone with a customer that you don't necessarily know if they're really at risk of canceling, you're taking some confidence and faith that an artificial intelligence system has said, "Go have a conversation this customer, they may be at risk of canceling", you need to have a conversation where you get into potentially a value-based conversation.
You can't lead with a value-based conversation because you don't want to reduce your margins or give away too much right from the outset, what you somehow want to probe and understand, is this customer genuinely at risk of canceling, yes or no?
If they truly are at risk of canceling, what type of proposition can we go back with so that we can save them, drive the additional loyalty, etc?
We did this previously at Elavon utilizing internal-only data, and with a very rudimentary statistical analysis model, that was the old way of trying to do at-risk cases, versus the new way, which is, of course, the Lattice score.
Less false positives
What we were finding is that there were a lot of false positives in terms of the old way of doing things. Actually, about 84% of the cases that we had flagged out to our account execs weren't really at risk, they'd just come to the system as being 'looks like this customer could be at risk of canceling' but weren't really at risk.
We still have a huge distance of improvement to travel, but in terms of Lattice, at least now we're getting to a point where it's near enough 60%. So six out of 10 are still not really valid cases, but it's a significant improvement versus where we were. Again, it's a much better use of account managers’ time when they're having that conversation.
When you get into saved cases, these are businesses that are still, from those that were genuine valid at risk, businesses that continue to transact three months down the line, again, the improvement there is very compelling.
Does this stack up at a macro level? Not yet. Are we seeing any significant improvement in terms of our overall churn rate at a European level? Not yet.
I think by 2019, having implemented this, having gotten proof of concept, a few results, I think that's really compelling. The big challenge now is how do we scale this up? How do we make this work in a much more effective way? Get people bought into it? Drive usage of it? And really accelerate the program?
I'd say that's one of the principal learnings that we have.
AI enables two of four pillars
Firstly, in terms of where our AI program has added value across that sort of ethos of right person, right message, right opportunity at the right time, it's really serving those final two pillars of sales enablement, which is fantastic.
Artificial intelligence, machine learning adoption, that is a classic sales operations issue.
There are compelling results here. There's clearly a 'what's in it for me?' message that can be taken over to our sales and account management agents, but in a classic way, they're busy, they've got other things to do, they don't necessarily want to engage with this program.
There's a lot of selling internally that we need to do to be able to make it happen and again, we're coming back to that push and pull mechanism of what carrots do we need to use? And of course, critically, potentially what sticks do we need to use as well?
We need to get value from this, yes, it's a third cheaper going at it with an outsource company as opposed to building it internally, but it still costs money. I've got to start proving real value from this on a scalable basis, and to do that, I do definitely need interactions across all of the areas of the company but in particular, with sales and account management.
Multi-channel communications strategy
We do need to get into a multi-channel communications strategy. Everything that I've talked about here is really phone-based. We're going through the program of building out our marketing technology as well.
I'd say as we get into the second half of next year, I think we'll be in a much more solid position where we're looking at a multi-channel strategy that helps drive a number of those interaction points.
One-to-many versus one-to-one is talked about a lot in our industry - we've started with the one-to-one interactions, what we really need to do is actually develop our technology stack to facilitate the one-to-many and actually flip between the one-to-many and the one to one as needed through the process of these use cases.
Partner value proposition
Finally, as we saw with Visa, and again within the context of our partner network, if we can do this and we've started to experiment with this with retail merchant services, RMS, we're a big provider for micro and small businesses in the UK, if we can start to drive value into their organization around AI and how AI can help us help them, then that helps to develop our partner proposition as well. That's good for us, good for them.