Skip to main content.

Transcript: Harnessing the power of AI to improve CRM and drive business value

Read the webinar transcript below

Jane_BigData_Hero Jane_BigData_Hero



Jane Moran:

Really happy to be here today. I am a huge believer in data and the power of AI and excited to be speaking on this topic. It's interesting because I started off as a data scientist and just looking at how this space has evolved in the last 15 years or so. When I was starting, we would do data science by starting with a blank notebook page, and we'd just start writing the algorithm and whatever our programming language of choice was.


And these days the algorithms are oftentimes already designed and already part of a package of whatever your favorite language is to be dropped in. And the challenge has become the full deployment of these solutions. And it's really exciting for me to see them taking off at scale in the way that they are.


And so what I'm going to talk you all about today is not only the types of AI solutions that we're bringing into the business and how we're making sure that they're really driving business value, but some of the challenges above and beyond just the technical implementation. How to get the organization to really accept some of these solutions, how to make sure that you're seeing the value for all the effort that is being put into the data and AI space.


So just to introduce the types of data sets that we're working with, I work for AIR MILES, one of the most recognized loyalty programs in Canada. We have nearly 11 million collector accounts representing two thirds of the Canadian households. We're adding just shy of a million new collectors every year, 750,000, $600 million in rewards value and cards are being swiped a thousand times a minute with rewards being redeemed every two seconds.


I like to show this slide just to get a sense of the scale. It's been both the blessing and the curse of working with a large loyalty program like this is that we get all this data, there's very rich transactional data that's associated with the activity that we're seeing AIR MILES collectors engage in. Whether they're shopping for groceries and swiping the card or it's LCBO or if it's fuel purchases. There's a volume of data there that we're getting that is very, very rich, but it also means that as we're employing solutions that we want to launch into the program, we're always doing it at scale and we're bringing it up to the million accounts a thousand times a minute often in real-time, which is a real interesting technical challenge but also super rewarding when we get it right.


The primary way that we're using AI right now is in CRM personalization. So this is making sure that we're managing at almost an individual level what people want to do as their next best action to fuel their purchasing behavior, to make sure that campaign investments are getting good return on investment. And that there's engagement not only with the Air Miles loyalty program, but even more importantly with the retailers that we're working with. So we're working with Sobeys and Metro as big grocery stores, we're working with Shell, we're working with Staples, we're working with LCBO, they're paying for their marketing execution, they want to make sure that their money is going in the best place.


So what we've done is we've created systems of personalized offers. This is as simple as product recommendation engines. If we know that every time you go grocery shopping, you're going to buy chips. Then we might want to give you a chips offer. If you're more established and you've been buying chips for a long time, you're engaged in those offers and we're trying to stretch you a little bit. Maybe we'll try to get you into something a little bit different. We'll add on tortillas with salsa for example, something like that.


So it's the thank you for visiting, reinforcing loyalty, it's stretch opportunities. It's trying to get you out of the snack food category and recommending that maybe you're going to need toothpaste or antacids or whatever it is that goes along with those. All of those different types of campaigns being done at a personalized level to manage the ongoing relationship between a consumer and a retailer.


But there's another type that we do as well, which is a stretch offer. Those are also being reinforced by AI in a different way. The stretch offers will say for example, that if we see that you're normally spending $75 every time you're going grocery shopping, we might try to stretch you to $80. We might stretch you to $85 or maybe even $90. And the amount of stretch is going to be complimented by an award or a bonus offer. So it might be a five mile incentive to get you to increase by $5 or a 20 mile incentive to get you to go up $10.


That whole algorithm is being reinforced through reinforcement learning. How often is the campaign being responded to based on the amount of stretch and the associated incentive, and it's being tuned based on the behaviors that we're seeing at an individual level. What this is really bringing to life is that we're getting better campaign ROI in two different ways.


First of all, we're able to target more precisely. We know which types of consumers are most likely to respond to each different type of campaign. And so we're reducing the email deployment or the app based or the digital advertising or social, whatever channel we're using. We're reducing the cost to deploy because we're getting better at targeting those who are going to respond. We're also making sure that once they are responding, that we're not giving too little or too much, that we're balancing the need to generate increased sales but to do it in a cost effective way.


Finally, less time and effort is huge. The more we can automate a lot of these decisions, the faster we can go. We can start to deploy millions of different campaigns or ongoing targeted offers at scale with less internal effort, which is important for the ongoing running of our business and those that we support. So lots of really good stuff in here but it's not easy to do it. It's not easy to deploy.


This is just an example of what this looks like when it comes to life. You're seeing here get 30 bonus miles coming up on your app. It's an app based individualized for that deployment. These are the types of offers that are actually coming to life for our Air Miles collectors in the market. And we are seeing real results. This is important to remember that these solutions now are producing real results at scale. We're seeing marketing campaign ROI increased by up to 240%. This is a combination of better targeting and better knowledge of how to balance those offers so that we're generating the maximal return on them.


And finally, we're seeing email open rates that are giving us good hope that what we're doing is targeting the right people. We're getting a 43% improvement in email open rates, which is meaning that we're delivering relevant messages and that we're maintaining positive momentum for our brands and the partner brands that we're working with. So that's what we're doing.


I want to spend the rest of my time talking about the challenges of actually doing this. And the fact of the matter is that it's not all just technical. Certainly there is a lot of technical infrastructure that is required to do this. We've migrated most of our database into the cloud. We're automating ETL processes. We are targeting the ETL process, the data management, the preprocessing, the deployment through the channels and then reinforcement learning based on the results is all flowing together, which is a difficult technical challenge.


But the more important challenge that we have been facing is how to really operationalize this at scale within the business. And the difficult challenge that we overcome is this chasm that often exists between our technical teams, our data science teams and our IT teams and the teams that are tasked with doing the actual business deployment and the operations. These are our account teams, our marketing teams, anyone who's actually providing the business service on the other end.


And what we've found as we've started to spin up some of these solutions is first of all lack of comfort with technical terms. The data scientists will go in and they'll want to talk about all the features of their algorithms and it'll be over the heads of most of the users who are trying to actually go out and sell these things. Sometimes it's risk aversion within the business, that the business understands that you can do this at scale but are they really willing to put all their eggs in this basket? Maybe, maybe not.


Sometimes in development, we're seeing difficulty in estimating timelines and costs. The newer the thing is that we're trying to do, the more difficult it is to estimate delivery timelines, because you've never done it before. Same thing in building business requirements, before you really know what it is that you want to do, how can you create detailed requirements?


Finally, the last few prioritizing your needs, managing multiple stakeholders, establishing change management are all some of the barriers that we've encountered, not all with the same project luckily. When things are going well, it's one of the things on this list is really the big barrier. When things are going badly, there's four or five of them happening simultaneously that we're trying to overcome.


One of the most common conversations that I have with my IT teams or data science teams, is that they say, "Look, I created a really good solution. I handed it off to the sales team or the marketing team. If they don't know what to do with it, then that's not my problem." And my constant talk track is, "No, that actually is your problem. That means that you haven't created something that the business can take and run with and sell. And you have to take accountability for part of the delivery of the solution at the end of the day." It's not just creating something that is really technically elegant, that works really well that drives the ROI. You also have to be able to communicate the value back to the business in order to deploy it. But there actually is, it's both sides of the bridge have to come together. It's not just one or the other. There's a whole element of communication that needs to happen in here.


The first thing that I like to talk about is making sure that there's accountability on both sides for the inherent risk in development and change. Humans are bad at calculating risk. Sometimes businesses are very risk averse, even when they're wanting to lean forward. It's one thing to say, when it's someone else's money, "Yes, I want to lean forward. I want to put this on the table." But when it's actually your job, when you're the one who has to come back at the end of the day and say three months in, "I have demonstrated that I'm generating more incremental revenue today than I was three months ago based on the fact that I adopted this solution."


Sometimes you're not really seeing the revenue at risk until you adopt it at scale. And so you have to invest before you see the return. One of the things that we've done is we've made sure that the technical teams actually have accountability for their results, just like their business counterparts.


So that they can't just say, "Look, I finished it, I'm throwing it over the fence. The accountability and the risk is no longer mine. I get to take on the development work but none of the risks." Creating that accountability starts to minimize some of this. There's always lopsided accountability, there's no way around it, but everyone has to have some skin in the game when it comes to this.


Organizational change is one of the things that we have found to be a big barrier. Not only organizational change in terms of structure, we have changed our structure. We're moving more and more towards development pods, cross-functional groups, where you have engineers and you have QA, but you also have your data scientists in there. You have some of the business accountability, you have UX and shared services who are contributing.


How do we create those teams that are cross-functional teams dedicated to the delivery of a particular experience or a particular product, but still maintaining their lines of expertise so that you're not completely isolated with a pod of other people in your organization who are not data scientists, right? A single data scientist in a pod of engineers can start to lose the best practices and the knowledge of being surrounded by other data scientists.


So it's moving to the matrix organization where you can get the best of both worlds. You're delivering according to a cross-functional team, but you're maintaining your standards and best practices in line with your area of expertise and your area of knowledge. Updating business processes is one of the last ones that I'm really going to touch on as we're already running out of time, but many areas of the organization and I feel like many organizations right now are moving from a waterfall process that involves a lot of upfront planning to more and more agile delivery processes.


And we have found that we keep swinging the pendulum, we get a little bit too waterfall where we're doing too much upfront planning, but then we get a little too agile where what's being interpreted as not needing to do any planning at all. "I'll just do two week sprints and I'll figure it out as I go." That's a little too lack of planning. There's a happy medium where you know what you want to do. You're delivering ongoing feature upgrades in a two-week delivery sprint framework, but you're also looking at the go-to-market strategy simultaneously. You've done enough planning and costing that you know when the timelines to deliver are expected, what the overall costs are and what the benefit are and that you're laser focused on that.


One of the biggest things that we've seen is as really trying to reinforce the need for a balance between you can't be fully waterfall but don't go too far down that agile pathway that you start to think that planning is not important. There is still a place for planning in agile processes.


Do want to acknowledge that it is possible to get through the breakthrough. There are many, many challenges throughout this of building up, deploying, scaling, but where we are right now is that over 40,000 personalized offers a month, we're increasing customer engagement, we're increasing profitability and ROI through the at scale deployment of these solutions. It absolutely is possible.


The power of data and AI right now is more exciting than ever, better than ever. So to all of you who are facing these and I think most people are facing some or all of these challenges, keep going, keep going. Don't be discouraged. It is super exciting right now. And with that, I will turn it over to some questions. Thank you very much.