Let's go through some comprehensive examples of how to link these non-financial metrics to financial performance. So here's another example, we've got an advertising firm. Okay, what happens as advertising firm is they have lots of local advertisements. This would be like the Yellow Pages, the local advertisement book with the phone numbers. So you can have it in various places in Philadelphia. It could be South Philadelphia or Center City Philadelphia, it could be each one of the little suburbs around Philadelphia has their own advertisements. So what company could decide I'm going to advertise in all the local locations, they can say I'm advertising only the big ones on this? So that's the setting here. Now again, their hypothesis which is with the business model this is, more satisfied customers buy more in the future. Now this could come through a lot of things. We're going to buy more of the same service so instead of me just buying in Downtown Philadelphia, I'm also going to buy one for each of the suburbs because I like your service. It could be, I'm going to upgrade. Instead of just having a black and white ad I'm going to add color, I'm going to add a picture here. It could be they like this service so well you're going to be able to cross-sell them something else. So there are some hypothesis about how a more satisfied customer is going to buy more in the future. So based on this, they spend a lot of money putting in these customer satisfaction issues and like a lot of companies, they do market research. They did this market research and in truth, they really didn't know whether these improvements in customer satisfaction ultimately were producing the financial results they wanted. We're spending money on this. Let's prove this. So one of the advantages of predictive analytics I found in companies is you can use this if you're not working in finance to help convince the chief financial officer that these investments are making sense. Ultimately they want to know show me the money. So that's what happened here, the CFO was pushing the people in marketing to show me that those customer satisfaction initiatives we're spending all the money on, actually are translating through to performance because we're using these metrics in our strategy. We're using this symmetric for performance evaluation and like that other company, their bonus plan were based on this performance symmetric. Again, it was what’s called the top box measure. This is a company were satisfaction ranges from one equals low to a hundred equals high. The bonus was based on what percentage of our costumers gave us a 90 or above. Let's push everybody to the highest side. And again if you think about target setting, this is one of those more is better philosophies. But the CFOs going to push us, so let's actually see if this is actually paying off. Now the nice thing here is this is a company where the customer sign a one year contract. So you can actually link up how satisfied was the customer this year and how much did they buy next year. That's the hypothesis we're trying to test here. So here is the analytics that they did and this is something called non-linear regression. It's basically we want to see how these things are linked together here. Now if you look at this, this is really great for about the first two-thirds, almost a nice perfect line between how satisfied were you last year and what are you going to buy this year. Then you get these strange little steps. You gotta ask what those are. Then finally after about 75, people were more satisfied but they didn't buy more. Now remember what the bonus was. Get everybody 90 or above. Now if I look at this, the first thing you gotta ask yourself is does that target make sense? Now obviously over the whole spectrum here, there is a relationship between customer satisfaction and future purchases. But we also have this target we have to worry about which is everybody pushing towards the top. Well if you look at this I would say the answer is no, it probably doesn't make sense. And it turns out when you talk to these people after about 75, they've bought everything they're going to buy. I don't need to buy anymore advertisements. So you can increase my satisfaction, you can spend money doing that. I'm not going to buy anymore, I don't need anymore. The other thing is you have these weird steps here. And again, here's where some qualitative analysis added to the predictive analytics helps. It turns out what those steps are, is instead of them buying in different cities. So we bought downtown Philadelphia and South Philadelphia and each one of the suburbs, they start upgrading the service. They start going to full page ads, they start adding color, they start throwing at a picture of their service. But you had to be a certain amount satisfied before you really upgraded. Which you're going to buy suppose you are just buying more than ads on this. But you can after a certain point there, it didn't get any more. So again the question is okay, if I am not going to set this target up there, you know move everybody above 90, where would you set it. Now remember, which you want is biggest bang for the bark. If I tell my employees to move some people from one box to the other one, where do I want them spending the money? Because they're going to go after whoever I want them to go after, or whoever I tell them to go after. Well, it turns out, you probably want to focus right there in the middle. If you believe this model, it says if I can move some people off of that nice straight line at the bottom and push them from, say, a 55 to a 65, that's when they're going to buy a lot more. Those are the employees I want to focus on. Not everybody. You do not have to focus on everybody. Let's focus on this group that, based on this analytics, if I can spend money to increase their satisfaction, those are the lines I should get the biggest bang for the buck. Now, again, something else you need to think about when you start doing that are the things that cause somebody who's giving you a score of 90 already, the things that are drivers of their satisfaction the same as the drivers of somebody who's given you a 50. Highly unlikely. Right, the people who already have 90 are satisfied on many dimensions, so asking about those is not going to help. The thing that's going to get me go from a 50 to 60 can be very different. So when I ask those satisfaction questions what I want to know is, what dimensions of satisfaction would cause somebody to move from 50 to 60? That's what you want to measure, not just are you satisfied. Are you satisfied on a dimension that somebody at 50 cares about, not on a dimension that somebody at 90 cares about, because you're not focusing on the 90s anymore. Okay, so based on that, let's focus on these people in the middle on there. So here's another example, we got a technology services company. Again we've got this issue of can you demonstrate that this money we're spending on non financials is actually paying off here, right? Are these non financial metrics actually related to future financial performance? Now here's one of the dilemmas you're going to run into if you start doing this. Again, data is not a problem in companies. But data is power. Information is power. What you're going to end up is what we call data fiefdoms. Lots of different parts of the companies own different parts of the data and they do not like to have you have access to it. My experience, the hardest people to get data from are the finance and accounting people. Right, they're going to claim that you won't understand it, it's proprietary. But if you really want to link non-financials to financials, we have to break down these what are called data fiefdoms. So we had data fiefdoms here. We have to break these down. Now, the financial outcomes they cared about was linking up these marketing metrics and these quality metrics. So they also had some operational metrics that had never been examined by the company. Now let me give an example of what this company does. Part of what they do is basically do cloud computing. They will do all the kind of back office computing for you, put in the cloud, they'll also implement systems for you. So we have two types of metrics. Some are metrics on we;ll ask the customers how satisfied they are on various dimensions. Other ones are operational or quality metrics. How much downtime do we have? How fast is our service? So we have two types of non-financial metrics that we have to worry about. Some are survey based. Some are really hardcore operational measures but they're not financial on this. Now, they had very strong intuition that these customers and quality metrics they had were related to future financial results. And here's kind of what you usually see. We know it's gotta be true. I've been in this business long enough that if you customers are satisfied on these dimensions they buy more. If these quality statistics are better, they're going to buy more. Well that's the intuition but again, an intuition is a hypothesis. So here's what they wanted to look at. Let's relate these customer metrics or operational metrics to financial outcomes. Now one thing you need to do is figure out what financial outcomes do you care about? They don't have to be the same for every company. Well here are the two things they really cared about is annual revenue growth. Okay because you have lots of costs that stay exactly the same, fixed costs. If I can grow the revenue, my profits are going to go up. And what I really want to know is what drives clients to have revenue growth greater than 15%. Those are the ones I want to focus on but I've got to figure out what the drivers are there. They also have these operation or quality metrics. This composite measure of things like downtime, service availability, and things like that, they we're looking at. The customer metrics, they ask you various things on five point scales. Overall how satisfied are you? Again, which of your willingness to reference this word of mouth? What value do you think you get out of this? And because we care about contract renewal, what's the likelihood you think you're actually going to buy our service again? Because it's easy enough to go to another cloud computing provider. So what you need to do when you these models is think about, you know what are the dimensions in your strategy you care about, specifically what are the financial outcomes? And here, it was really revenue growth not overall profitability. So here's what you find. Let's look, and again, we want leading indicators. Let's relate what kind of satisfaction scores you give us on these various dimensions. And future revenue growth, which is what we want as leading indicators. Well it turns out, a lot of these, the green ones are statistically significant when you run the data analysis. The big once appear to be overall satisfaction value. Look at service quality though. It's actually negative, it said as service quality goes up, future revenue growth goes down. Now again we going to worry about this all correlation versus causation here, what is causing this? And here's where you want to dig in. Well, what it turned out this is a non-linear relationship. In cloud computing, you have to have a certain amount of security. You have to have a certain amount of uptime. If you don't meet that, the customer's not going to be there. Once you meet it though, they're not willing to pay for to make it higher. If you've got 99.9% reliability and you're going to charge them a lot more to get to 99.9999% reliability. They don't want it. It is not worth it to them. So what you had here is they were pushing that spectrum of service quality too high. Just didn't pay off. Not that service quality didn't matter. In fact, it's what's called a satisfizer. If you don't reach a certain level, you're out of business, but above that it may not make too much sense to actually improve too much. You've got to worry about whether your customer is willing to pay for it. Now again, the other thing they really cared about was not just revenue growth. Tell me the people that have high future revenue growth. So here you're doing something on a model where the thing you care about is zero one. Did you have high future growth or not? So you got to use different statistical methods. But again what you see is this negative on the service quality. When you run it just as a straight model you'd have to estimate the non-linearities here. The thing that really mattered here, value. Right. For some reason, if people think this is a great value, that's when they're actually going to have higher revenue growth. Now, always the big question then is, how do I get higher value. What do they mean whey they answer this? And that's where you got to go into the peeling the onion bag to say what dimensions of value could use a manager actually do. To actually make this thing pay off.
