Data analysis for organizations -- framework and concepts
Today we'll be answering a simple but hard question which is how can data analysis improve my organization or improve the value proposition that. I'm delivering to my to to my customers now. It seems like a very simple question but it's actually quite hard and the reason why. I'd say it's a very hard question was would be if you ask this question to different people say within an organization whether it's a lion employee your senior management. You're going to get responses that are all over the place so how can we. How can we establish a framework to properly understand. How analytics is actually going to help us. The first thing that we need to do is to differentiate the purpose of our data so there's data that could be for a product or for a decision that's being made now if it's for a product now think of you know it's code or the code execution as part of something that you deliver to the customer so think of Google search or you know Facebook's facial recognition the AI and there's there are other things that you create and the code is there to deliver directly to the customer now we need to separate this out from data and data analysis. That's used with the purpose of making a decision now. In these types of scenarios we're trying to improve decisions or processes within an organization now examples of that would be selecting a strategy properly understanding and making decisions within our supply chain and inventory making a decision of which what's the next product to offer any given customer. What is it that they're going to need. Need next what can we offer for them. Now these are there's these are two very separate categories. And what we're doing today. We're focusing just on the decision end and putting product aside for now now these. This decision focused. Data analysis needs to be decision focused now that sounds tautological and and it is but frequently. This is something that's forgotten. It's something that we've we've pushed to the side and an analyst frequently is doing an L where the outcome of their analysis does not meaningfully impact a decision.
So we need to take a step back from any analysis that we're doing and we could say as a result of this analysis one. What decision will be made differently. The second point is how much value can potentially be created and captured from this improved decision. So first you know what decision is this going to affect if it's not going to affect any decision then there's no way it could create any value and second okay. It's going to change and improve this decision fantastic. That's great well. How much is that worth those. Are these two key questions and the pushback from this is that that you know frequently and I say this because I frequently fall into this temptation is there's lots of really really cool interesting stuff in the data that might be marginally potentially useful. But it's really really cool. It's this constant temptation that just you know lots of curious. People have but we can't fall into that we need to go back to this. These fundamental two questions of what decision is going to be made differently and how much value is that actually going to create and capture. Now let's focus first on question one about good analysis so good analysis should change decisions now one way to approach this is to think about okay. The output of this data analysis dashboard but whatever it is is going to be this. This is our data output and with this output. We will do what with it. Take which actions. How will things change now. I wanted to start out with some bad examples. One of them is political cover. Although this is a mixed can't be bad. I think it's usually bad occasionally. It can't be good but essentially we're doing this this this analysis and the purpose of it is so that we can confirm something and provide political cover for something that we need to do. That's sometimes done for say layoffs or some other areas. There can be good reasons for it but I think 95% of the time it's bad another one that I've seen is secure blankets.
This report makes me feel good inside. There are more of those than you realize if you start walking through if you got a hold of every report in your organization and thought about well why does this report exist. Will it make somebody feel good and makes them feel secure. Great know that those are bad examples and and the last ones a very cynical one which is job security where we potentially create something and it seems like it's important and seems like it's impactful but it's not really doing anything so how do we. How do we evaluate. How do we test. If there's if-then analysis is decision focused well what we can do. We can pretest it. So imagine that there's a model that has three outputs it could say the result is a B or C. And it's always going to say a now at the end of this if you say well if the model says a then. I'm going to take action X and if the model says B I will take action Y and if the model says C I will take action Z. Now if you do this analysis and at the end it's always says I will do X for both a B and C. Then it's not changing my decision so it's not doing anything and so what we're doing we're doing this analysis and verifying that different actions are being taken based off of this output. So if you walk through and pretest you can frequently end up with better models and better outputs because if you pre test something. Let's say you walk through all of these steps and you realize you know what like that's that's that's kind of helpful but not terribly helpful. We're going to change our model around or capture different data to distinguish better and to provide better output. So that a better and more defined action can be taken with that information so pre testing not only. Lets you know whether you should do something it frequently improves. The analysis itself now a quick aside about value creation and value capture. Now too many of us this can be a kind of a foreign or strange concept but it's actually very straightforward fundamentally.
You need to understand. We need to understand how much values is going to be created captured from this improved decision that we've we've created because of our successful data analysis. Now what is value creation value creation were providing a benefit to our customers or reducing costs in our organization and value capture is the side. So we've created this value and the portion of it that benefits you know some of that benefits say going to go to the customer. Some of it's going to stay with us. As an organization. The captured portion is the portion of the value creation that we get to keep generally in the form of higher profits. Now there's a formula for this for data analysis and it's it's very straightforward and. It's very intuitive as soon as you see it but it's fundamentally it's there's this value that you capture from your improved decision because of this analysis my decision has improved by X amount but then you need to subtract from it the value that would have been captured if you had done no analysis so you take the value from in the improved decision and subtract it without the from the improved decision with the analysis. Subtract it without the analysis plus it also took time to do the data analysis so there are costs involved there including the opportunity cost of not doing other projects and you end up with the net value of a data analysis and this net value. If it's positive it means because of our analysis we've we've improved the organization we've created value we've captured value and it's you know net present value positive and in financial terms. Now if it's negative then we shouldn't have done it in the first place. It didn't help us so that that's a good formula to keep in mind. You imagined that there's this improved decision you compare it to the status quo without the analysis and then you subtract out you know the cost and time involved now let's home in more on this concept of value capture and value capture really is comparing the costs and benefits for our organization first on the costs.
I did analysis computed. There are the direct costs of you. Know salary benefit system processing blah blah blah. But you know you're familiar with but is also administrative time. How much time has been spent on this analysis like what else could have been done with that time. What are other initiatives that could have. We could have focused managerial time and talent up so this is this can be very very expensive for data analysis but also the benefits are frequently under countered as well the benefits can be enormous and this is especially the case when something is perpetual or long-term where we have a certain process of analysis that embeds in the organization and creates these process improvements on an ongoing basis. So what it does. It's it's say let's say it's not enormous. It's improving some process. Say in your warehouse. That's happening a thousand times a day and we're proving each one of those thousand times a little bit and that happens this month. That happens this year it happens next year happens five years from now. It's a perpetual or a long-term improvement so that benefit can be enormous. You could you also see this with big decisions. So we're gonna bet the company well. We need to do the math before we bet the company and we also need to do it correctly and so these benefits can be enormous whether they're big decisions or frequent smaller decisions. We frequently undercount the benefits and we also frequently undercount the costs so we just need a proper understanding of both the benefits and the costs in order to in order to apply the formula that we've just gone over so one of the one great advice that I've heard is especially toward the beginning if you're doing an analysis and it's not for something that's either perpetual or big then stop doing it and start working on something that's perpetual or big instead it's it's it's very is basically.
There are very very important things you could be working on that. Really move the needle for the company so focus on that instead of small potatoes. It's an obvious point. But it's one that we need to remind ourselves of frequently so that we don't fall into that trap of focusing on the more trivial items another way of viewing this value frequency so here on the y-axis we have value and on the x-axis we have frequency and it breaks up into either low value at a low frequency scaling all the way to high value and high frequency decisions now high value and high frequency decisions that they're there generally involves smaller companies. But not always. It's when you're betting the company on something very very important now with smaller companies. This happens frequently. With larger companies it can happen especially in certain industries. Let's say fashion or even within technology when you're really seeing new models of your product if you botch a release of you know the latest iteration of your product that could jeopardize the company potentially. I mean I mean. Fashion is an obvious example. Where every year you need to come out with you know latest and greatest whatever the best product is or else you could fail before next year hits so these high value high-frequency stakes occur especially in some industries. But it's very industry specific so if that applies to you of course with your data analysis initiative you're gonna be focusing on yellow first and foremost and everything else gonna be pushed to the side for now and then then second you're gonna be focusing on your greens like we just went over high value low frequency events. Say you're acquiring another company now. You're not doing this every day every week every month even maybe not even every year and this is a very high value decision needs to be done. Well needs to be done properly. We could also think of low value high frequency like in our warehouse where say we're picking orders and we're going to improve this order picking process in our warehouse say buy we're gonna shave off a few seconds by implementing a certain model from our data analysis group.
Yes we've saved this time but over the year over the next five years over ten years we're going to see huge dividends from this and one of the the dangers that happens that we identify a project and it seems like green or yellow but over time the longer we work on it the more it started sliding into the Reds either because this Skov keep shifting or you know the data is not there or the models not delivering the value. That the that you you need and so you settle for something. That's not quite as impactful and so. There's this natural decline toward that red zone. It's something that we just need to do. A check on ourselves is this project sliding in that direction or not and of course to some degree they frequently are but if it does - it - to a limited extent at least but if it does to a greater extent we just need to be ready to kill a project if we need to. Let's recap for today. Analysis should change decisions and if they don't don't do the analysis and the second big point is we need to weigh the costs and benefits of an improved decision so one the costs are much larger than we think and second the benefits could be much larger than we think if we select the right types of projects to work on so it's more of a value frequency framework that we can apply and recognize we where we can make the biggest impact.
So we need to take a step back from any analysis that we're doing and we could say as a result of this analysis one. What decision will be made differently. The second point is how much value can potentially be created and captured from this improved decision. So first you know what decision is this going to affect if it's not going to affect any decision then there's no way it could create any value and second okay. It's going to change and improve this decision fantastic. That's great well. How much is that worth those. Are these two key questions and the pushback from this is that that you know frequently and I say this because I frequently fall into this temptation is there's lots of really really cool interesting stuff in the data that might be marginally potentially useful. But it's really really cool. It's this constant temptation that just you know lots of curious. People have but we can't fall into that we need to go back to this. These fundamental two questions of what decision is going to be made differently and how much value is that actually going to create and capture. Now let's focus first on question one about good analysis so good analysis should change decisions now one way to approach this is to think about okay. The output of this data analysis dashboard but whatever it is is going to be this. This is our data output and with this output. We will do what with it. Take which actions. How will things change now. I wanted to start out with some bad examples. One of them is political cover. Although this is a mixed can't be bad. I think it's usually bad occasionally. It can't be good but essentially we're doing this this this analysis and the purpose of it is so that we can confirm something and provide political cover for something that we need to do. That's sometimes done for say layoffs or some other areas. There can be good reasons for it but I think 95% of the time it's bad another one that I've seen is secure blankets.
This report makes me feel good inside. There are more of those than you realize if you start walking through if you got a hold of every report in your organization and thought about well why does this report exist. Will it make somebody feel good and makes them feel secure. Great know that those are bad examples and and the last ones a very cynical one which is job security where we potentially create something and it seems like it's important and seems like it's impactful but it's not really doing anything so how do we. How do we evaluate. How do we test. If there's if-then analysis is decision focused well what we can do. We can pretest it. So imagine that there's a model that has three outputs it could say the result is a B or C. And it's always going to say a now at the end of this if you say well if the model says a then. I'm going to take action X and if the model says B I will take action Y and if the model says C I will take action Z. Now if you do this analysis and at the end it's always says I will do X for both a B and C. Then it's not changing my decision so it's not doing anything and so what we're doing we're doing this analysis and verifying that different actions are being taken based off of this output. So if you walk through and pretest you can frequently end up with better models and better outputs because if you pre test something. Let's say you walk through all of these steps and you realize you know what like that's that's that's kind of helpful but not terribly helpful. We're going to change our model around or capture different data to distinguish better and to provide better output. So that a better and more defined action can be taken with that information so pre testing not only. Lets you know whether you should do something it frequently improves. The analysis itself now a quick aside about value creation and value capture. Now too many of us this can be a kind of a foreign or strange concept but it's actually very straightforward fundamentally.
You need to understand. We need to understand how much values is going to be created captured from this improved decision that we've we've created because of our successful data analysis. Now what is value creation value creation were providing a benefit to our customers or reducing costs in our organization and value capture is the side. So we've created this value and the portion of it that benefits you know some of that benefits say going to go to the customer. Some of it's going to stay with us. As an organization. The captured portion is the portion of the value creation that we get to keep generally in the form of higher profits. Now there's a formula for this for data analysis and it's it's very straightforward and. It's very intuitive as soon as you see it but it's fundamentally it's there's this value that you capture from your improved decision because of this analysis my decision has improved by X amount but then you need to subtract from it the value that would have been captured if you had done no analysis so you take the value from in the improved decision and subtract it without the from the improved decision with the analysis. Subtract it without the analysis plus it also took time to do the data analysis so there are costs involved there including the opportunity cost of not doing other projects and you end up with the net value of a data analysis and this net value. If it's positive it means because of our analysis we've we've improved the organization we've created value we've captured value and it's you know net present value positive and in financial terms. Now if it's negative then we shouldn't have done it in the first place. It didn't help us so that that's a good formula to keep in mind. You imagined that there's this improved decision you compare it to the status quo without the analysis and then you subtract out you know the cost and time involved now let's home in more on this concept of value capture and value capture really is comparing the costs and benefits for our organization first on the costs.
I did analysis computed. There are the direct costs of you. Know salary benefit system processing blah blah blah. But you know you're familiar with but is also administrative time. How much time has been spent on this analysis like what else could have been done with that time. What are other initiatives that could have. We could have focused managerial time and talent up so this is this can be very very expensive for data analysis but also the benefits are frequently under countered as well the benefits can be enormous and this is especially the case when something is perpetual or long-term where we have a certain process of analysis that embeds in the organization and creates these process improvements on an ongoing basis. So what it does. It's it's say let's say it's not enormous. It's improving some process. Say in your warehouse. That's happening a thousand times a day and we're proving each one of those thousand times a little bit and that happens this month. That happens this year it happens next year happens five years from now. It's a perpetual or a long-term improvement so that benefit can be enormous. You could you also see this with big decisions. So we're gonna bet the company well. We need to do the math before we bet the company and we also need to do it correctly and so these benefits can be enormous whether they're big decisions or frequent smaller decisions. We frequently undercount the benefits and we also frequently undercount the costs so we just need a proper understanding of both the benefits and the costs in order to in order to apply the formula that we've just gone over so one of the one great advice that I've heard is especially toward the beginning if you're doing an analysis and it's not for something that's either perpetual or big then stop doing it and start working on something that's perpetual or big instead it's it's it's very is basically.
There are very very important things you could be working on that. Really move the needle for the company so focus on that instead of small potatoes. It's an obvious point. But it's one that we need to remind ourselves of frequently so that we don't fall into that trap of focusing on the more trivial items another way of viewing this value frequency so here on the y-axis we have value and on the x-axis we have frequency and it breaks up into either low value at a low frequency scaling all the way to high value and high frequency decisions now high value and high frequency decisions that they're there generally involves smaller companies. But not always. It's when you're betting the company on something very very important now with smaller companies. This happens frequently. With larger companies it can happen especially in certain industries. Let's say fashion or even within technology when you're really seeing new models of your product if you botch a release of you know the latest iteration of your product that could jeopardize the company potentially. I mean I mean. Fashion is an obvious example. Where every year you need to come out with you know latest and greatest whatever the best product is or else you could fail before next year hits so these high value high-frequency stakes occur especially in some industries. But it's very industry specific so if that applies to you of course with your data analysis initiative you're gonna be focusing on yellow first and foremost and everything else gonna be pushed to the side for now and then then second you're gonna be focusing on your greens like we just went over high value low frequency events. Say you're acquiring another company now. You're not doing this every day every week every month even maybe not even every year and this is a very high value decision needs to be done. Well needs to be done properly. We could also think of low value high frequency like in our warehouse where say we're picking orders and we're going to improve this order picking process in our warehouse say buy we're gonna shave off a few seconds by implementing a certain model from our data analysis group.
Yes we've saved this time but over the year over the next five years over ten years we're going to see huge dividends from this and one of the the dangers that happens that we identify a project and it seems like green or yellow but over time the longer we work on it the more it started sliding into the Reds either because this Skov keep shifting or you know the data is not there or the models not delivering the value. That the that you you need and so you settle for something. That's not quite as impactful and so. There's this natural decline toward that red zone. It's something that we just need to do. A check on ourselves is this project sliding in that direction or not and of course to some degree they frequently are but if it does - it - to a limited extent at least but if it does to a greater extent we just need to be ready to kill a project if we need to. Let's recap for today. Analysis should change decisions and if they don't don't do the analysis and the second big point is we need to weigh the costs and benefits of an improved decision so one the costs are much larger than we think and second the benefits could be much larger than we think if we select the right types of projects to work on so it's more of a value frequency framework that we can apply and recognize we where we can make the biggest impact.