ÃÛ¶¹ÊÓƵ

Transcript

Hi, my name is Nils Engel and I’m a Solutions Consultant here at ÃÛ¶¹ÊÓƵ. Today I’m going to be talking about ÃÛ¶¹ÊÓƵ Mix Modeler. ÃÛ¶¹ÊÓƵ Mix Modeler is a unified measurement technology that allows clients to understand how their marketing initiatives are performing and helps them to adjust their marketing spend based on what’s working versus what’s not working. It takes two methodologies into play as part of that unified measurement. It allows for multi-touch attribution. This is what we call our bottoms up approach and it’s where we can identify a person who was interacting with the campaign and that same person converted. We can then start giving credit to those marketing touches based on their influence on driving that conversion. As well as marketing mix modeling where we may not have that view, we may not understand that somebody touched the marketing campaign. When we start looking at data coming for example from walled gardens such as Facebook, we don’t get that impression level data from them and so it gets more difficult to understand the impact of those campaigns. That’s where marketing mix modeling comes into play where it uses more of a top-down approach where we take this data in aggregate and find relationships in as we increase spend in Facebook for example. What’s you know do we see a relationship in an increase in conversions around the same time period or some point in the future. What’s nice about these technologies and the reason it’s unified is that the multi-touch attribution outputs can inform the marketing mix model and vice versa. We call this capability transfer learning. Okay so let’s pop into the interface and get a taste of how this ÃÛ¶¹ÊÓƵ Mix Modeler actually works. For starters I’m in an overview pane that gives us a view into all of the data that’s been ingested in and this is a great place to start understanding exactly what data has been loaded in. It’s giving me my top kind of key PIs if I will so I can understand exactly how much spend I’ve loaded in, how many impressions I’ve loaded in, so on and so forth. Once we have data loaded in and ÃÛ¶¹ÊÓƵ Mix Modeler has access to really any data that’s been ingested into the ÃÛ¶¹ÊÓƵ Experience platform. Once that data has been ingested into ÃÛ¶¹ÊÓƵ Mix Modeler we can then go through and start what we call harmonizing the data and this is a very key capability that ÃÛ¶¹ÊÓƵ supports which allows us to come up with a common nomenclature across the variety of sources of data that we’re loading in so that we have a consistent representation of what a channel means in the data. Right and this may be coming across from ÃÛ¶¹ÊÓƵ Analytics data, Facebook impression data, outbound email data, so on and so forth. So all of those sources can be normalized into a single view into that data. We have a common naming convention for clicks across all those different data sources and impressions and costs and any of the other KPIs that have been loaded in. This is something that a marketer can do without any assistance from IT which becomes very powerful self-service interface for performing this. Now once we have that data loaded we can then build models and I’m going to start by showing you the output of a model. The models allow us to understand the performance of these marketing initiatives and within here I can see what data was included in the model for the date range that we’re reporting on. So I can see how many conversions occurred and what the spend was over time. I can see what that spend looked like over each channel. I can then understand broken out by date how much spend was brought in and then the actual volume of touches that were brought in for each channel and I can select which channel I want to be analyzing for. This is just a great way of kind of validating the data that’s being used in the model. Out of that we then get model insights and this is where it’s allowing us to start understanding the effectiveness of these marketing channels for in driving whatever our conversion event might be. First thing we do is we figure out the baseline. The baseline represents how much marketing or how many conversions would occur without any marketing. So if we completely turn marketing off this is our baseline. And then what the model is determining is how many conversions incremental conversions are we getting as a result of marketing and we can see this broken out by our non-spend channels as well as our spend channels. We can see that contribution broken out by channel and we can even understand the return on investment that each of these channels get. This is a huge win for organizations to start understanding where they should be aligning their marketing spend. The data scientists will be happy because we provide model quality views that allow us to show how well the model fit with the data. Low scores mean that the model did not have a good fit with the data. We want a good fit with the data. Now I showed you the output of an existing model but I can build brand new models on the fly. So if I want to start asking a new question of this data I can very quickly create a new model give it a name that makes sense hit next and I can start defining things like what’s the conversion goal that I want to apply this model for. So maybe I want to use orders as a conversion goal and I can specify based on that harmonized data what represents a conversion in the data. I also have the ability to define what represents marketing touches in the data. So I might want to say I want one for email and I can go in and based on that harmonized data I can specify what represents an email touch in the data. I also have the ability to specify of all the data do I want the model to run against everybody in the data or just a subset of them. Right and so I can filter out certain types of data from them maybe certain customers or certain geographic location out of the model. And then I can also include external factors things like the S&P 500 index or employment numbers. Right I can include that if I think it may have an impact on my model. And then I also can include internal factors things like employee count store counts inventory levels right all of that data can be included as part of the model. When I then click on next for the model I then have the ability to specify how long of a training window I want to use and if I have run models in the past and I want I know certain channels have an impact I can include that as part of the model. If I were to click on finish now this would run the model and I would get to see the performance of this data based on that model. As well as modeling we also have the ability to perform scenario planning and this becomes a really key capability as clients have a certain monetary amount that they want to spend they can on marketing they can use ÃÛ¶¹ÊÓƵ Mix Modeler to perform scenario planning to test to start understanding if I invest x amount of dollars how much of a return can I expect on that. And in this case I’m comparing two models against each other that have differing budgets one of them that’s using custom plan where I’ve defined some constraints of how much I want to spend in each of the channels and another where it’s just allowing us to use what the AI recommends. I can see the spend breakdown for each that the model recommends as well as what my forecasted return on investment is and even a return on investment breakdown by channel and ultimately my forecasted return. I can run these scenario plans on the fly compare them against each other and understand which one works best. And then lastly in the plans insight perspective as I now start executing on these plans I can now start understanding how I’m performing to plan am I over budget am I over target that might be okay if I’m over budget right what’s my return on investment looking like and what do the conversions or orders look like. This is a very fast run through on ÃÛ¶¹ÊÓƵ Mix Modeler hopefully it was helpful thank you very much.