Blog

There are many great conferences hosted on the subject of data science by different programming and community groups along with industry heavy hitters. Since it’s data science month at Cardinal Path, we sat down with our data science team to find out about some of their favorite conferences of 2016.

Danika:

broom: Converting statistical models to tidy data frames – useR! International R User Conference
Those that know me, know that I hate Excel. But I understand it has a purpose. Frequently, data is stored in Excel sheets in formats that include formulas, images, multiple headers, hyperlinks, and perhaps even random values way off in the 1,000,000th column. I loved this talk because it is about getting your data out of Excel and into R in a clean way: “Jailbreakr- Get out of Excel free”. I’m curious to see what kind of ugly data it can handle. Jenny Bryan also worked to bring us the R package googlesheets, so she is helping make spreadsheets easier to work with in R in more ways than one!
jailbreakr: Get out of Excel, free – useR! International R User Conference
Having data that looks clean is necessary for most modelling and data visualization use cases. Most of the time, we want the data to be ‘tidy’. In this case, tidy means that you should have 1 row per observation, 1 column per variable, and 1 table per observational unit. However, the output of R models, such as regression with lm(), is pretty messy so the resulting output is not tidy. Basically, this talk is about making the output of models tidy, so that it can be used in later data visualizations. David Robinson developed the R package broom to deal with the messy output of R statistical models.

Charlotte:

Machine Learning & Art – Google I/O
This presentation is by The Google Cultural Institute from this year’s Google Developer conference. It shows what fun can be had with machine learning from an artistic angle.

Sunspring – Ars Technica
A great follow up video to the Google video on machine learning and art is this short sci-fi film published by Ars Technica – it’s not a conference talk, but it’s directly related to the theme of machine learning and art. This is what happens when you let artificial intelligence write a movie. Gizmondo did a great write up of how the neural network was developed for this on their site.

Jas:

Size of Datasets for Analytics and Implications for R
A quote that stood out to me from this presentation was: “It takes a big man to admit his data is small” — @jcheng. Big data is a buzz word that doesn’t seem to be going away anytime soon; the majority of data analysts/scientists deal with at most datasets of several Gigabytes. Interestingly, the growth of RAM year over year has outpaced the growth seen in the average size of datasets. Szilard Pafka discusses why it might be wise to stick with R for even relatively large datasets and why using immature big data tools may in fact be counter-productive.

FiveThirtyEight’s data journalism workflow with R
Have you ever visited fiverthirtyeight.com (founded by Nate Silver of 2012 US election prediction fame) and noticed the charts look a lot like ggplot2 charts in R….Well, that’s because they are! The only missing piece is that they hand over the ggplots to their visual journalism team which makes the charts “sexier” using Illustrator*. In fact, R is used in every other step of the journalism pipeline; Andrew Flowers, Quantitative Editor at 538, elaborates on why this is the case.
*Andrew doesn’t actually go into the details on what is done in Illustrator, but this video shows how you can export ggplots into Illustrator and make some quick, visually appealing changes.

———————————————————————————————————-

If you are interested in learning more about all things data science, don’t forget to join us on November 3rd at 10:00 am PT/ 1:00 pm ET for our webinar hosted by the American Marketing Association: “Applying Data Science Methods for Marketing Lift“. You can register here.

Businesses that belong to the B2B space have unique KPIs and reporting needs. This is because of the way they sell, it’s different, which means that the way that they measure is also different. Instead of thinking about visits, engagement and conversions alone, B2B companies also think about prospects, opportunities and win rates. There are multiple ways to customize your current dashboards to not only represent specific B2B needs, but to also to integrate predictive analytics to answer specific questions for this type of business. In this post, we’re going to use data science techniques to visualize the answers to the three different questions below in a Tableau dashboard format:

  • Who are your customers?
  • Will they respond to you?
  • What will they respond to?

Who Are Your Customers?

Before you start tackling prediction tasks, it helps to understand what you’re already working with. Who are your current customers? What characteristics are common to them? Are there certain segments that look and act similarly who should have common marketing and outreach strategies?

A descriptive data mining task will allow you to understand what your current customer base looks like. In this case, we’re referring to a cluster analysis. This will allow you to see visually how different segments of customers exist within your CRM. Clustering is incredibly flexible, and you can cluster customers according to whatever makes sense for your business. This could include:

  • Company characteristics: industry, size
  • Individual characteristics: title / function
  • Website analytics: content consumption, frequency or recency of site visits
  • Lead gen data: lead score, name of sales contact, pipeline stage, time in stage, number of opportunities, total win value, win rate, lifetime value

Once you put it in dashboard form, you’ll be able to see who the major segments of your customer base are and how this is changing over time.

This is a question you can answer in Tableau. With the release of Tableau 10 earlier this year, cluster analysis is now an in-product feature. Underneath the hood, Tableau is using a k-means clustering algorithm to find commonalities and groupings in your data, based on their similarity across whichever features are most useful to you.

Will They Respond to Your Campaigns?

Once you’ve got a better understanding of who your customers are, you can start thinking about how your marketing tactics are reaching and engaging them.

The idea of “response” can be defined very broadly here. For a site with online e-commerce functionality, customer response is usually defined as a purchase. But for B2B sites who typically have a longer sales cycle, the response could be a softer conversion point: did they reach out to sales? Complete an inquiry form? Download a product data sheet? Advance down the sales pipeline? The response outcome selected here should align to the realities of your customer’s buying process.

To put this analysis into dashboard form, you’ll need to build a response model in R that predicts the likelihood of each individual customer to respond to a campaign outreach effort. The Tableau – R integration is the way you get the customer level analysis out of R and into your dashboard efforts. For reporting purposes, it doesn’t make sense to report on individuals, so a rolled up and average of the findings across major customer segments and demographics would be of use.

Once you’ve got a calculated response rate, you can also start projecting the ROAS (return on advertising spending) of our outreach efforts or even determine our break-even point with some basic cost information. Tableau let’s you build in dynamic functionality so that you can conduct your own what-if analysis to see the impact of a changing ad cost on ROAS.

What Will They Respond To?

Digging in even deeper, you could also take a look at what creative elements your customers are likely to respond to. This is a more challenging analysis to build, mainly due to the fact that the required data isn’t always easy to get a hold of. There are several ways you could build out this dataset, either through carefully segmented A/B tests, or through careful tracking of historical response rate information for individual customers.

The dataset for this requires as much information as possible about the ad creative: what was the creative like? The calls to action? Did it have images? What was the messaging in the text? Were there other incentives in the creative? Once you’ve built out a robust data set, the analysis occurs the same way as before; by using the R integration in Tableau to build out the likelihood to respond to creative and then feeding that back to Tableau for visualization and dashboarding purposes.

If you put all of these components together: an understanding of your customer base, their likelihood to respond to outreach efforts, and details on what exactly they respond to, you’ll be well on your way to be able to optimize your marketing time and efforts. One advantage of putting predictive analytics into a dashboard format is that it allows you to scale easily across brands, geographies or products. The data set would change for each different dashboard, but the visualizations and R code fueling the analysis could typically remain unchanged.

If you’d like to find out more about predictive analytics techniques along with other advanced industry leading data science applications, check out our upcoming webinar with the AMA: “Applying Data Science Methods for Marketing Lift”, on November 3rd, 2016 at 12- 1pm PST. You can register for free here.

For many organizations, even those with sophisticated tools and practices, getting the right data, and more importantly, the insights – to the right people at the right time is difficult.  With multiple stakeholders, various data sources, and the ever increasing pressure to ensure that the data and insights are actionable, one thing is clear: marketers feel challenged when it comes to delivering on the promise of dashboards.

To help frame up the topic of dashboard execution, it helps to understand why dashboards fail. We undertook a dashboard inventory at Cardinal Path this year, where we polled everyone who works here across both our US and Canadian offices. We asked “have you completed a dashboard project for a client”, “is that dashboard still in use?” “and if not, why?”.

The following represents some of the most common reasons we saw why dashboards were no longer in use.

  • The website was updated but the dashboard was not. The problem here ties back to the need for a consistent taxonomy. Just because your site changes doesn’t mean your metrics have to; with a site change you should still largely be able to keep the same metrics and dimensions mapped to the same place. And maintaining a dashboard becomes much easier if those metrics don’t change just because the site has.
  • The analytics platform was not advanced enough.
  • The dashboard platform was not user friendly. The second and third point here both map to the same underlying concern: a mismatch between the needs of the business and the capabilities of the technology. This really speaks to the necessity to carefully evaluate the technology to make sure you’re getting the correct solution in place.
  • The dashboard takes more work to update than value it provides. I see this all.the.time. I would much rather put an extra effort into the dashboard creation process just to make sure that a dashboard has an efficient design. If a dashboard is easy to update you have more time available for analysis and insights.

So with that being said, here are  some best practices to help you get started with your own dashboards.

Best Practices

1. Take Action

For many organizations, this represents the first step in actually using all their data rather than simply collecting and then drowning in it. Dashboards can help improve both performance and the decision-making process, and can make complex data inviting and accessible with low barriers to entry. But the first step is deciding to make them a part of your plan.

2. Define the metrics to track

Spend time agreeing on the metrics that matter, and build reports for those metrics that can feed dashboards. A KPI is only as valuable as the action it inspires. Too often, organizations blindly adopt industry-recognized KPIs and wonder why that KPI doesn’t reflect their own business and fails to affect any positive change. This can take time, but don’t underestimate the value it adds. This is where you are building buy-in for your entire project.

3. Match dashboards to requirements

Ask yourself as you build your dashboard are stakeholder questions getting answered? Are the KPI’s appropriate and in line with your stakeholders needs? The CMO will have very different metrics than the channel managers. When looking at specific tools, you should also be asking whether the dashboarding tool is appropriate- is it too advanced? Does it provide enough interactivity? Too much interactivity?

That being said, it is still common that even with the best laid plans, the purpose of your dashboard will get lost. This often happens because the data that is being presented is for the wrong people, or presented in the wrong format.

Organizations with data analysis programs that don’t plan for delivering their insights in easy-to-access-and-evaluate formats fall into several common traps:

  • Paying for the wrong tools and analysis driving up costs, reducing flexibility, increasing turnaround times and too often yielding only partial results
  • Lack of knowledge transfer and capacity-building radically minimizes ROI from hard-won marketing technology investments;
  • Loss of productivity, as internal and external stakeholders find it difficult to collaborate with each other to gain a collective perspective via a single source of truth

4. Share

A dashboard is only meaningful if people use it. Be sure to promote and disseminate to everyone who could potentially benefit from the data. Dashboards can make you look good, and put you in a position to help other departments. When you share dashboards, be sure to include information on where the data has been gathered from and who they need to contact if they have questions – it helps users understand the data and know who they can go to for more information.

5. Iterate

It’s not always going to be perfect in its first incarnation. Once implemented, start gathering feedback from the people using it, and seek to make improvements over and over again. Your dashboard should be a dynamic platform that is continually changing and adapting to meet the needs of its users.

6. Hone in on what matters

Dashboards need to be live and real-time view of your data, you should be able to drill down to underlying reports, data and customer segments. If you find irregularities, you should be able to click through to discover the source. This ensures that the data is actionable- which, if you’ve ever read any of our other materials, you will know, is the foundation of what we do at Cardinal Path.

7. Make it pretty

This is the place to start thinking about tools. We’ve all been there. Staring blankly at a mind numbing spreadsheet which tells the untrained eye, basically, a whole lot of nothing. An aesthetically pleasing and well-designed tool is actually more functional, and more effective. It gets used more often, and provides users with more delight.

If you’d like to take the first step by visualizing your own Google Analytics data, try this free trial from Klipfolio! Also, check out these great resources on common pitfalls of dashboard design and on visual communication.

 

 

 

In our most recent webinar, Marketer’s Guide to Digital Dashboards, Cardinal Path’s panel of experts, which included Charlotte Bourne, Mark Tallman, and special guest Stephane Hamel, shared insights into some of the most common reasons why so many dashboards fail, and how to save yourself from these pitfalls to set up dashboards that deliver actionable insights for business users.

Judging by the excellent questions posed by our audience, many are looking for ways to get valuable insights from their dashboards. With multiple stakeholders, various data sources, and increasing pressure to ensure that the data & insights are actionable, one thing is clear: marketers feel challenged when it comes to delivering on the promise of dashboards.

Below are some of the great questions asked during the webinar, along with the answers.

1. We have a lack of non-vanity goals/objectives from our stakeholders. They want too much data. How do you arrive at a consensus on dashboard priorities?
A: Stephane Hamel: In a perfect world, you would have addressed the KPIs right at the beginning of the project, when defining the objectives. The reality is that this is often not the case. You could use an investigative approach and answer the 5W+1H questions – which will help to frame the needs & scope of your dashboard. I’m a fervent advocate of breaking something into smaller parts in order to gain a better understanding of it. If you can at least control the scope and start by agreeing on a single KPI, then you should go for it. There are many benefits to having a more agile approach with multiple iterations. You can call it a “pilot project” if you’d like, but whatever you call it, make sure that it gives you a “right to fail learn”! For one client, they wanted a very sophisticated dashboard which merged multiple data sources. By using this“pilot project” approach, we were able to scope it down and at least start providing value in a short amount of time. However, the real benefit came from standardizing the KPIs terminology, campaign taxonomy, and getting everyone on board. In the end, change management was actually the most difficult part, once we had that in place, building the dashboard was easy!

2. How do you identify the difference between ‘nice-to-know’ and actionable data?
A: Stephane Hamel: KPIs are usually supported by additional metrics (eg. If you show “conversion rate”, the underlying “Transactions” and “Sessions” should be visible too. Keep in mind what is a “nice to know” might be actually be a critical component for someone else. The number of Facebook likes might be more important to someone who is operational, but could be only a ‘nice-to-know’ for your executive-level stakeholders. In the end, there are three ultimate KPIs: satisfaction (of customers, employees, investors, partners), costs, and revenue. To be considered a KPI, the metric should clearly tie back to one of those. Is “Number of Facebook Likes” a good KPI? I’m sorry to say this to you but, ‘it depends’. It depends on the answers to the 5W+1H!

 

Screenshot_2
3. I’m using Google Data Studio and trying to bring all sources – web analytics, email metrics, industry ad data and social media engagement all together into a dashboard. Any advice or experience on doing this?
A: Mark Tallman: Good question! There are a number of ways you can tackle this, but it all comes down to how you want to connect to your data. Currently, Google Data Studio can connect to Google Only data sources. This means Google Analytics, Google AdWords, Attribution, Big Query, YouTube, and Google Sheets. They recently announced a connection to Cloud SQL and MySQL databases as well. So, if you only want to connect to Google-owned data sources like Analytics and AdWords, just use the direct connector! If you want to use additional data sources like Facebook, Bing, Twitter, MailChimp, or others, you have to get crafty…you can either push that data into Google Sheets (less technical) or a database (Cloud SQL or MySQL) and pull it into the dashboard from there. You can either do this manually (export the data from each platform, then paste into Sheets) OR programmatically with something like Analytics Canvas.

4. Several people have said they are struggling with managing the needs of multiple stakeholders for a single dashboard. What’s your advice?
A: Stephane Hamel: Stephen Few defined a dashboard as fitting on a single page (or screen), but the reality is that it’s not always going to fit on a single page. It must, however, have consistency in the way you define and present the same KPI on different dashboards. Personally, I would aim to have a single dashboard even if it has to be larger – this way, all of your stakeholders are “exposed” to what is important for their peers, which can facilitate the growth of a data culture, and even spark interesting conversation, maybe even surfacing interesting ideas and solutions. To quote Stephen: “A dashboard is a visual display of the most important information needed to achieve one or more objectives; consolidated and arranged on a single screen so the information can be monitored at a glance”. -Stephen Few, Information Dashboard Design.

5. Which tool, besides Tableau, can you recommend that will allow me to pull in social and web analytics data side by side? The trouble with Tableau is that it’s hard to share dashboards since licenses are required to even view these reports.
A: Charlotte Bourne: If you have programming skills or developer resources you can leverage programming languages like Python and various R packages to access the different APIs. The challenging part with those processes is often the authentication step, which can be harder than creating the actual data pulls. Other advantages there are that you can use those languages for data cleaning and wrangling, to join your data sets and to push it into a final format.

If you don’t have programming skills there is quite a market for various connectors. You’ll find a whole range of Excel-based connectors that allow you to re-pull and refresh your data quite quickly. A few options in this area is Next Analytics, Excellent Analytics, Supermetrics, etc. These serve as easy entry points for non-technical resources and have different capabilities with respect to data sources you can access.

Also keep in mind that Tableau dashboard creation requires a licence but Tableau dashboard viewing does not. There is a free dashboard viewing tool called Tableau Reader which allows end users to access the dashboards you create with all of the same interactive functionality. (This would be like the difference between Adobe Acrobat vs. the Acrobat Reader).

Some other options are Klipfolio, PowerBI, Google Sheets, and Google Data Studio, but all come with their own advantages and limitations.

6. What are some examples of companies that provide ETL (Extract, Transform, Load) software?
A: Charlotte Bourne: Analytics Canvas comes to mind (specialized for Google Analytics), Informatica, Talend (open source) or Pentaho (open source) are other well known vendors. There is a Gartner magic quadrant for this as well.

7. We have many data sources, so the big obstacle is automating the data extracts while not yet having a centralized, managed data mart.
A: Charlotte Bourne: I really understand this challenge. You’re looking to get as much automation as possible to have an efficient update process which is difficult as the number of datasource increase. If a data mart build isn’t a good solution, I would do a gap analysis. First, on the data side: what kind of automation can come from the data platforms you currently use. Look into what automation you have available to you in terms of scheduled reports and report formats. Does that give you sufficient data for your dashboard? Are additional joins or data cleaning needed?

At this point, you should think this through from the dashboard visualization side. What automation features are available through your dashboard tool? What data connections are available?

Finally, look at the gaps. If you find that you cannot get to the degree of automation and efficiency using the native capabilities of your data sources, and your data visualization tool alone, that’s when you should start evaluating a data integration tool or ETL tool in order to bridge that gap. That allows you to then work on streaming your dashboard process without resorting to the all-out step of building a datamart to support your dashboard project.

8. Does Cardinal Path offer this software? 
A: Charlotte Bourne: While Cardinal Path doesn’t offer the actual dashboard software, we do partner with leading vendors like Tableau and Klipfolio and provide the services to pull in data, define KPIs, enable integrations and automations, conduct analyses, and ongoing reporting in order to draw out the insights that matter and maximize return on a dashboard investment.

9. I’m using Google Analytics but before to plug the data into my dashboards I need to modify them using Excel CSV. Do you know if there is way for the data to be automatically modified before I plug them into my dashboard?
A: Mark Tallman: Yes, Analytics Canvas! It does however, depend on which transformations you’d like to make. Most of the time, you can skip Excel completely.

10. Does Analytics Canvas connect to Google Data Studio?
A: Mark Tallman: Not exactly. In Analytics Canvas, the best path forward to connect to Google Data Studio would be to export your dataset into Google Sheets and use that as your point of connection into Google Data Studio. There is no direct connector, but the currently available output formats can still fit the needs of Google’s product.

 

Data Driven Attribution and Media Mix Modelling (MMM) both help marketers understand marketing channel performance and return on investment. While it’s true that both help to measure marketing success, they differ in their data requirements and capabilities. If you are trying to figure out which one is best suited to your organizational needs, read on.

Below is a table which summarizes the main differences between the two.

DDA or MMM?

Model Type

Media Mix Modelling typically uses a regression or some similar statistical or econometric technique to determine how much each channel impacts your conversion goal. This is more of a correlational measure, as we don’t have a direct link to be certain that any converted users have seen the media. Basically, the data is aggregated and not at the user level.

Data Driven Attribution modelling is commonly viewed as a more causal method, as we can find a  direct link between a user seeing a media channel and converting (or not converting). However, the actual algorithms used to measure the lift from each marketing channel will differ by provider, and may use a range of statistical and machine learning modelling techniques. Basically, the results you will receive may be different.

Data and Data Collection

Media mix modelling (MMM) has a one-time collection process, so if results are needed only once, or even yearly, this is going to be the better option. MMM is typically a better option when you have historical data for all potential drivers of your business readily available. MMM also works well with offline data, such as TV impressions or the number of stores open in a particular region.

Data Driven Attribution works best for the collection of data from online channels and for the ongoing collection and refreshing of models, so if frequent report updates are required, this would be the preferred option. Depending on the attribution provider, tags may be used to collect data or, attribution may work with data collected by your existing analytics platforms. Typically, there will be some time in between  when you initiate data collection, and when there is enough data available for attribution models to run so that  you can get your initial results.

Generally, Data Driven Attribution is the best option for measuring online conversion paths, though some attribution providers have connectors to offline media or offline conversions.

Online or Offline Conversions?

If the majority of your conversions occur offline, MMM makes more sense than Data Driven Attribution. Sure, some attribution providers have integrations available with data collection partners or other connectors to offline data. But the econometric models in MMM readily accommodate offline data. For MMM, the conversion type can be whatever is relevant to your business, provided the historic data is available (typically, this would be sales volumes (online and/or offline)).

Data Driven Attribution is the strongest candidate for online conversion types and can work well with hard conversion goals, such as purchases, and softer conversion goals, such as newsletter downloads.

Reporting Deliverables

If marketing strategy is constantly evolving, and budgets can be shifted easily between channels on a monthly or even weekly basis, then a Data Driven Attribution provider, which allows for daily report updates, is superior.  Reports may include detailed impacts of online effects such as lift analysis, recommended frequency capping, viewability, conversion path reports, and more.
MMM can be refreshed approximately once a year to keep your data recent and relevant, though the cadence of updates will depend on the nature of your business. If budgeting is typically done once a year, this approach can work well. Additionally, it does have the ability to test different budgeting scenarios and their resulting outcomes, whereas Data Driven Attribution may not have this baked in. Depending on the provider of your MMM model, and even though the inherent model is fixed, you may also have the capacity to play out different spending scenarios.  That is, your MMM model may allow you to forecast the ROI of channels based on different spend inputs.

In general, Media Mix Modelling is an approach that works well for companies with significant offline media spend while Data Driven Attribution modelling is most effective for companies with the majority of their marketing activities running online and in large volumes. I hope that this comparison has been helpful in your research! If you have further questions about how you are measuring your marketing success, please contact us.

Data Analysis

The Cardinal Path Data Science Team Weighs in on Their Favorite Conference Talks of 2016

There are many great conferences hosted on the subject of data science by different programming and community groups along with industry heavy hitters. Since it’s data science month at Cardinal Path, we sat down with our data science team to find out about some of their favorite conferences of 2016. Danika: broom: Converting statistical models … Read Full Post

Supercharge Your B2B Dashboards with Predictive Analytics!

Businesses that belong to the B2B space have unique KPIs and reporting needs. This is because of the way they sell, it’s different, which means that the way that they measure is also different. Instead of thinking about visits, engagement and conversions alone, B2B companies also think about prospects, opportunities and win rates. There are multiple … Read Full Post

Dashboard Best Practices

For many organizations, even those with sophisticated tools and practices, getting the right data, and more importantly, the insights – to the right people at the right time is difficult.  With multiple stakeholders, various data sources, and the ever increasing pressure to ensure that the data and insights are actionable, one thing is clear: marketers … Read Full Post

Marketer’s Guide to Digital Dashboards- Webinar Q&A

In our most recent webinar, Marketer’s Guide to Digital Dashboards, Cardinal Path’s panel of experts, which included Charlotte Bourne, Mark Tallman, and special guest Stephane Hamel, shared insights into some of the most common reasons why so many dashboards fail, and how to save yourself from these pitfalls to set up dashboards that deliver actionable insights … Read Full Post

Data Driven Attribution or Media Mix Modelling?

Data Driven Attribution and Media Mix Modelling (MMM) both help marketers understand marketing channel performance and return on investment. While it’s true that both help to measure marketing success, they differ in their data requirements and capabilities. If you are trying to figure out which one is best suited to your organizational needs, read on. … Read Full Post

Benchmark Your Marketing Analytics Maturity

See how your marketing analytics performs against thousands of organizations. (Approx. 5 minutes).