Data science for marketing

Share this article

Are you wondering what problems data science teams can help marketing teams solve? Or maybe you are more interested in exploring specific examples of data science projects for marketing? Well either way, you are in the right place!

In this article, we tell you everything you need to know about data science for marketing. We start out by going over the types of problems that data science teams can help marketing teams solve. After that, we discuss some metrics that data science teams can help marketing teams move. Next, we provide specific examples of data science projects for marketing. Finally, we discuss what types of data are most important for data science teams that support marketing teams.

Note that throughout this article, we primarily focus on marketing efforts that are targeted at new customer acquisition. We will cover topics such as lifecycle marketing and reengagement of customers that are already actively using a product in a separate article. The reason that we separate these topics into different articles is because there is a large difference between the type of data that is available on customers that are being acquired for the first time and the type of data that is available for existing customers.

Why is data science important for marketing?

What kind of problems can data science teams help marketing teams solve?

What are some examples of problems that data science teams can help marketing teams solve? In this section, we will provide information on specific problems that data science teams can help marketing teams solve.

  • How to measure campaign performance. The first problem data science teams can help to measure campaign performance and provide visibility on how different campaigns are performing. This is crucial for helping marketing teams understand how their campaigns are performing and which marketing strategies are most effective.
  • Who to target in marketing campaigns. The next problem that data science teams can help marketing teams solve is deciding who should be targeted in marketing campaigns. Certain populations are more likely to respond well to marketing campaigns than others and data science teams can help marketing teams understand who those populations are.
  • What messaging to use in marketing campaigns. The next problem that data science teams can help marketing teams solve revolves around what messaging and creatives to use for specific marketing campaigns.
  • When to target specific prospects. Data science teams can also help marketing teams understand when they should be targeting specific customers. Just knowing which customers to target is often not enough, you also have to understand when the best time to target specific customers is so that you can target them right when they need your product.

Metrics data science teams can help marketing teams move

What kind of metrics can data science teams help marketing teams move? Here are some examples of metrics that data science teams can help marketing teams improve.

  • Number of leads generated. One important metric that data science teams can help marketing teams improve is the number of leads generated. If marketing teams are able to target the right prospects with the right messaging at the right time then they will increase the number of leads that are entering their customer acquisition funnel.
  • Number of or percentage of leads who become paying customers. It is not enough for a lead to just enter your customer acquisition funnel. The presence of that lead in your funnel does not contribute much to the business unless the lead actually onboards onto your product and becomes a paying customer. Data science teams can also help to optimize the number of leads that become paying customers by helping to ensure that marketing teams are investing in marketing channels that bring in serious prospects. Data science teams can also help design retargeting campaigns that encourage leads to progress through the customer acquisition funnel.
  • Return on investment. You can always get more leads into your funnel by spending more money on your marketing efforts. Another important metric that data science teams can help marketing teams optimize is the return on investment that the team gets on their marketing spend.

Examples of data science projects for marketing

What are some examples of data science projects for marketing teams? In this article, we will provide multiple examples of data science projects that can help marketing teams operate more efficiently and effectively. We will specifically focus on projects that require advanced modeling or a deep knowledge of statistics. We will break these projects into different categories based on the main problem that they help marketing teams solve.

Problem 1: How to measure campaign performance

We will start out by discussing how data science teams can help marketing teams better understand the performance of the campaigns that they are running. We will specifically focus on data science projects that may require advanced modeling or knowledge of statistics. This means that we will not discuss things like straightforward experimentation and AB testing.

Project 1: Distributing credit for leads that saw multiple campaigns

Understanding which customers can be attributed to which marketing campaign is an important part of evaluating how your marketing channels and campaigns are performing. It is the most straightforward way that you can understand which marketing channels are bringing in serious customers that are actually signing up for your product.

It is fairly straightforward to just take the last marketing campaign that a lead saw before signing on to your product and give all of the credit for bringing that lead in to that final campaign. However, this is a flawed approach because it does not account for the ways that the other campaigns influenced the customer. Maybe the customer would not have actually signed up unless they had already seen a few previous campaigns that convinced them of the value of your product.

A multi-touch attribution model is a common type of data science model that solves exactly this problem. It can be used to distribute credit to the different touchpoints that merchants had with different marketing campaigns. These types of models are generally built using data on which marketing campaigns leads interacted with before they signed up for the product.

Project 2: Understand the impact of campaigns that cannot be run with simple AB tests

The next way that data science teams can solve problems that marketing teams face is by helping the marketing team understand the impact of more complex campaigns or events that cannot be analyzed using simple AB tests. For example, if your marketing team is planning a campaign of TV commercials where you will not be able to tell exactly which incoming leads were exposed to the campaigns and which were not, you might need to do some more advanced modeling to understand the impact of the TV campaign.

In these situations, it is common for data scientists that have experience in causal inference to intervene and help to measure the performance of the campaign. For this specific example about TV campaigns or campaigns where you cannot determine exactly which leads saw the treatment and which leads did not, interrupted time series models are sometimes used to understand whether there was an uptick in the number of leads in areas where the commercials were aired compared to areas where the commercials were not aired.

Project 3: Estimate the value that leads will bring

Return on investment (ROI) is an important metric that must be kept in mind for paid marketing campaigns. Most marketing teams are interested in understanding things like the amount of revenue that leads will bring to the company so that they can better estimate the amount of revenue that is brought in per dollar of marketing spend.

The problem with calculating return on investment for leads that are brought in via paid marketing channels is that you have to wait a whole year (or longer if you are truly interested in lifetime revenue) to see how much revenue a customer actually brings in. It is not feasible to wait this long to understand your return on investment because then you will not be able to update the way you distribute your paid marketing budget to make things more efficient.

A customer lifetime value model is a model that is created based on information that is known about a customer to estimate how much value they will bring into the company before they actually bring in that money. This type of model is often used to help marketing teams understand how much value a customer who has only been using your product for a few weeks will bring in over a longer period of time. This allows marketing teams to better understand the return on investment they get on their paid marketing campaigns.

The exact data that is used to calculate lifetime value for leads can vary from use case to use case. In general, these models are built using information that is gathered during the customer acquisition process like demographic data that is collected via sales interactions or application forms and website interaction data. It may also include some information on what the customer did in their first few weeks on the platform.

Problem 2: Who to target in marketing campaigns

We will now discuss data science projects that can help marketing teams understand who to target in marketing campaigns. This includes projects that help marketing teams determine which specific individuals they should be targeting as well as projects that help marketing teams understand which marketing channels they should be focusing their effort on.

Project 1: Understanding which marketing channels drive outcomes

For some marketing channels, it is straightforward to understand what kind of return you are getting for your investment. For example, if you are running an email campaign to retarget leads who have already provided their email address then you can probably see which leads opened and clicked on your emails. This means it is relatively straightforward to see which leads engaged with the email and returned to your customer acquisition funnel after the email was sent out.

For other marketing channels, it is not so simple to attribute specific leads back to traffic acquisition sources. For example, if you run TV advertising campaigns then it is hard to know exactly which leads saw your advertisement. This makes it hard to calculate your return on investment and even harder to understand how your return on investment compares for different marketing channels.

So how can data science teams help to solve this problem? Data science teams can help bring more visibility to which marketing channels are driving outcomes by building a marketing mix model. This type of model usually takes the amount of money spent on each channel and other exogenous variables that may affect the number of leads coming to a website as an input. The output is the outcome that the marketing team is being measured by such as the amount of revenue that is brought in in that time period. By estimating the coefficients that link the input variables to the output variables, you can estimate the relationship between the amount of money that is spent on each paid marketing channel and the outcomes that are generated.

Project 2: Automate bidding strategy

If your company does a lot of paid marketing then the marketing team is likely spending a lot of time adjusting the bidding strategy to determine what customers are being targeted. If you are advertising on search engines like Google, this might involve determining which search terms you want to show up for and how much money you are willing to pay to show up for a particular search term. If you advertise on facebook, it might mean determining how much of your budget you want to spend on people who fall in a given demographic.

If your marketing team is manually performing calculations to update bidding strategy, there is likely a lot of work that can be done to help automate the bidding strategies. Some automated bidding systems are more reactionary and look at historical data to adjust bidding strategies. Some are more proactive and attempt to predict what the return on investment will be for a given bidding strategy in a given time period.

Problem 3: What messaging to use in marketing campaigns

In addition to helping marketing teams understand which customers they should be targeting, data science teams can also help marketing teams understand what kind of meaning they should use to target those customers.

Project 1: Understand who your customers and leads are

The first way that data science teams can help marketing teams understand what messaging they should use in marketing campaigns is by helping the marketing team understand who their customers and leads are. Once marketing teams have an understanding of what types of customers they serve and what their unique problems are, they can better understand what type of messaging will resonate with them.

Models that help the marketing team understand who they are reaching out to are usually built using unsupervised machine learning models like k-means clustering models. These models group similar subjects together so that data scientists can analyze those groups of customers and understand what their defining characteristics are. These models can be created using many different types of data that are available on customers and leads. The exact formulation of the model will vary from use case to use case.

Project 2: Test how different combinations of components in your creative work together

Simple AB testing is useful in cases where you have a small pool of creatives that you want to test against one another, but it is not as suitable for situations where you want to isolate different components of your creative and test how different combinations of components work together. For example, imagine that you were designing an email campaign and you had three different options for the email title, seven different options for the main call to action, and six different options for the main body image. Now imagine that you wanted to test different combinations of these different components to understand which title, call to action, and image combination was most effective in driving conversions.

How would you evaluate which combination worked best? One option would be to run a simple AB test where each combination of the three components was one treatment. However, this would be inefficient and would require a huge sample size because this setup would not take into account the similarities between different combinations. For example, it would not take into account the fact that a few of the combinations had the same title and pool information from those combinations together to more efficiently estimate the impact of that title.

So what do you do in these situations? This is another great example of a situation where the data science team can help the marketing team run more advanced experiments. In this situation, data science teams can build multi-armed bandit models that pool information over combinations that share similarities to help evaluate which combination is best. These types of experiments are typically run in stages where a small batch of combinations is tested then outcome information from that round of testing is used to determine how many people should see each combination in the next round of testing.

Project 3: Optimize discounts and incentives

Discounts and financial incentives are often included in marketing content to help entice prospects who are on the fence about using a product. Deciding exactly what kind of discounts and incentives should be used in marketing campaigns can be tricky. Marketing teams want to ensure that the discount is just large enough to convince prospects to sign on. If the discount is too small, it will not have any impact on conversion rates. If the discount is too large, revenue will take an unnecessary hit.

This is another problem that data science teams can help solve through either advanced experimentation or by building predictive models. There are multiple different ways that this type of problem can be approached, so it is hard to say that there is one classic model or family of models that is used in this situation.

Problem 4: When to target specific prospects

Data science teams can also help marketing teams to determine when they should target specific prospects and when they should not. This is an area where data science teams that work on lifecycle marketing for customers who have already used a product generally have more options than data science teams that work on new customer acquisition. That being said, there are still some use cases that can be addressed in the new customer acquisition realm.

Project 1: Decide when to retarget leads who have not progressed

One way that data science teams can help marketing teams is by helping them to determine when they should retarget leads that have entered the customer acquisition funnel but have stopped moving through the funnel. These may be leads who were initially considering signing on to the product but have not made a final decision about whether they are interested.

Marketing teams should generally be careful about how frequently they target these leads. If they send too many emails at the wrong times, the leads may become annoyed and block or ignore further emails. If you fail to send an email to someone who is in a critical moment whether they can be influenced to sign on to the product, then you missed an opportunity to convert a lead.

Data science teams can help marketing teams to understand when it is the best time to reengage with specific leads. This is another problem that can be approached in multiple different ways. One popular way to approach this problem is to use uplift modeling to identify leads that will be particularly receptive to retargeting.

What kind of data is used by marketing data scientists?

What kind of data is used by marketing data scientists that work with new customer acquisition? Here are some examples of types of data that are useful for marketing data scientists.

  • Paid ad spend. The first type of data that is useful for marketing data scientists is data on paid ad spend. Specifically, it is useful for data scientists to have an understanding of how much money was spent on each paid ad campaign that was run on each marketing channel.
  • Paid ad performance. The next source of information that is useful for marketing data scientists is the performance of paid ad campaigns. The exact type of information that is available will vary by channel. For paid ads that are run on the internet, this might include the number of ad impressions shown and the number of ad clicks for each campaign. For TV ads, this might just be an estimate of the number of people who viewed an ad.
  • Customer attribution. The next type of data that is important for marketing data scientists is customer attribution data or data on which campaigns a specific customer saw before they signed on to the product. For some channels like email marketing, there will be high visibility on which customers received an email. For other channels like TV commercials, this data might not be available.
  • Customer interactions and self-identified attributes. It is also useful for marking data scientists to have visibility on what kind of interactions leads had with the website landing page or sales team.
  • Conversions. In addition to this, marketing data science teams should also have access to information on conversions and outcomes for leads. They should be able to see which leads made it all the way through the funnel and became paying customers and which leads did not.
  • Customer attributes from external sources. If you use third party vendors to enrich the data that your sales and marketing teams have on leads or prospects, then this is another thing that the marketing data science teams should have access to.

Related articles


Share this article

Leave a Comment

Your email address will not be published. Required fields are marked *