Data Science

Forecasting Marketing Data At Scale

Written By Annmarie Stockinger & Lindsay Stecklein

What if we told you that you can make informed decisions using a type of data analytics called “forecasting” for the most common business predicaments? And what if we sweetened the deal by saying you could have the answers to these questions constantly at your beck and call?

  • “We’re setting budgets tomorrow – can you have projections ready for the 9AM meeting?”
     – Your Boss
  • “I need you to determine how to allocate all our Paid Search budgets across our 10 business units. Can you have that by tomorrow, too?”
    – Your Boss back again!

It’s common to struggle with allocating proper budgets and setting attainable goals. Often, we find ourselves basing these plans off of strategic guessing, or merely “guesstimates” and a shrug.

It’s NOT common to have a predictive forecasting tool at your fingertips based on a model built for your business’s goals that understands seasonality, trends, and predicts performance.

Stand out from the crowd by having the most informed budgets of them all. See where trends are going ahead of time, and move budgets proactively, impressing your bosses and colleagues. Maybe even start a rumor that you’re a psychic for your knack for accuracy in prediction. We won’t tell them the truth. It can be our secret!


What is Forecasting in Data Science?

While the results may be magical, no psychic powers are needed to succeed. Our solution lies in Data Science. At Seer, we have developed a forecasting approach that lets us aid our clients through meaningful, accurate modeling.

Born out of a need to better understand seasonality and trends, forecasting has allowed us to take a deeper look at anomaly detection, seasonality, and what-if analysis in ways that weren’t previously possible.

Case in point – a while back we had a few clients come to us with questions like “How many sessions can I expect next year” or “Is this spike an anomaly or due to a change we made?” and while we had some analysis skills that helped us answer them, we realized we needed a more robust solution. We needed a solution that could proactively get ahead of these questions.

To answer these questions at scale, we utilized what is called time series forecasting, a statistical technique that looks at historical trends and ebbs and flows and uses that to create a future-looking view of performance.

How Forecasting Works

  1. Understanding (modeling) past trends and occurrences
  2. Predicting (forecasting) the future developed from an understanding of the past

A good way to look at this is, if every year (Year over Year, or YoY), data trends are up and tend to spike at a specific time of year, the forecasting model would likely predict a slight increase and that would be larger during the in-season.

📣 Forecasting can be done across any industry. Here’s a few examples of time series forecasting in practice:

  • Forecasting when doctors from various service lines (Primary Care, Pediatrics, etc) will have the most appointments
  • Forecasting the amount of decaffeinated professionals that will be wanting coffee at a particular coffee shop at various time of year and day
  • Forecasting how many marketing qualified leads will come in daily for a B2B company
  • Forecasting how many sessions and impressions will generate from particular PPC campaigns

Common Challenges of Forecasting

Finding the Right Tool

While some methods are great at finding trends for a single metric like sessions over time, when we started adding in things like channel and product, we realized that out of the box tools in BI platforms like Tableau and Power BI just weren’t cutting it.

Using the Right Data

We also faced challenges when it came to identifying the data we needed to forecast. While the data available was limitless in some cases, we found ourselves struggling to select the data points we needed to check most and isolating where data might be too correlated with another metric for us to get value out of it.

Setting Goals & Expectations

In a forecasting project, there are many parties involved. Depending on who’s involved, there could be a data science team building the forecast, a team that owns the product or products being forecasted, various marketing teams that allocate budgets for these products…just to name 3 key stakeholders at minimum. That means there could be multiple parties all with very different views of what matters most in a forecast, and very different expectations of what a forecast looks like.

Making Decisions with Results

Sure, the need for forecasting had been identified. But how do we make sure that our clients can actually run with this output, and that we can measure results to see if we’re getting the results we want and expect?

How Seer Solved It

Finding the Right Tool

When it came to selecting a model and writing code that could handle combinations of metrics and dimensions we turned towards Facebook’s FBProphet and Google Cloud.

Using the Right Data

With computing issues mostly resolved, we then turned to our Analytics team to define a process that would allow us to monitor the KPIs that were the most important. Using a combination of business sense, data mining, and trial & error we were able to design a sufficient approach.

Setting Goals & Expectations

Talking and asking questions. It sounds simple enough, but it’s easy to forget that what you are expecting on any side of a project can be completely different than what others are.

We had many ongoing conversations with our client stakeholders and internal teams to make sure that everyone was on the same page about outcomes, expectations, and overarching goals. One of the biggest things to keep in mind is that this process will be iterative, and conversations will need to be ongoing.

Making Decisions with Results

This is heavily dependent on conversation and communication. First, we identified how this would be most useful for the client. Then, we built an auto-updating report where the forecast continuously updates itself as time goes on.

Next up, we analyzed this report in conjunction with our Seer PPC team to see how PPC could best plug in and take advantage of trends. Finally, we built PPC recommendations from our analysis, presented them to the client, and are now running with PPC recommendations built by data science!

How are we Using Forecasting Now?

Seer has been using forecasting more and more as our teams and clients hear about this offering. The way we use forecasting varies by client, depending on their industry, organization, and what they are aiming to gain from their forecast.

The first step to using forecasting is opportunity identification and evaluation. This step is crucial to setting a strong foundation for a focused forecasting project. From there, we’ve been able to forecast for clients with very different industries and needs.

Example Forecast – Healthcare PPC

We forecasted sessions, impressions, impressions share, PPC spend, PPC cost, and appointments for a healthcare client to understand how PPC could capitalize on trends, and allocate budgets proactively moving with seasonality.

This forecasted for specific service lines to give granular insight into where to shift budgets depending on the forecasted performance of all involved service lines.

Forecasted Service Line Sessions – Highlighting Seasonal Trends Throughout The Year

Capture

Daily Session Seasonality Across Two Service Lines

Capture 1

Top 3 Tips for Conducting and Using Forecasting

1. Over Communicate with Everyone Involved

Consistently communicate what you are doing (and why!) with all stakeholders involved. Talk to them again. And again. Don’t forget to check in with them regularly, especially at major phases of the process to make sure that everyone is on the same page. Keep them involved as you are choosing your metrics to forecast, as you are writing your forecasting code, as you are building your visual, and as you are conducting your forecasting analysis.

2. Stay Determined in order to Overcome Blockers

Forecasting, while invaluable, has its challenges. Like any data science project, it’s important to remember that it’s an iterative process. There will be blockers that might be derailing, but that’s okay. All blockers can be dealt with – it’s your determination to find a solution that will continue driving the project forward.

3. Keep Your Business Goals at the Forefront

Remember to tackle each phase or step in the process with your mind laser-focused on gaining actionable insights from the data. The more strategic you approach your analysis, the more likely your forecast will be an incredibly valuable resource that helps you make decisions the impact your bottom-line.


🚀 Want to learn more about how we’re using forecasting to help businesses make better decisions? Contact us here. If you are curious about what it’s like to work on the Seer Analytics & Data Science team, we’re hiring Sr. Data Engineers and Analytics Managers in Philadelphia, San Diego, and remote! Apply ASAP.

VIEW OPEN POSITIONS