The “IT” Factor: Why Alternative Data is Key to ML Marketing Automation
Machine learning (ML) and marketing operations go hand in hand. But are you confident you have all the data you need to support your marketing automation tools? If you can’t find it from inside the organization, do you know where else to look?
The chances of being able to answer all your business-critical questions with only internal data are slim. Increasingly, you need to incorporate external, alternative data sources to compete and thrive. To absolutely smash your data science marketing automation efforts, your marketing team needs a data acquisition strategy, too.
Machine learning requires great data
Marketing automation (and the ML models that drive these processes) requires accurate, relevant, joined-up data to work properly. And it needs a lot of it.
You need to take control of your own data, leveraging it for predictive insights you can rely on. But you also need a clear understanding of where the gaps are in your internal datasets, so that you can supplement and augment these datasets for better results.
Why do we need a data acquisition strategy?
A data acquisition strategy is really just a fancy way of saying that you’ve actually taken the time to think ahead about what data you’ll need to make your models as accurate and effective as possible. That you’ve figured out in advance how and where to get that data — and how you’ll go about combining external, alternative data sources with what you have in-house.
Valuable information that helps you better understand the behavior of your customers and the performance of your business can come from a broad range of external sources. That includes data marketplaces, data catalogs, other companies, think tanks and research bodies, PoS transactions, social media platforms, official government websites that provide demographic and census data or other statistics and reports, and IoT sensors of all kinds.
This can feel overwhelming, but the key is to follow the concept of “mapping the territory”.
Start with your key hypothesis and/or the main questions you would need to ask the data, to clarify whether or not your hunch is correct. Note that it’s important that you frame these questions as data analytics or data science queries — and that you understand which problems belong to the field of data analytics, and which to data science.
Once you’re crystal clear on the questions you want to answer, you can start thinking about where to get the information you need.
What data would you need in order to answer each of these questions? Who is most likely to have that data? If it’s not something you can access within your own organization, then who else might possibly collect and store that kind of data? In what format? And how can you get to it?
Mapping out exactly who knows what, how these points of information interlink, and how that translates into specific types of datasets will help you to approach the problem strategically, ensuring there are no glaring gaps in your marketing automation machine learning data.
Can’t we just find the data when we need it?
Well, sure, you could do that. But that’s like asking whether you really need to buy everything on your shopping list before you start baking a cake when you could just run to the store every time you reach a point on the recipe that asks for a new ingredient. Nothing’s stopping you from taking an ad hoc approach to sourcing data if you really want to do that, but it’s hard to see the benefit. It’s incredibly inefficient.
It’s much better to draw up a proper strategy, set up robust, built-in processes, and take a proactive approach to sourcing all the data you need before you start building a model. Not only will this ML project take less time to build overall, but you’ll also have set up a system that can be scaled or replicated for future marketing automation applications, too.
What types of data might we need to drive our marketing automation?
Some types of data you can use to supplement or enhance your internal datasets include:
- Footfall (also called traffic, shopper counting, people counting). This measures how many people enter a mall or store, as well as general traffic around a specific region.
- Geospatial. This gives you information about the physical area, whether that’s a district, city, neighborhood, specific building or natural feature.
- Economic data tracks the economic performance of a nation, region, state, or industry. Usually, it’s presented in time-series form and may cover GDP, debt/government expenditure, total manufacturing production, etc.
- Financial. This relates to the financial activities of an individual person, company, or other kind of entity. Typically this covers income, borrowing, repayment history, assets owned, and so on.
- Event data (also called an audit trail) refers to all “events” logged by a computer system. An event is basically anything that happens, and each tiny action performed by a system is recorded in its relevant log. Taken together, you can use event data to analyze system performance, track the impact of external factors, figure out where bottlenecks occur, and so on.
Final thoughts: marketing automation should never slow you down
Fundamentally, marketing automation is supposed to save you time and hassle. It’s supposed to make you more efficient, stretching your vital resources further. It’s supposed to make you more agile and responsive, so that you can adapt to rapidly changing market conditions and seize opportunities as they emerge.
None of that is going to happen if the very processes you use to build your ML-driven marketing automation tools are dragging you to a halt. Or you’re spending so much time tracking down the datasets you need and finding ways to use them that by the time you have any predictions, you’ve missed your moment.
That’s why it’s absolutely essential that you get this right from the outset. That you have technologies in place that serve your goals, rather than creating more problems. A strategy that streamlines your workflow. A data science platform that facilitates and automates connections to external datasets. It’s all achievable, but you must have a plan.