Guest Blogs

The Struggle is Real – 3 Considerations to Make Machine Learning More Effective

customer experience services

Can machine learning enable success? As marketers, we all want to provide great customer experiences, scale our programs, drive improved outcomes, and yes – be more efficient. It’s hard for me to think of any marketer that doesn’t want these things. The key is identifying the best way to do so. Machine learning solutions are often discussed as a popular approach because they are based on data and -in theory- the intelligence can be applied quickly. A simple example is providing someone the next best offer based on a prior action. Yet, many of us can feel stuck getting the most out of machine learning initiatives. Here are some of the common roadblocks:

  • Special skills required – i.e. what if I don’t have a team of data scientists?
  • Extremely time-consuming – I may have the people, but the data cleansing is taking forever.
  • Huge deployment challenges – ‘nuff said – you get it if you have done this before!
  • The bar has been raised on data quality – the model depends on good inputs – what if my data is messy?

For machine learning to be a go-to approach, here are a few simple considerations that I have seen make a big difference when evaluating or using machine learning tools.

1.    It all starts with the quality and diversity of your data. If your data is a mess (not unified) or is incomplete (only a few sources), your machine learning output will be suboptimal. That’s why you have to get data collection right. It is important to include first-party and third-party data – or any key sources where your organization is interacting with a buyer (ex: call center data). Make sure the data is tagged correctly and then is normalized and enriched. Even the best models will generate undesirable outputs if the input dataset is incomplete or the data has not been standardized.

2.    It helps to have a hypothesis. You don’t want to just start wading around in data wondering what might be interesting. Machine learning algorithms look for patterns in the data, but you’re likely aware of many of these patterns—it’s just too much data for humans to sift through. Start with a feasible premise and then let that drive the model. Sometimes teams are looking for big wins, which are great – but it’s important to have a mix of small wins too – incremental improvements are impactful too.

3.    Set your technology up to be flexible. Your business is not static, so it is smart to anticipate and plan for change. It is important to select machine learning solutions that can adapt to whatever new challenge or technology could be brought in.

In summary, machine learning will be part of your strategy at some point if it is not already. All of us want to be able to automate manual tasks, discover insights that will improve the business, and ultimately work more efficiently. It is just like building a house, we all want that impressive kitchen, but if the foundation is shaky – no one is using the stove to make a soufflé. Make sure you have a very thoughtful data strategy that looks to the fundamental need of any machine learning algorithm—good input data—and your results will definitely be better. And, remember it is never too late to get this going. If you need help, check this out.


    Heidi Bullock
    Heidi is the Chief Marketing officer at Tealium. She has immense expertise in marketing SaaS products. She has prime expertise in product marketing and revenue generation across the customer life cycle (acquisition marketing, up-sell cross-sell).

    Previous ArticleNext Article