Feb 6, 2022
Are All ML Models for Support Created Equal?
AI for supportCustomer SuccessCustomer SupportData ScienceData Scientistmachine learningmachine learning for supportNLP
Machine learning (ML) is the buzzword of the decade, with virtually every company claiming (or planning) to use it in one way or another. Whether you are interested in buying an ML-based product or service or building models in-house, you need to be confident in your model’s predictions for you to really leverage them to transform your business. After all, the value machine learning can provide for your business will be limited by how actionable the insights it provides are and how much confidence you have in those insights in order for you to act on them.
In this article I’ll provide you with three key points to focus on to ensure that the ML model you’re looking at is right for your business. They are:
- Leverage high-quality data
- Ensure alignment between the people who create the model and the people who will use it
- Learn and adapt to your specific business needs over time
Let’s get to it!
Leverage high-quality data
What does it mean to leverage high-quality data – isn’t the data you have good enough? Well, the “dirty little secret” that data scientists don’t often talk about is that almost all machine learning models used today are the same! The secret sauce that really sets a model apart is the quality and quantity of data used to train it. Back to your question: is the data you have good enough to make ML predictions? Maybe so. Is it good enough to make good ML predictions that you can rely on to make business-critical decisions? That is an entirely different question.
For example, let’s say that you would like to predict if a support case is likely to escalate ahead of time. How much do you trust the data in your CRM? Does your business have the mature data infrastructure to provide data scientists with accurate, up-to-date information from your CRM in a format they are able to leverage? If you are able to say yes to both of these questions, you are off to a good start, but you are just at the starting line. And if not, it could be months of “digital transformation” for you to even get to that starting line – and your team is overwhelmed with escalations today. Engaging a team experienced in working with support teams and CRM data to get reliable, actionable data in place quickly could be the key path to reducing your time to value – and relief.
What about ensuring alignment between the people who create ML models and those who will use its predictions? This sounds simple – maybe it’s just a few meetings, what’s the problem?
It means that in a time in which more data is being created than ever, (and more all the time) it’s important to spotlight the right data to use in order to filter out all the noise. Furthermore, data scientists will need to know all about where your data is, what the data means, how to interpret it, and what context may contribute to it.
Let’s continue our previous example of predicting support escalations. Starting from the top – what does it mean for a support case to escalate? This process may be obvious to you, but to a data scientist who has never handled a ticket in their life it may be a time-consuming learning process. Maybe the consulting firm that you hired to create the model says they know, but this is one of hundreds of different domains they’re in at any time, and they may not take the time to peel all of the factors that contribute to escalations apart.
How many meetings – and how many weeks gone by – do you want to spend educating data scientists on the ins and outs of your processes and your data? How much time, energy, and dollars would you save working with a data-driven team who knows support better than any other team you can find today?
Learn and adapt to your business needs
Your business – and therefore your data – is constantly changing. Anything from reorgs, new products or support tiers, organic growth of your business, all the way down to simple changes in your CRM itself will change your data.
Importantly, while your data changes, your ML model will not. It will continue making predictions that get less and less accurate over time until enough users complain or maybe even stop trusting your application and look for another solution.
Being able to faithfully rely on machine learning means more than just getting a decent model: it means being able to leverage the power of an adaptive ML platform that continues to deliver accurate results as it becomes more and more critical to the operation of your business. An ML platform provides regular inspection, maintenance, and adjustments to your ML models (yes – you will need more than one!) just like a trusty mechanic keeps your dad’s favorite old car running.
You may have an ML model in hand, but who is going to monitor and tell you when the data has changed and a new model is needed? Who will make sure that it’s making accurate predictions the week of Thanksgiving when you want to rely on it to get your team a much-needed break? Who will retrain it when your “Product” field changes to a “Product Category” field and the predictions go haywire? Worse yet, now your company’s data scientists don’t respond to your emails to train another model for the same old problem when they’re excited about the new research project they just started for someone else. Or that “one-time” consulting fee for the first model doesn’t seem like such a good deal when you have to go back to them every quarter. Wouldn’t it be easier to have a service that maintained your ML model for you? The answer is YES!
Leveraging high-quality data, ensuring alignment, and adapting to your business needs are three areas you should focus on to have the confidence in the ML model. It’s challenging to hit all three. Instead of trying to do this yourself, you want to rely on a group of people who are laser-focused on ensuring that you achieve the gains in customer experience and efficiency you should expect from your ML model.
The SupportLogic SX platform has been created by people who live and breathe support and have been pursuing the best ways for AI, ML, and NLP to transform the support industry for years. SupportLogic extracts Signals from your key customer interactions, predicting with unparalleled accuracy which cases are likely to escalate and at the right time, allowing your service and support teams to engage proactively to optimize the service experience, drive satisfaction and loyalty, and protect and grow revenue – all while reducing the burden support managers face day-to-day which we know all too well.
The SX platform’s machine learning models interpret more than 30 different data points at any time in order to understand the context of a case before determining if it is likely to escalate. The SupportLogic team has a tight web of support and data science backgrounds, allowing for models to be not only accurate but designed for the intricacies of individual support workflows. By offering predictions as part of a cohesive support experience platform, our models are able to learn from the actions you take on these predictions in order to become more accurate and more targeted for your ideal outcomes over time. The result is that your common and avoidable escalations are managed before they arise, leaving support teams to rally in a truly proactive manner for non-preventable escalations. In both situations, your customers receive an elevated support experience.