5 Machine Learning Challenges for Customer Support
Do you run a support organization that would like to improve by applying machine learning technology to your operations?
Do you want to make your team more efficient while at the same time delivering a better and faster customer experience?
You need specialized tools to make this happen. You need machine learning that is purpose built for customer support.
Here are 5 problems that Y Meadows solves, which make it the optimal solution for the customer support environment:
1. Multiple Products
Support messages for different products need to be directed to different support teams, so their resolutions depend on the product involved.
We can build a “hierarchy of models”. The first model will determine the product. Then a second model takes over to determine the exact category. Think of it like a library system — first you determine the major topic, then the subtopic.
2. Signature Blocks
Some customers' signature blocks may contain whole paragraphs of text (i.e. confidentiality), which can "confuse" standard machine learning models.
The Y Meadows processing pipeline can strip out the signature block language. It won’t confuse the models when they get trained, and we continue to strip out the phrases in operation.
3. Unbalanced Categories
Certain categories get many more messages than others, but that doesn’t mean that they are more important. Having some large categories and some small categories will lead machine learning to “overweight” the large topics.
The Y Meadows model training system will augment the smaller categories to more evenly distribute data across all topics. We do this by synthesizing data that looks very similar to the actual messages. The result is a model that can correctly understand messages in all types of categories.
4. Very Sensitive Topics
You might have certain topics that need urgent attention, and you never want to miss those. Messages about potential security issues are a good example.
Y Meadows allows you to set the “sensitivity” level separately for each topic.
5. Extracting Information
To answer questions, we need to understand the topic, but we also need to extract information from the message: names, dates, phone numbers, amounts, etc. Entity extraction is not part of most machine learning models.
Y Meadows includes a three layer extraction system. It automatically finds standard information (i.e. phone numbers), uses pattern recognition to identify entities (i.e. product codes), and can be trained to gather other types of relevant data. Once extracted, the information becomes available and used by our engine to resolve customer issues.