Zero to ML Hero: Incremental Approaches to Machine Learning

The potential applications of machine learning (ML) are nearly limitless. And when done correctly, leveraging artificial intelligence (AI) within a business context can drive strategic growth. However, getting started with machine learning doesn’t have to be an overwhelming undertaking. AI/ML products are just features. Bringing one to market doesn’t mean you need to redefine your whole business; it will just make your current product offering better and your customer’s lives easier.

In this blog post, we’ll uncomplicate incorporating machine learning into your product offering. From chatbots to improved search capabilities, you’ll learn how to start small and deliver iteratively so you can leverage this powerful technology to drive big value.

Collect User Feedback to Identify Where ML Might Help

We’ve said it before and we’ll say it again: the best time to plant a tree is ten years ago, and the next best time is today. Ideally you should already be collecting customer data and user feedback, but if not, start doing so immediately. 

ML algorithms learn from data, so implement tools like Pendo or Google Analytics to track user interactions in your product. This will help you identify friction points in your user journey and the tasks within your product that users find monotonous or time consuming, such as redundant data entry.

You can even start simpler and conduct user interviews or surveys. Reach out to your customers and get as many as you can on Zoom meetings. Have them tell you a story about the last time they used your product and ask where they were getting hung up trying to achieve something with your tool. You can collect similar qualitative and quantitative data using Google Forms, SurveyMonkey, or Typeform. You may even find that your users are more likely to participate in a short survey than hop on a Zoom, but you can usually suss out friction points better when speaking to them live.

Remember, practicing user-centered design and speaking to actual users is a team effort. Engineers and product managers can help conduct user research in addition to product designers. Engage with your users early and often so you can find innovative solutions (like implementing AI/ML into your product offering) that meet both user and business needs.

Use ML to Improve an Existing Feature or Build a New One

Once you’ve identified friction points in your user journey, use our PEMDAS for Product Management framework to prioritize which features to improve/build next. If the majority of your users have vocalized the same frustrations within your product and solving those challenges with machine learning will drive business value, start there. Below are some ideas for how to improve existing features or build new ones using ML.

Search

If you have search anywhere in your product, use vector embeddings and databases to improve search relevance and make better recommendations to your users. Vector databases make it easier for machine learning models to remember previous inputs, and embeddings capture relationships and meanings of words. You can build your own vector database (or we can help!), but there are a lot of existing providers out there like Pinecone.

Smart(er) Autocomplete

You can also improve the search functionality of your product by building autocomplete-powered user interfaces. This open source, product-ready JavaScript library allows users to type an input and the autocomplete “finishes” their thoughts. It minimizes typing, helps users find what they’re looking for faster, and exposes them to other interesting searches they may not have initially thought of. Autocomplete is pretty much a universal feature across search engines, so why not give your users something they’ve almost come to expect?

Anomaly Detection

If you sell a finance or healthcare product, anomaly detection can help identify unusual data patterns that could be indicative of fraud or health changes. These ML algorithms can filter through unstructured data and determine what is normal vs. not, helping you get ahead of issues and communicate important information to your users. Anomaly detection has broader applications beyond cybersecurity including quality control in manufacturing and food production.

Chatbot

AI-driven chatbots are designed to mimic human conversations and streamline the process of your users receiving product support. Chatbots will scan your website, your help center, or other resources you designate to generate helpful answers to your users’ questions. They’ve come a long way since the days of getting frustrated with generic, robotic responses and when done right, can now act as virtual assistants. We have experience building these too and actually built our own Slackbot to help with user research.

Transcription

If your users have a need to convert audio or video to text in real time, AI-powered transcription can do so much more than turn recordings into editable text. Implementing speech-to-text tools can arm your users with better insights and action items. These language processing models can recognize patterns in speech and help your users improve the way they talk on sales calls, better remember meeting takeaways, and more. There are plenty of APIs/open-source speech recognition systems out there (like Whisper) that you can customize to fit your needs. We’re actually working on something exciting that uses an existing model framework. Stay tuned!  

Image to Text (and Text to Image)

If your users are frequently uploading PDFs, documents, or even pictures of handwritten notes to your software, make their lives easier by implementing an image-to-text AI tool. For example, a user may collect business cards at a conference and they need a list of email addresses to send a mass follow-up email to. These deep learning models will extract the text from images and store that data in your software. The inverse also exists in the form of text-to-image generators, which can turn text prompts into images. We’ve actually built one of these ourselves.

Churn Detection

Incorporating ML into your product offering should make your users’ lives easier, but it can also have business benefits. You can use AI to analyze customer interactions within your product and detect behaviors that indicate churn risks, such as less transactions, no logins over a given time period, etc. Churn detection goes a step further than user analytics and helps you implement retention strategies to improve customer health and increase ROI.

Companies That Have Incorporated ML Into Their Product Offerings

There are a lot of big-name tech companies out there that also understand this notion that AI/ML doesn’t need to redefine your product offering, but it can change your business for the better. By improving or introducing one small feature, each company has been able to make their customers’ lives easier while still staying true to the core value their solution offers.

HubSpot recently introduced content assistant, which helps marketers write more engaging emails that will resonate with potential customers. You can rewrite text, expand or summarize it, or change the tone.

Dovetail is a customer insights hub and one of our favorite tools here at Crafted. Last year, they incorporated AI-powered transcription into their offering. It helps users make sense of large amounts of data by automatically clustering qualitative customer feedback.

Adobe is no longer just a PDF reader and photoshop software. They recently released Firefly, which allows users to use generative AI and text-to-image prompts to create unique images and take photoshop to the next level.

Conclusion

We hope that by sharing these incremental approaches to machine learning, you’re less intimidated by the endless possibilities with AI and are able to hone in on solutions that will enhance your user experience and drive business value.

And while there are a lot of open source models/APIs out there that you can customize yourself, the Crafted team has experience implementing ML models across business of all shapes and sizes. Whether you need help getting your data ready, identifying the right model for your business, or productizing a model, reach out and we’ll take you from zero to ML hero!