The Effectiveness of Bundling, Don't Fall in Love with Technology
Tweet of the Week
Companies like Epic and Microsoft beat out competitors because they provide the full ecosystem of tools that conveniently integrate with each other so the users stay locked in. We see this same phenomenon with the Python programming language. It became one of the most popular programming languages in the world because of its strong ecosystem in data and web development which led to the rise of its adoption.
In the NLP space, we also see this with companies like Hugging Face whose transformers library has reached over 60K Github stars, the product-market fit metric of open-source software. They provide modules for every step in the NLP engineer’s toolkit that integrates with each other well.
Musing of the Week - Don’t Fall in Love with Technology
As an engineer, it’s too easy to fall in love with technology. Being in a field like NLP doesn’t make it any easier. However, it’s important not to fall into the trap that better technology results in a better product or improved business outcome. It’s important to deeply understand customers’ workflows, the tools they rely on, how they make decisions, and what metrics they care about. These are the table stakes to applying ML successfully.
For example, I’m exploring a use case for a SaaS product I’m building that involves helping customer support reps resolve support tickets faster with the semantic search of past support tickets. On the surface, it would seem logical to provide a separate tool with a beautiful UI that allows reps to search past similar cases/tickets with great accuracy and learn from them to solve their current support ticket.
Not so fast.
Support reps mainly work in the ticketing software and they need answers fast. So if you’re going to deploy a new search product, you need to do it within the tools they already heavily work with to actually get engagement and provide answers quickly. Otherwise, you negate any benefit you provide with advanced search technology. You also don’t want to search against every ticket because some don’t contain high-quality information to learn from and they’ll just end up polluting the search results. Neither of these problems is an advanced NLP problem but they are just as if not more necessary to the overall solution.
It’s a good reminder that engineers need to be thoughtful not only when they choose technology but how they go about deploying it.