Most companies are actually quite overstaffed for various reasons, freeing up X employees in department Y doesn't mean they slide over to department Z, Z already has more than enough employees usually.
My simple web app, which took virtually no mental effort to create other than learning the source code system and some vanilla Web App / SQL, was treated as a god-send to those employees. It was still being used 6 years later, though I've lost track relatively recently. Always interesting to hear interesting stories like that out there.
My brother's workplace (from my distant vantage point, keep in mind) could have saved millions upon millions of dollars had it had a dedicated technical engineer working on crafting automated reports that connected their massive databases of data to its operational scientists instead of spending 2-4 hours on building the report themselves during the worst time-crunches.
Although in all likelihood, they were fired.
You're unlikely to hit millions like you would from a VC, but you can be solving very tangible problems rather than the stereotypical "Uber for X". And you would be amazed at how much companies are happy to spend on products that work.
The main issue with deep learning is that for a small/medium client, there isn't enough data to train on. Lot of clients want classification software of some kind, but often it's binary and you don't need anything as complex as a CNN.
There are a few big frameworks for vision, like Halcon, which include SVM and neural net implementations (and loads of other stuff). You can provide training examples and it'll do the rest for you. It's not cheap and you have to charge for licenses, but they even have their own little scripting language as well as hooks for C++. The idea is you can use any camera, drag and drop functions and get your solution out. I've never used it, but I've seen their sales demos and they're quite slick.