Identifying a SaaS Market That Machine Learning Can Address | Hacker News Reader

Identifying a SaaS Market That Machine Learning Can Address

tomtunguz.com

61 pointsgk122 comments

I'd say that by far the greatest potential in process optimization still lies with improvements in areas that don't require machine learning, such as:

- fully or partially paper-based processes

- digital systems that can't easily exchange data with one another

It always makes me want to cry when I sees systems where due to the lack of a proper interface between software components the API between these components essentially can be described as "Print it out on paper and hand it to someone else to type it into another software."

You don't need machine learning for that. You need APIs and first and foremost you need people to talk to each other and agree upon joint requirements for different components and the system as a whole.

chances are that if they haven't automated it, it's not revenue generating or important enough to invest capital to fix it

Or it is worth fixing but nobody can be bothered to collect/review the data showing the cost of doing nothing.

Could be but then machine learning doesn't help either.

There's still a lot to fix and automate though with a significant ROI.

Did you know for example that the paperwork required for sending a shipping container from Shanghai to Rotterdam is more expensive than the actual transport of the container itself?

Freight forwarder Maersk considers this to be a painful enough problem to try and solve it with Blockchain technology now (one of the few cases where using a Blockchain actually makes sense because not all participants involved in the process necessarily trust each other).

This is demonstrably false, since it can be said of the problems addressed by any new product, and new products succeed all the time.

are there still problems like this? doesn't everyone already know the this is a saas startup now? wouldn't have all of the inefficiencies already been poached? if not I'd love to know about some

There is absolutely still a host of problems like this and no, there are still many inefficiencies waiting to be poached.

Often these are internal, sometimes highly customer-specific products though.

In many cases writing a simple interface between two existing components unfortunately won't do because these components were either never meant to have external interfaces in the first place or the original programmers aren't available anymore and nobody really knows how to change these components.

So, your only option is rebuilding the whole system from scratch, which is no easy undertaking, given that these kinds of systems often have accumulated decades worth of knowledge and business logic.

> Consequently, if you looking to build a machine learning based SaaS company, find a really expensive internal process and automate it.

so basically the "how to" is "find a problem and improve it"

this falls short of the expectation I had going into the article...hoping this was some sort of framework to identify markets

I feel like this is clickbait

Most of Tunguz’s posts are like this. His analysis of SaaS startups are very meta and a few sound like a armchair quarterback. His style is not for everyone and if you are an entrepreur, his advice is rarely practical and specific.

TL;DR “find a really expensive internal process and automate it.”

Do you need AI for that ?

No but you can charge more if you call it AI.

The snazzy misnomer of the day is “robotics”, short for the more accurate “robotic process automation” or “RPA”:

https://en.m.wikipedia.org/wiki/Robotic_process_automation

// Note: If you’re interested in getting paid to drive automation efficiencies at massive scale with any of RPA, AI, ML, whatever, I’m hiring.

We'd been hearing about this "robotization". At some point our team leader talked in depth with some prospects and brought back the good news -- hey, this is what we were trying to sell all along.

So now we're robot makers apparently.

almost any software company?

> To date, most software imbued with machine learning reduces costs rather than increase revenues.

Source?

ML includes trading algorithms, recommendation systems, self driving cars, search, translation, asset pricing, etc., all of which are revenue generating.

The point is that all of those processes were already being done, just in a less efficient manner.

That can be said of any product, not just ML.

This article is nonsense. Grammerly or Amazon Echo are examples of how AI can do _new_ things in a frictionless way. Reducing friction is a huge use case for machine learning. Lowering costs in this splintered economy of haves and have nots is certainly valuable but hardly the most interesting or valuable opportunity for ML.

Amazon echo isn't really AI. They use something similar to intents https://dialogflow.com/docs/intents for chat bots, if I'm not mistaken

I think people, including the author, vastly underestimate what will happen when things get cheaper. I think smart speakers are still a case of transcription/understanding being cheap enough that we can do it for trivial things like turning on the lights.

But I think ML is even more powerful, in situations where we learn from real world data rather than some human/programmatic data labeling, we can exceed the abilities of people. Google's recent ML for health stuff comes to mind where it's much better at humans at certain tasks.

With Machine Learning, obviously!

Meta Machine Learning.

Reply on news.ycombinator.com