McKinsey report on application of Neural Network tech in industry


#1

This is a pretty useful report that gives good insight into various use-cases and the opportunity. We can likely expect that use cases with a more viable path to return will be easier to pitch. Also worth noting is that the authors recognise the outlay in inhouse dev, so suggest AIaaS as a potential path to follow but limit this to “cloud vendors”. Finally it’s worth reflecting on the challenges regarding regulatory concern, data privacy, etc. I would suggest that in the agent classification ontology we don’t just include functionality, interfaces, data types, etc. but other aspects for selection e.g. hosting security, hosting location (for EU data concerns), datasheet on the data (From Gabriel’s link), etc.

Anyway The Report


#2

Hi @Matthew,

Have you explored some of the other accounting big 4 or major consulting firm reports on AI?

I think you’d find them really interesting. There are a ton of market opportunities to use AI to automate workflows, and they’re just scratching the surface.


#3

I have indeed. I get quite a lot of stuff from them and Gartner, etc.

Here’s what I’m seeing. RPA is basically following hard in the footsteps of outsourcing. These companies vary from snake oil to actually using some ML to provide a degree of intelligence in the processes. What I’m seeing consistently, though, is that the pitch is pretty much only cost reduction.

With outsourcing there was this whole “get access to skills, focus on you business, we can deliver it better, not just cheaper”. The vast majority of the times the business case was tested and it came down to a finance manager shouting “I was promised $x take-out 3 months from now” leaving some poor Indian dudes to pick up how to deliver 24x7 with 3 grads in Bangalore, but still, at least they walked in with bigger ideas.

Talking to the RPA firms the best fig leaf they can offer is “this will free up people to do more value add things”… my arse it will.

I’m also seeing good application within specific verticals with more “spot” solutions. Interesting stuff but I think the next 18 months will see these meet with the RPA/ML offerings.

It’s a bit the same as computers where they kind of got the home market and trivial business users with PCs, whilst at the same time address niche challenges in various verticals where the mainframe was installed with a specific use-case in mind.

IMHO what joined these two markets and formed the current market where everyone in the company has a laptop is the democratisation of professional tools. Microsoft Office is incredibly key here. Before it was widely used a manager would have a secretary for typing letters, a PA for keeping the diary and sending letters (much reduced nowadays, though not eliminated), a design department that would produce presentations, the guys crunching the numbers, etc. Tools like Photoshop, etc had an impact, but for me MS Office is the big one.

This has had a downside too. Previously there would be a handful of presentations required and a few letters per day. Now managers spend hours creating and viewing PowerPoint, and reading hundreds of emails. Value add? Not sure.

Where does this apply in AI? We need to put the tools into the hands of average users. Those power applications are great, as are the trivial use cases and semi-AI uses, but the big power comes in giving those tools to everyone. Your average business user having the power to just get human-beating decision calls in a portable device would be incredibly powerful. But here’s the rub: coupled with process automation, it wipes out so much of the work done in a company. Departments like treasury, HR, strategy, MarComms all get wiped out when a CEO can get better hedging, hire/fire, market projections, messaging etc. from an AI.

We’re a way from being there, but for me the winner in AI is going to be a company who can provide that “portal” to the average user where the power is in their hands. FOr me that’s not about being the best at the technical back-end, necessarily, in the same way as Apple uses 3rd party hardware. I think SNet can get there, but will need laser-like focus to do so. That middle ground is the key.


#4

@Matthew you are onto something here. I’ve often thought the biggest issue most companies face when dealing with data is actually the creation and storage of data sets around particular problems that need solved. For example I work in a very high volume semiconductor manufacturing plant and you would think data organization is something that would be highly valued but whenever a problem occurs the human element still needs to rummage through multiple different data locations just to compile a data set for analysis. Then translate that information into some form of report usually power point, just to get stored in some other file system that has no relevance to like data. We have more analytical tools than you can shake a stick at but the real time consumer is the data set itself and this tends to drive under production due to the repeating nature of the issue.