Comments on: AI Is a Forecasting Silver Bullet, So Why Aren’t You Using It? https://demand-planning.com/2018/12/10/ai-is-a-forecasting-silver-bullet-so-why-arent-you-using-it/ S&OP/ IBP, Demand Planning, Supply Chain Planning, Business Forecasting Blog Wed, 12 Dec 2018 16:30:19 +0000 hourly 1 https://wordpress.org/?v=6.6.4 By: Stefan de Kok https://demand-planning.com/2018/12/10/ai-is-a-forecasting-silver-bullet-so-why-arent-you-using-it/#comment-12250 Wed, 12 Dec 2018 16:30:19 +0000 https://demand-planning.com/?p=7456#comment-12250 The title has two parts. The first is false. AI is no silver bullet, or if it is one, it is missing a gun.

The second, a question” why aren’t you using it?” has some clear answers:
1 – There is some good AI out there. But it is drowned in an ocean of bad offerings. As a potential user, it is impossible to determine the good from the bad.
2 – This is made even more difficult by the enormous hype. Everyone is offering AI. Most don’t have a clue what it is, let alone what are the proper forms to use for the specific types of problems they claim to solve. As a potential user, you cannot even tell who really has AI. Articles claiming AI is a silver bullet do not help in this regard. The author places himself on the wrong side of the hype-vs-credible line.
3 – AI has specific use cases. Using it stand-alone for time-series forecasting does not beat traditional forecasting techniques. Not yet anyway. Under conditions of causality or lack of historical time-series data however, AI can be used to turbo-charge any time-series mechanism to produce great improvements in accuracy. Very few vendors, let alone potential users, have figured out when to use what and when to pass on AI.
4 – One-size-fits-all AI is not ready for consumption, and may never be. Whether automated or not. The first creates a lot of corner cases that go unnoticed but need to be handled by exception. The second requires an army of experts that are highly demanded by everyone. What is needed is specific algorithms, chosen and tuned for specific types of problems. And yes, they would be packaged for automated consumption.
5 – Data data data. AI loves data. Even if the engine is easy to install and run, creating the mechanisms to gather all this structured and unstructured data is the big hurdle. Time will make it happen. It is just not that easy yet.
6 – Sorry to say most companies don’t even have the basics right yet. They need to learn to crawl, then walk, then run. There will be companies on the bleeding edge. These will be experimenting with AI right now. The vast majority are laggards, most have been running 30 years behind the edge forever. Don’t hold your breath waiting for these companies to go to AI.

Apologies for being a wet blanket. But let’s keep the hype down.

]]>
By: Klaus Spicher https://demand-planning.com/2018/12/10/ai-is-a-forecasting-silver-bullet-so-why-arent-you-using-it/#comment-12177 Tue, 11 Dec 2018 19:07:26 +0000 https://demand-planning.com/?p=7456#comment-12177 The report represents a good job selling AI. There are busibess areas (e.g. Financial Data) in which AI-Forecasting now has just has reached the accuracy level of competitiveness compared with traditional methods. (Read Makridakis recent report comparing 8 AI-FC-Systems and 8 traditional systems on 1.000 time series). – AI-FC-Systems are Black Boxes. So you can believe the output – but you will not understand exceptional results. I think in 10 years AI-FC-Systems might outperform the tradional ones. At the moment the AI-FC hype is driving papers not products.

]]>