Top two skills an AI Business Owner must have in 2020 — Says Ashok Samal from Microsoft

Source: Deep Learning on Medium

Top two skills an AI Business Owner must have in 2020 — Says Ashok Samal from Microsoft

Let me start with my personal experience. Its been more than 5 years, I have been doing ML & AI in different capacities. And according to my statistics out of 10 AI initiatives/projects, 2 of them doesn’t survive more than 3 months, 5 of them never comes to live / production (even if the results from POC or POV was super exciting), 1 run into AI ethics and value prop conversation and eventually get killed and rest 2 goes live with a lot of *limitations and challenges. (*limitations and challenges — to know more about AI production limitations and challenges, wait for my next post).

Artificial Intelligence Adoption and success rate in commercial industries

I know, it sounds ridiculous and it results in a huge waste of time, money and manpower. The bigger question is: Why the numbers are so bad and what can we do about it?

Well, after doing multiple brainstorming on this problem and many retrospectives I concluded, the biggest challenge in AI is not technology or python coding or deep learning knowledge. It is a problem that lies in the business layer and most of us don’t spend enough time to address it.

I ‘ll try to take 1 simple example to demonstrate this. During my conversation with a Spanish glass items manufacturing company (anonymized) we stepped into a use-case to detect defective products in the production belt before it could go to the next stage of processing or to the market. The problem was: once a defective product is delivered to the market, the whole shipment has to return back to the warehouse for further inspection and this process costs a lot of money and reputation is at stake. In other words, it is a case of “Revenue leakage”. So, as an AI expert you would jump into the problem and try to fit in some algorithm and vision AI to identify defective products before it leaves the production belt. You would take some sample pictures of good items and bad items and ll run a POC to take some pics and classify it as a good item or a defective item. So far so good and trust me, you can get up to 99% accuracy if you are using the correct algorithm(s). Everyone is happy especially the business owner(s). But the fun begins now. Keep reading…

A typical manufacturing + Production unit.

Now that you have a successful POC in had you want to bring it to live. So, as the very first thing, you need some live pics when the product is on the belt. How would you do that?

  1. Will you put one camera or 5 cameras on the production belt to capture all the sides of the product?
  2. Will you send these pics to the cloud to get it classified or you will run this model on an edge device?
  3. Let’s say you identified a defective item, what you ll do next? By the time AI has identified the item, it would have gone a little ahead in the belt. Will you ask a human to pull this out of the production belt or will you use a robotic hand with automation to deal with the situation.

It turned out, the production house needs to invest significantly to install cameras and robotic hand and synchronize the entire flow. This would also change significantly the way the work has been scheduled and the way workers have been trained to do the job. Turned out, it cant be done now because of huge investment in infrastructure and a huge investment of time & manpower. The board decided the push this to the next fiscal year to consider again.

What just happened here? A POC with 99% accuracy wasn’t able to go live and it is not because of technology. So what went wrong and how could we improve it? If you focus deep on the problem in hand you ‘ll see two major problems in the state of execution.

  1. An ability to Zoom-in & Zoom-Out
  2. ROI Calculation at the beginning phase of planning.

Zoom-in & Zoom-Out: Business owners need to zoom in to see and understand the real issue in hand but also need to zoom out to see the bigger picture and integration of a use-case with rest of the business ecosystem. A smart business owner would bring out the above 3 questions at the idea evaluation phase by using the zoom out technique and force the stakeholders to think about this at the beginning of the planning phase. Zoom-in should be used as a problem-solving strategy whereas zoom out is a strategy to measure consequences once the solution is in hand.

2. Return on Investment

ROI Calculation: Business owners need to understand the real cost of an AI engagement. Sadly 70–80% of them miscalculate this cost which comes in 3 fold.

  1. Cost of the AI solution (Data Engineer, ML Specialist, Algorithms, etc)
  2. Cost of Integration (Integrating to the rest of the business ecosystem. eg. CRM, Apps, hardware, automation, etc)
  3. Cost of production management (Security, Scalability, Audit, performance, accuracy, usability, etc)

These three main factors will decide the total cost of implementing an AI solution. Once calculated, then you have to calculate the outcome of this AI Solution which is generally termed as: Scoping the Impact. In order to do that you need to mainly take care of three factors.

  1. Market Impact because of your solution
  2. Customer impact because of your solution
  3. Revenue Impact because of your solution

Once you calculate both cost & impact of your AI solution then you can easily calculate the ROI to take a decision on whether to go ahead with the use-case or not and remember, this calculation you have to do it at the idea evaluation phase.

In my view, this is the top two skills an AI business owner needs to have in 2020 in order to be successful. Without this, the outcome of the engagement is totally on luck. I hope you enjoyed the post and feel free to like and share it. In case of questions, please comment below or reach out to me on / / +46–734083209.

AI holds an end-less pool of opportunities but the real question is — what you are going to do with it. — Ashok Samal, Microsoft