Why is AI so difficult to implement in Companies?

It’s our fault. We (the AI industry) mythicise it.

AI is not magic. What is mostly known as AI today, is actually Machine Learning. Machine Learning is a computation paradigm that allows one to “teach” a computer something, based on examples.

We don’t teach the computer the theory, we teach it practical examples. And we do that long enough, until we are happy with how well the computer is doing, whenever a new example comes.

That’s it.

So, why is it so difficult? Because data scientists, engineers, the media mythicise this. And because knowledge is power, the experts protect it, because they want to keep their power. Machine Learning is just another discipline of Computer Science.

Translating a business problem to a Machine Learning problem is just another skill that business analysts need to develop.

In my experience, it helps to have a process that supports the journey. The following steps are useful in looking at efficiency challenges and identifying opportunities.
This is not meant to be a recipe for a holistic vision of the future of your business, but a non-exhaustive list of steps that have proven to be useful in the past.

1. Identifying the business challenges

Identifying the biggest opportunities for improvement is the first step

  • Which areas of your organization are the most inefficient?
  • How could you better serve your customers?
  • How can you save money?

2. Come up with possible solutions

After identifying the challenges and opportunities, it’s time to brainstorm possible solutions and approaches.

In the #industrialmanufacturing, a typical challenge is how much energy is spent during the process

Energy is often one of the largest costs in production

  • Can I save energy with better data? What if you could make better energy-saving decisions?
  • If you were to solve the problem, what would you do, how much effort would it take, and how much money would it cost?
  • In insurance, typical questions are: how much does it cost me to evaluate a claim? How can I make this process more efficient?

3. Set priorities

There will be many ideas of how to increase efficiencies, save money, and introduce new revenue streams out of the second step, so the third step is for each of those cases to be prioritised.

A good exercise is to use an impact vs. effort matrix.

So-called low hanging fruit are those that have a high impact with a low effort. These are where you should begin.

 

4. Proof of concept (PoC)

In this stage, you take one of those lowest hanging fruits and implement a Proof of Concept. The goal is to prove that the problem can be solved in the way that it was conceived. A better and more accurate estimate can be done of the effort in the cast of involvement in such a case.

5. Minimum Viable Product (Go live)

Once the proof of concept has been proven, the idea is to go live with the solution. A data-driven solution has several stages of implementation:

  • Data gathering and integration
  • Designing a Data Lake (or a Data Warehouse)
  • Developing Predictive and Machine Learning models to bring added value
  • Deliver insights in the form of visualizations, alarms, recommendations, or integration with current IT systems and business processes

6. Improve. Improve. Improve.

 

Going live is not the end of your work. It is only the beginning. You’ll only understand the benefits of the new solution once it’s in your users’ hands.

After that, a continuous improvement process is key. Save budget for this, since it is as important as the conceptual and project phase.

Paulo Nunes

I'm an entrepreneur and AI enthusiast, CEO of Two Impulse