Adopting AI In Your Business –  How To Get Started

A quick guide for companies that want to explore the possibilities of implementing Artificial Intelligence (AI) in their business, but don’t know where to start.

The following guide was produced, based on our experience and insight as an AI – vendor. See our cases here. FPM (First Principles Modelling) offers visual AI (Computer Vision) solutions, which is our reference frame, but the guide can be transferred to most AI-projects. 

The Importance Of The First AI-Project

The first Computer Vision project often defines how your organisation evaluates the value of AI. Successful implementation and positive value creation will clear the way for future projects. Therefore, the optimal first project should have a measurable impact but first and foremost ensure successful implementation.

We must focus on the long term Return On Investment(ROI), rather than the value of the first implementation. Building up long term competence is more valuable than the short term business value, therefor be careful solely focussing on classical accounting ratios when evaluating ROI.

The Process – Implementing AI

The process of implementing AI in your business can be overwhelming, but it does not have to be.

The structure of the process is linear, which means that the first steps of the process will define the later ones. In essence, each step must be fully understood, because it serves as the foundations for the subsequent one.

As a company, you can hire Deep Learning consulting firms, for the various parts of the process but your company should always be involved with capacity in the following steps: “Define Problem”, “Decide on AI Task” and “Understand Constraint”. This is because these steps success comes from understanding the business.

The steps that involve developing the right (machine learning) model and embedding it into a device will often be a part of the process where the external consultants take over unless your company has the expertise in-house.

STEP 1

Define Problem

The first natural step is to find a suitable problem, that you would like to use AI to address. When searching for the most suitable problem, some understanding of the capabilities of the AI-technology is beneficial. The potential value creation becomes clear when you can connect your business with the technology. You do not need to become an expert but even a small amount of technological knowledge combined with deep, hands-on understanding of your business will do the trick.

This is also the reason why on our homepage we offer basic insight, that can create a fundamental understanding, wherefrom ideas can rise. Read more.

The first problems you address will surely not alter the fundamentals of your business, but it is the beginning of the evolution of your company and realization of its AI potential.

STEP 2

Decide on AI Task

An expert level of understanding of the business problem is essential when we combine the problem and AI-technology. This the important part, when you merge a thorough understanding of the business with technological capabilities and create the methods to be used. There are numerous tasks that can be solved with AI; we list some diverse examples of Computer Vision solutions here.

STEP 3

Understand Constraints – Relate It To The Problem

The understanding of the constraints that comes from applying technology to a business problem is important. The technology has limitations and the performance will, in certain situations (e.g. compute intense AI), be a trade-off of speed, precision and size of model.

To illustrate the trade-off challenge some examples have been listed:

  • If we want a diagnostic tool for cancer, we will focus on accuracy more than speed.
  • If a production line goes at a certain speed, we must value speed and accuracy over the size of model.
  • If we want to embed the solution into a small offline device, we must focus on a smaller size model (algorithm) and less on speed/accuracy.
  • If you want the fastest, highly accurate and complex AI-solution embedded to a small offline device = Not possible
  • In most cases, solutions require for a mix of the key facets listed above, but you should always evaluate what is important for your company now and in the future regarding performance.

STEP 4

Develop the right Machine Learning Model

When we understand the constraints, we can start to develop the right Machine Learning model, that in the most optimal way complements the demand for performance.

Depending on the problem you want to solve, it’s worth noting that some topics have been scrutinized for years and there are sometimes free models available for download. These models are often trained to do a lot of different tasks, which also means they are not optimized for a specific task, which lowers performance. To implement these free models, it also requires knowledge of technology. The value creation of these free models often does not match expectations.

Most of the problems that companies want solved are not standard but related to case-specific tasks. This means a solution must be created that can perform the desired tasks. When developing models with Deep Learning, they are optimized to solve a specific task and thereby greatly increase the performance. As an example, developing a visual quality inspection will only work on the case-specific object(s).

The good news is that technology has evolved, in a way that dramatically reduces the price of custom Machine Learning model creation when using Deep Learning, which means it is broadly accessible.

FINAL – STEP 5

Embed Machine Learning Algorithm Into Device

When the model has been created it is time to embed it into the desired device.

Depending on the size of the model, you can embed it into a local server, cloud solution, single-board computer or microchip. To maximize speed and accuracy, you can use cloud or a server and, when implementing into smaller offline devices (edge computing), you can use a single-board computer or microchip. At present, the possibility of embedding useful solutions into a microchip is mostly in the experimental phase

Good Advice Before You Start With AI

At the current time, the AI-marketplace is overflowing with buzzwords and broken promises. Yes, there good consultants and vendors, but there are also a large number of people with only a little to offer. This is an important fact to realize when dealing with vendors and consultants. One main problem when buying an AI solution is that a product may work in the testing phase, but fail when applied in the real world. Successful live testing should always be a condition when buying a product.

Now, Start Using AI!

The world of AI has evolved in a way that lowers prices and enhances performance. It is not only accessible to technologically-advanced companies but is available for everyone who wants to realize their AI potential.

In the coming years, AI will become widespread and an a essential competitive differentiator across industries. For the most part, AI will not simply be plug-and-play. Instead, you must develop skills within your company that can drive forward its evolution. The easiest way to get know-how is by implementing AI-projects and then building on the experience gleaned. The earlier you start, the further ahead you will be compared to your competitors. How long will you wait?