The 3 questions everyone is asking about artificial intelligence

Aren't you intrigued by AI’s almost infinite possibilities? Yes, but? We know that some questions still tickle the back of your head. When we attend the many trade shows we go to every year, we listen to you, and we understand. That's why, today, we've decided to address the three most asked questions!

1. What is data? Do I have enough to start a project?

Data is the foundation of artificial intelligence. It refers to the information that AI relies on to learn and make decisions. For instance, if you want an AI model to predict sales, you need to provide it with data on previous years' sales and other variables that may affect the product's sales, such as the weather, the unemployment rate, and so on.

Ensuring high-quality data is vital. If the model is trained with biased or inaccurate information, it may lead to incorrect outcomes and fail to achieve its full potential. Basically, it works like our brains: to give the correct answer to a multiplication problem, we must first learn how to calculate accurately!

That's why data collection and management are so important: data is the raw material AI needs to operate! Data preparation is the most time-consuming step and often the most underestimated. On average, it is estimated that 80% of a Data Scientist's time is dedicated to cleaning and preparing the data. This involves filtering and organizing the data to ensure its accuracy, relevance, and uniformity of format. It's a process that takes time and patience but one that is absolutely essential for training reliable models.

OK, but how much data is usually needed to start an AI project?

This is a question that comes up often. Unfortunately, there's no precise answer: it depends. The fact is, because every company and every project varies enormously, the amount of data you need isn't always the same.

Now, more is always better than less, but don't worry if you haven't started collecting them yet! It's never too late to start storing and classifying your company's data. An AI model may not be perfect at first due to a lack of data, but it can be re-trained over the years and optimized over time with new information. Just one thing to bear in mind: quality is key if you want to achieve reliable results.

Moreover, if you don't have enough data on hand, you can always source it from external databases, either for free or for a fee. You can combine this data with your own to train the algorithms and test the solution.

We can, however, provide you with some guidelines regarding the amount of data required for specific applications. For anything involving machine learning, such as image recognition or a model used to predict when maintenance will be required on equipment, you need an extensive data history with at least 3 years' worth of accumulated data.

To make accurate sales predictions, it is essential to have at least five years of relevant data. The data you use must also be related to the type of model you want to build. For instance, if you wish to forecast monthly sales, your data backups must also be arranged by month. This ensures that you can effectively capture trends and patterns that can help you make informed sales predictions.

In contrast to some AI projects that require vast amounts of data, certain applications, such as operational research, do not need as much. For instance, if you aim to optimize a factory's activities to maximize the use of available resources and time, you mostly need to have precise knowledge of all the variables and constraints to create an effective system.

2. How long does it usually take to implement an AI project from start to finish?

Once again, it depends. The type of project will influence its duration. If you already have a lot of organized data, a proof of concept can be generated fairly quickly. If you don't, systems will have to be set up, and AI scientists will have to fetch the data manually, lengthening the process.

Consider a company that is already using business intelligence tools and has a good sales history. In such a scenario, obtaining a sales prediction model could be done in as little as 2 to 6 weeks. However, suppose a company aims to automate its entire production line and incorporate specialized equipment like sensors or robotic cells. In that case, it will be a considerably longer process that could take a year or even two.

With the right teams and sufficient quality data, an AI project can usually be put into production within a year. At Vooban, projects typically last between 6 and 18 months, plus post-implementation support.

That's all well and good, but what factors are likely to extend the development time of my AI project?

Well, we can identify a whole bunch of them! One major factor is your company's technological maturity. Yes, if you haven't updated the systems you've been using since the 80s, it may be challenging to integrate 2024 technology into your existing processes! Our AI Scientists often emphasize the importance of ensuring that your company's digital transformation is up to date before embarking on any artificial intelligence project!

Secondly, and you may have guessed it, incomplete and poorly organized data can cause project delays. Also, when companies use software that requires logging in to extract data, it complicates and lengthens processes.

Similarly, when businesses impose strict limitations on access to their data, such as requiring specialists to work on their computers within their premises for security purposes, it can hinder productivity. The truth is, at Vooban, we have the necessary infrastructure in place to facilitate secure data sharing.

It's important to remember that AI is often only a small part of a larger project, potentially requiring a team of web developers. The more people working on a project, the longer it takes!

Ultimately, it depends on how much the company needs. It's quite rare that we're dealing with a single siloed project. As we explore the possibilities, we often discover multiple tasks that need to be addressed beforehand or opportunities that can be deployed in parallel.

3. What percentage of ROI can we generally expect from an AI project, and how quickly?

Generally speaking, it is believed that a successful AI project should generate a return on investment (ROI) within 1.5 to 2 years. Having established this, you can guess that the exact timeframe depends on each project. In fact, the question of ROI is a vital one and one of the first that needs to be addressed.

In order to determine the potential ROI that a project can offer, you need to compare the cost of the system against the potential benefits it can bring. You can then adjust the variables, such as the number of people who will be affected, what you want to automate, the type of system you want to implement, and other factors. This will allow you to make an initial ROI calculation, which will help you to compare the pros and cons of different potential projects for your company.?

Well, that sounds easier said than done. Fortunately, Vooban has the expertise to ask the right questions and define the perfect project for you. Questions like what, you ask? What business are you in, what do you do, how many people work for your company, and how much is your annual turnover?

At Vooban, we have developed the Innovation Sprint and Project Plan approaches. At the very start of a digital transformation, this stage takes the form of strategic workshops. These meetings provide an in-depth understanding of the company's specific needs and optimization opportunities, as well as identifying and prioritizing the key challenges to be addressed. During the Project Plan phase, ROI is calculated in detail to ensure the project will be profitable quickly.

What is important to understand is that there are economies of scale in artificial intelligence. For example, for a sales prediction system, the ROI will depend on how many people are currently working on this task and the company's turnover. Suppose one person is in charge of sales planning, and the company makes eight million dollars annually. In that case, the ROI will be smaller than if four employees are responsible for forecasting sales for a company that generates two hundred and fifty million dollars in sales.

In production planning, ROI will depend on the complexity of the planning. The more complex it is, the more it costs, but the more significant its impact on revenues. You get the idea.

And what about subsidies?

We're lucky in Quebec because companies can benefit from government subsidies to integrate AI into their processes. Some subsidies even cover up to 50% of the eligible costs for proof of concept and project development!

I've heard of quick wins...

Some technology projects can indeed come to fruition quickly. Quick wins are often associated with technologies such as Business Intelligence (BI), connectors between various systems within your company, sales prediction, or even mobile applications that can replace paper-based processes. The good news is that your company can implement all these technologies rapidly and generate benefits quickly.?

To begin with, we recommend starting with a quick win, which refers to a simple project capable of generating a good ROI. While we understand the strong appeal of AI as a solution to all problems, we absolutely must resist the temptation to solve everything at once. Our approach is to solve one problem at a time and then reassess the needs for the next project. Adaptation is necessary after each implementation, and proceeding with one project at a time ensures a good experience with the technology.

If we had to sum it all up...

Data is the information required to train your AI model to answer your queries accurately. It's always better to have more data than not enough, and it has to be quality. It's never too late to start collecting it!?

Bringing an AI project to realization can be quick, but it can also take one to two years. Several factors influence can influence the timeline.

Finally, ROI is often seen within a year or two. That said, certain variables specific to each company influence the exact figure. In addition, government subsidies can minimize the initial investment.

These were just three of the most frequently asked questions. We know you have a lot more on your mind, and that's good because we have the answers! Let’s talk!