Artificial intelligence is on everyone’s lips. It is probably one of the most revolutionary technologies ever, promising an impact perhaps greater than that introduced by electricity and the internet. A technology that companies of all sizes want to invest in to ride the wave and not fall behind. Since the release and proliferation of tools like ChatGPT, AI agents and Large Language Models (LLMs) have been at the center of investments by corporations, SMEs, and startups, yet most of these result in failed projects with no real impact on the business. But if AI can truly impact your company’s business processes, why is it so difficult to truly benefit from it? The answer is often very simple yet painful: AI is a tool, and as such, it must be used (and not abused) in the right way, where and when it is truly needed.
Working alongside corporations in various sectors (highway, insurance, food & beverage, agritech, infrastructure and transportation, and manufacturing), at Synapsi we have developed a structured process in 5 fundamental steps that helps us avoid these mistakes.
Start with Problems, Not Technology
Imagine being a company that produces electric drills and wants to revolutionize the market. Knowing the sector well, you understand that most customers look for a new drill when the previous one breaks. You decide to invest in developing a new titanium-coated drill bit to make it 40% more durable than those on the market. Sales increase for a while, until the launch of a new product starts eating into your revenue. This product is 3M’s adhesive strips for hanging pictures, which allow your potential customers to solve the problem of hanging a picture without the hassle of drilling a hole in the wall. You have just learned one of the most famous lessons in marketing, elaborated by Theodore Levitt.
“People don't want to buy a drill with a 6mm bit. They want a 6mm hole in the wall.”
It’s a simple concept, yet often overlooked by those proposing innovation and technology to businesses: SMEs are not looking for chatbots, algorithms, or generative models as ends in themselves. What they really need is to ensure operational continuity, reduce errors, serve customers faster, and manage costs effectively. Artificial intelligence can certainly be a valuable tool to achieve these goals, but it is never the starting point. When this fundamental dynamic is lost sight of, the risk is that AI becomes yet another generic promise, a budget cost with no real return on investment.
The challenge is not to be the first to apply artificial intelligence. The real challenge is to understand where you are employing people to perform tasks that a machine could do better, faster, or with fewer errors? Only from here can a true reasoning about AI adoption begin. Otherwise, you end up integrating technologies that no one needed, just to avoid seeming late. AI cannot replace people, but it can empower them by making them more productive.
Imagine being a company in the insurance sector that employs assessors to analyze images of claims. Each time the operator analyzes a new claim, they must look at multiple images to verify the license plate and chassis number of the car. They must then assess the damage and manually enter the data into the system. Imagine this process repeated dozens or hundreds of times a day. How does this impact the business? Now imagine being able to extract all this information automatically, so that the processing time for a claim is more than halved. This is a reduction in annual costs and process efficiency that allows for faster service to customers. This example, inspired by a use case from one of our previous clients, is a clear example of how AI can have a real positive impact on business.
Set Clear and Achievable Goals
A false myth to debunk is that artificial intelligence can solve any problem. Too many think that simply feeding any input into these systems will make them learn to solve our problems. This is absolutely not the case; in fact, among experts in the field, a mantra often prevails: “garbage in, garbage out.” If we input any kind of data into an AI system without ensuring quality and without a structured process, we will get a system that returns answers of the same quality. If you have identified a business process to work on, the first goal is to understand whether the data you have is usable for training an AI algorithm or if you can use one already trained on your data. If the answer is no, you must start from the basics. A well-structured data acquisition process can bring enormous benefits. In this case, you can identify areas for improvement to obtain cleaner data. For example, imagine having videos captured by a camera installed on a fleet of vehicles. In an initial phase, the cameras were installed inside the vehicles. Analyzing the data, you notice that most videos have reflections, image quality that is too low, and an extremely limited field of view. If you think AI can work miracles by detecting even the minutest details, know that you will likely only waste your budget. Rather than immediately starting an automation project, you could invest part of that budget to relocate the cameras outside the vehicles or replace them with higher-quality cameras. With quality data, you will certainly achieve better results.
Once you understand this first fundamental step, the next step is to set clear and measurable goals. What performance must the system achieve to have a tangible impact on your business? Which processes does it affect? Without KPIs, you will navigate in the dark. Whether it’s improving customer satisfaction, reducing operational costs, or increasing employee productivity, ensure these KPIs reflect the impact you expect from each AI integration. If you don’t know where to start, you could develop a benchmark of AI solutions already on the market or open-source. Alternatively, you can take human performance in the activity you want to automate as the value the system should aim for.
Start with a Pilot Project
The adoption of artificial intelligence is a journey, not a destination. You don’t need to reach the goal in one go. Every AI project is unique and should always be characterized by an experimental process. Our suggestion is to always start with the development of a Proof of Concept (PoC), a pilot project that allows teams to work toward clear goals in a limited context. The goal of a PoC is not to integrate AI from day zero, but rather to understand its limitations. It is an iterative process, where you experiment, identify areas for improvement, and test the real potential of an AI system in an operational context. It may seem like a useless waste of time, but AI is not a magic box and will not work as you expect from the moment you apply it. A pilot project helps you understand how to best leverage it and paves the way for more conscious adoption in business processes.
Returning to the example of the insurance company, instead of aiming to extract all the information you need from the start, you could focus your energy on developing an AI system capable of reading the license plates of damaged vehicles. Start with European vehicles and measure the system’s performance. After an initial testing phase, you might discover, for example, that the system struggles to distinguish certain characters. For instance, the system might frequently confuse similar characters like “D” and “B”, “0” and “O”, or “C” and “G”. This is a common problem in this type of task. Focus on this aspect and improve the system’s performance. Once the KPIs are reached, you can easily make the system capable of recognizing license plates from non-EU countries. The PoC serves to structure the system in the right configuration and identify a series of actions to make it robust. Only once you have verified the effectiveness of artificial intelligence can you integrate it into business processes.
Pair the System with the Current Process
Before automating the current process with AI, ensure that it maintains good performance over time. Like a new employee, AI will need a training period where you can monitor the system’s performance in a real-world scenario for a sufficiently representative period. For example, if you are a highway company and your goal is to analyze traffic, you should ensure that the new AI system is paired for a period long enough to cover “normal” traffic periods and busier periods like holidays. This phase ensures that the system remains reliable in every operational scenario and corrects any unexpected behaviors.
Integrate It into the Business and Scale the Solution
Once the system has been sufficiently tested, you can finally integrate it into business processes. In this final step, it is crucial to remember that AI is a tool, and like any other tool integrated into your company, it must have an impact on the business. In this phase, it is essential to constantly monitor the KPIs you defined in step 2. Conduct periodic review sessions to evaluate the use and effectiveness of the tools. It’s not just about monitoring metrics; the goal is to understand the story behind the numbers. How are AI tools reshaping workflows, decision-making, and customer interactions? Are they meeting, exceeding, or falling short of expectations? This continuous evaluation cycle will help you identify success patterns, areas for improvement, and opportunities for further AI exploration.
If you’re still reading these lines, you will have understood that integrating AI into your business is an iterative process, where at each step you can measure the impact it is having on your company. It is certainly a complex process that can be intimidating, but it leads to concrete and tangible results. We at Synapsi know this well, and if you want, we can guide you through this process step by step.