From business goals and processes to data, infrastructure and oversight: how to connect individual AI solutions into a whole that delivers measurable business results.
Introducing artificial intelligence into a large company is much more than a matter of choosing the best AI tool. The key challenge is how to connect business goals (the impact of the rollout), processes, data, AI solutions, infrastructure, and the management and oversight of the entire environment into a working whole. Do you know how many different AI tools your employees are using today? How much do they cost you, what data do they access, and what business impact do they actually create?
These are questions every chief executive of a large company should be asking today. Not because they need to know the differences between individual AI models, but because a fragmented approach to AI can quickly create a new layer of complexity, costs and risks.
“Large companies already have hundreds of applications today. Add a multitude of AI tools, different models and ways of using them, and you get a new problem that someone will have to manage,” warns Ivo Švrljuga of Smartis.
What does a systematic AI rollout mean?
It means that a company does not introduce artificial intelligence as a collection of disconnected tools, but as part of a business system that connects business goals, processes, data, AI solutions, infrastructure and the management of the entire environment (from defining the rules to overseeing their execution).
For company leadership, this means at least six questions.
1. What do we actually want to achieve with AI?
The biggest mistake can happen right at the start: a company sees what AI can do and starts looking for places to use it. But technical feasibility is not yet a business case for investment.
“If you have 50 people using a solution, but the company doesn’t need a single person less and doesn’t generate more revenue because of it, all you’ve got is an additional cost,” says Švrljuga.
The first question, therefore, is not what AI can do, but what we want to achieve with it: more revenue, lower costs, faster processes or better service. If we cannot define the impact in advance, we will struggle to prove it later.
“If your employees use AI, that is not yet proof that the company is performing better because of it. As a CEO, you need to know what you have improved with it, how much it costs you and whether you are in control of how it operates. If you don’t have these answers, you need a clear plan, not just new tools. In our free webinar, we will show how to connect individual projects into a whole and measure whether your investment in AI is paying off.”
IVO ŠVRLJUGA, Smartis
2. Is the process we want to automate well designed in the first place?
The next trap is automating a bad process. If we simply speed up an existing process with AI, we may very efficiently automate something that should not exist at all.
“A process that was designed for a human is not necessarily suitable for automation. If you want a truly automated process, you have to take another look at it and redesign it if necessary,” warns Švrljuga.
AI therefore does not just change the speed of work. It also changes the sequence of steps, responsibilities and the organisation of the process.
3. Is our data good enough for AI?
A person can often correct bad data using experience. An automated system does not have that context.
“You can build a great automation, but if your CRM doesn’t have the correct phone number, email address or other key piece of data, the process fails,” says Švrljuga.
AI therefore quickly exposes data quality problems. Data needs to be cleansed, structured and connected, and the same applies to the company’s knowledge: the documentation, contracts, instructions and other sources on which AI is to operate need to be prepared, classified and labelled with their security level, made accessible to digital tools, and governed by access rules based on user permissions.
4. How will we connect different AI solutions into a whole?
The real value of AI lies not in a standalone chatbot, but in its connection with existing systems, data and business processes.
If sales uses one AI solution, finance another, procurement a third and customer service a fourth, the company has not necessarily become four times more efficient. It may mainly have gained four new technology silos.
“Customers don’t want to piece together different platforms, vendors and technologies themselves. They expect a complete solution,” says Švrljuga.
This is exactly where an AI systems integrator comes in: it has to connect different models, applications, data and existing information systems into a whole that supports the business process.
According to Švrljuga, the next phase of AI adoption in large companies will therefore increasingly be an integration problem, rather than merely a question of choosing the best tool. Over its many years in business, Smartis has built strong relationships with many partners and is therefore uniquely positioned to offer each customer the most suitable set of tools, tailored to that customer’s specific business and technology needs. Its partners include the technology giant IBM, which has extensive experience in building and managing AI and a range of solutions built around the IBM WatsonxTM platform, and with which Smartis has many years of experience and successful cooperation.
5. Do we have a technology foundation that can handle all of this?
Models and AI agents must operate in an environment that is sufficiently powerful, secure and cost-manageable. The company must therefore decide what it will run in-house, what in the cloud and how everything will be connected.
“Even with external infrastructure, you need internal architecture, rules and oversight,” stresses Švrljuga.
Cost control is also important with AI. More models, more usage and more disconnected solutions can quickly mean higher costs, but not necessarily more business value.
6. How will we organise AI governance and oversight?
If an AI agent accesses documents, who decides which ones it may see? If it prepares a decision or performs an action in an information system, who sets its limits and who bears responsibility?
With serious automation, it is therefore not enough to monitor whether the system works. The company must know what data AI accesses, what it does with it and what actions it can perform.
Roles, access rights, policies and responsibilities must be defined, especially for personal data, contracts, payroll and trade secrets.
“Because of its probabilistic nature, AI is far more dynamic than classic deterministic systems. Models change, agents perform new activities, perform the same activities in new ways, and do everything quickly and often, so you have to monitor and adapt the environment,” says Švrljuga.
Governance is therefore not a one-off check when AI is introduced, but an ongoing task for the company.
Why is this becoming a CEO issue?
Because AI does not stay within a single department.
Sales, finance, procurement, production and support will each look for their own ways to use it, while IT will have to ensure security, stability and connectivity. This will lead to tensions between departments and increased business risk if these tensions are not properly managed.
Supervisory boards will not be interested in why internal activities were misaligned if serious incidents occur with high financial and business impact. Given the level of autonomy and automation that an AI environment enables, such events can happen quickly and their consequences can spread fast, like a summer wildfire in strong wind. Ultimate responsibility lies with the CEO, who is the ultimate link across the entire company and is first in line for managing systemic risks. AI is, and will be, a systemic risk if the company wants to harness its full capacity for business success.
Another important responsibility of the CEO is to ensure a consistent rollout of the AI environment across the entire company, as only this will deliver the maximum business impact. If everyone optimises only their own part, the company may end up with a multitude of successful AI projects, but not necessarily a successful AI transformation.
“AI is at its most powerful when it goes beyond a single department. You need someone who connects business goals, processes, technology and risks,” says Švrljuga.
This is exactly why the role of the systems integrator is changing as well. With AI, it is no longer enough to implement individual technologies. The business case (the expected business impact of the rollout), processes, data, AI solutions, infrastructure and governance must be connected, and this whole must be made to work in a coordinated and controlled way over the long term as well.
This is the approach Smartis is establishing as an AI systems integrator: from consulting and selecting the right use cases, choosing and integrating different solutions, preparing and managing the necessary infrastructure, and planning and ensuring an appropriate level of security, to the continuous management and adaptation of the AI environment. The real question for the CEO is therefore no longer just how quickly we are introducing AI, but above all: are we introducing it in a way that will genuinely make the company better, while still keeping it under our control?
