Expert opinion from the EvolutecC engineering team
Artificial intelligence is no longer a technology companies are merely considering. Today the challenge isn't trying it out, but making it work sustainably within day-to-day operations.
A McKinsey study published in June 2026 found that almost 90% of organizations are already experimenting with AI, yet only 7% have managed to scale it across the enterprise. The gap between testing AI and turning it into a real business capability remains enormous.
It's a situation we see more and more often: the company has already tried AI, but the results aren't what it expected. Maybe it has a chatbot up and running, an assistant connected to internal documentation, automations built on generative models, or even agent pilots. The problem arises when taking that solution into production: connecting it to existing systems, feeding it reliable information, maintaining it over time, and proving that it cuts costs, reduces errors, saves work hours, or improves a business metric.
At that point the conversation changes. The question is no longer “which AI tool should we implement?” but “does our company have the technological and operational foundation needed to make the most of it?”
Adopting AI is getting easier; scaling it is still the challenge
The barriers to experimenting with AI have dropped dramatically. Today you can test assistants, generative models, automations, and agents without building the technology from scratch. But a working proof of concept and an enterprise solution are two different things.
The same McKinsey study argues that companies that move forward combine technology with operational discipline, internal capabilities, and changes in how processes are executed. A second analysis by the firm, published in July 2026, reinforces the idea: greater individual use of AI tools does not, on its own, create enterprise value. The most advanced organizations redesign workflows, roles, and operating models around what AI makes possible, instead of adding one more tool to the existing process.
The difference is fundamental. The AI model can be part of the solution, but it is rarely the whole solution. To operate, it needs data, integrations, infrastructure, security, defined processes, governance, and metrics. When any of those layers fails, AI inherits the problem.
What may really be holding back your AI project
Seen from the systems side, many of the difficulties in an AI project begin long before the model is chosen. AI simply exposes them.
1. Data that isn't ready for AI yet
Imagine a company where the CRM identifies a customer one way, the ERP uses a different reference, the ecommerce platform stores different information, and several departments keep parallel Excel files to fill in what they can't find in the main systems. Then someone suggests: “Let's connect an AI to analyze our customers and generate recommendations.”
The model may be excellent, but a much more basic question comes first: which of all those sources has the correct customer information?
This problem becomes more prominent as organizations try to move from pilots to enterprise solutions. In June 2026, McKinsey published an analysis dedicated to data readiness for AI, whose central conclusion is that, as pilots scale, data is becoming a critical constraint. To use AI reliably, organizations need to connect structured and unstructured information on a governed, reusable foundation that can be considered a trusted source.
EY reached a similar conclusion in a July 2026 guide on data architectures for AI: fragmentation, inconsistent governance, and difficulty managing structured and unstructured information are holding back enterprise AI.
That's why it isn't enough to ask whether we have data. We need to ask whether it's reliable, up to date, and connected, who governs it, and whether we can use it securely. AI doesn't magically turn messy data into reliable information.
2. Fragmented systems that were never fully connected
This is another common scenario. An organization may have an ERP, CRM, ecommerce, accounting software, a customer service platform, marketing tools, internal applications, and custom systems. Each works well on its own; the problem starts when they need to work together.
One employee exports data from the ERP and organizes it by hand in Excel; someone else corrects a few fields; the file moves to another department, and in the end someone uploads the data into yet another system. Then the proposal appears: “Let's automate the process with AI.” But that problem probably doesn't need AI yet. It needs integration.
IBM put it bluntly in March 2026 when analyzing architectural readiness for AI: fragile integrations, failed syncs, manual processes, and systems that don't communicate well can become barriers to scaling. Putting an intelligent layer on top of fragmented systems doesn't eliminate the fragmentation; sometimes it just builds a more modern interface on top of the same structural problem. Before adding AI, you need to understand how information currently flows between the company's systems.
3. Processes that still aren't well defined
Another common mistake is trying to automate a process before fully understanding it. A company comes in asking “can we automate this with AI?”, but when the process is reviewed, questions appear that have no shared answer:
-
Who starts it, and what information do they need?
-
Who makes the decision, and which systems are involved?
-
What rules exist, and what exceptions must be considered?
-
What happens when information is missing?
-
Who validates the result, and when does the process end?
-
Which metric will tell us it improved?
If the answers depend on who performs the task, the process isn't structured. And automating an ambiguous process doesn't improve it: it may just make its errors happen faster.
McKinsey notes in its July 2026 analysis that moving from adoption to impact requires redesigning workflows around AI's capabilities, not inserting tools into existing dynamics. That's why order matters: understand, structure, integrate, automate, and only then bring in AI where it creates value.
4. Infrastructure that wasn't designed for what we now want to build
Some companies try to build advanced AI capabilities on top of architectures that have been accumulating changes for years: legacy services, monolithic applications, databases that are hard to integrate, no APIs, outdated dependencies, insufficient processing capacity, performance issues, or environments without a clear separation between development, testing, and production.
This doesn't mean you have to rebuild the entire technology ecosystem before using AI. It means we need to know how ready the current architecture is to support what we want to implement.
Deloitte's State of AI in the Enterprise 2026 report reveals a telling gap: although 42% of companies consider their strategy highly prepared for AI adoption, they feel less prepared in operational dimensions such as infrastructure, data, risk, and talent. We can be clear about what we want to do with AI and, at the same time, not yet have the technical capacity to do it at scale. Strategy can move faster than infrastructure.
5. Projects without a clear business outcome
“Implementing AI” shouldn't be a goal. A goal looks more like this: reducing the processing time of a request, cutting manual errors, eliminating repetitive tasks, improving response times, automating a certain percentage of inquiries, increasing operational capacity without growing the team at the same rate, improving information classification or retrieval, or detecting anomalies before they become problems.
The difference seems small, but it completely changes the architecture of the solution. An AI can answer correctly and still be a bad investment. If we don't know which metric should improve, it will be very hard to prove later that the technology created value.
In fact, according to McKinsey's July 2026 study, most of the organizations analyzed still don't report significant enterprise value from AI in performance, cost reduction, employee experience, or customer outcomes. So the question isn't just “does it work?”, but also “what changed in the business because it works?”
6. Lack of capabilities to maintain what gets built
Launch is not the end. An enterprise AI solution needs to evolve: models, APIs, and costs change; information sources evolve; new security risks emerge; users discover new use cases; integrations require maintenance, and business processes transform too.
That's why it pays to define from the start who will be responsible for operating, monitoring, and evolving the solution. Deloitte identifies the skills gap as one of the main barriers to integrating AI into organizations. Building isn't enough: you have to be able to operate, measure, govern, maintain, and evolve.
It's not just a small-business problem
A bigger budget doesn't eliminate these difficulties. A large company probably has more infrastructure and specialized talent, but also more legacy systems, databases, vendors, departments, integrations, processes, regulatory constraints, and technology dependencies.
Deloitte found in 2026 that adoption keeps advancing, but only some organizations are truly rethinking the way they operate: 34% say they are reimagining the business, while gaps persist in infrastructure, data, risk, and talent. Meanwhile, McKinsey found in July 2026 that only 11% of surveyed leaders place their organization at the most advanced stage of transformation —which it calls reinvention—, where AI is already starting to reshape roles, workflows, and operating models.
So it isn't a matter of size, but of technological and operational maturity.
AI amplifies the operation you already have
This is perhaps the most important idea of the whole analysis: artificial intelligence doesn't fix a disorganized operation; it amplifies it. IBM summed it up in May 2026 when discussing how to build AI-ready data foundations: without contextualized, reliable information, AI can end up amplifying the fragmentation that already exists in organizations.
If a company has reliable data, connected systems, clear processes, defined owners, a ready architecture, governance, and measurable metrics, AI can multiply capabilities that already work. But if it has contradictory data, manual processes, fragmented and unintegrated systems, duplicated information, and blurred responsibilities, AI will have to operate within those same limitations.
Sometimes the bottleneck isn't artificial intelligence, but what lies beneath it.
What does being ready for AI really mean?
It doesn't mean having the most expensive ERP on the market, migrating absolutely everything to the cloud, replacing every application, or rebuilding the entire architecture. It means understanding five dimensions well enough:
-
Data. Where is the information? Is it reliable? Are there duplicates? Who manages it? What is the official source? Can we use it securely?
-
Systems and integrations. Which platforms take part in the process, and how do they communicate? Are there APIs? Where do manual handoffs persist? Is information duplicated across systems?
-
Processes. Is the flow documented? Who is involved? What rules and exceptions exist? Which parts are worth automating?
-
Architecture and infrastructure. Does the current architecture allow new capabilities to be integrated? Can it scale? Does it have security or performance issues? Is there enough observability to know what happens when something fails?
-
Objectives. Which metric do we want to change, and by how much? How will we measure it? When will we consider the initiative a success?
Only after answering these questions does it make sense to decide whether the solution requires automation, integration, modernization, data work, custom development, artificial intelligence, or a combination of several.
Not every problem needs AI
It may sound strange coming from a company that builds AI solutions, but that's exactly the difference between selling technology and solving a problem.
If a task can be solved by properly connecting two APIs, you probably don't need a generative model. If the problem is duplicated information, the data has to be cleaned up first. If the process isn't defined, it has to be understood first. If the infrastructure is unstable, it has to be stabilized first. And when AI is the right tool, we can bring it in on a foundation capable of making the most of it.
The question shouldn't be “where can we put AI?”, but “what problem do we need to solve, and what technology architecture makes sense to solve it?”
That's why at EvolutecC we don't necessarily start by building
When a company comes to us with an AI, automation, or integration need, jumping straight into development can be the most expensive decision. First we need to understand what's happening today, which systems are involved, how information flows, how reliable the data is, what dependencies exist, where the bottlenecks are, what technology risks there are, and what outcome the business expects.
That assessment lets us separate three things: what's urgent, what's necessary, and what really creates value. Sometimes the result will be implementing an AI solution; other times, integrating systems, modernizing the architecture, organizing data, automating processes, or building a custom solution. In the most complex projects, it will probably mean combining several of these actions in a phased roadmap.
Technology should be the result of the assessment, not its starting point.
Before switching tools, review the foundation
If your company has already implemented AI or automations and the results haven't been what you expected, switching platforms right away may not solve anything. First, it's worth answering one question: were our data, systems, processes, and infrastructure really ready to take advantage of that technology?
A technical audit evaluates architecture, infrastructure, security, performance, integrations, and scalability to identify risks, dependencies, and opportunities for improvement. If the problem also involves the quality, structure, or governance of information, the assessment can be complemented with a specific data analysis.
The result should give you clarity on three points: what to fix now, what to prepare before scaling, and where it really makes sense to invest in artificial intelligence. Because the ultimate goal isn't to add more technology, but to build the right technology to solve a real problem.
Before your next AI investment, find out whether the problem lies in the tool or in the foundation it needs to run on.
Schedule a technical audit with EvolutecC. Assess the state of your architecture, data, integrations, and infrastructure, and build a roadmap to understand what your company really needs before scaling automation and artificial intelligence.