Adding AI to a bad process does not make a company “intelligent”

Artificial intelligence promises considerable productivity gains. It can accelerate the search for information, produce analyses, prepare documents, assist employees in their decisions and, with the arrival of AI agents, progressively carry out increasingly complex work sequences. Faced with these possibilities, many organizations naturally seek to determine where to integrate AI into their operations.

This approach nevertheless carries a trap. When a company starts by looking for the tasks it could hand over to artificial intelligence, it risks automating processes that first deserved to be questioned. Unnecessary steps can be accelerated, validations with no real value can become automatic, and data can circulate even faster through an organization whose processes remain complex or poorly structured.

A company that manages to speed up an inefficient process by 40% still has an inefficient process. It simply runs it 40% faster.

The arrival of artificial intelligence should therefore be seen as an opportunity to rethink the way work is organized. Before asking what AI can automate, an organization should return to a far more fundamental question: why do we do this work this way?

Automation can preserve processes that should have disappeared

Organizational processes are rarely built from a blank page. They evolve with the company. A step is added after a problem, an approval appears following an error, an Excel file is created to work around a system limitation, and an additional check is introduced because information is missing somewhere. Over time, these adaptations accumulate.

Each step generally has a historical explanation. That does not mean it still has an operational justification.

It is thus common to see employees enter information into a system, export it to a file, and pass it on to another person who checks it before entering it into another application. A third person may then produce a report from that same data. When AI arrives, the temptation is to automate the data entry, the verification, or the preparation of the report.

The technology can indeed make the whole thing much faster. Yet the most important question remains entirely open: why does the information have to travel through all these steps?

Automating without questioning the process can then produce a particularly modern form of waste: doing extremely efficiently something that perhaps should no longer be done at all.

Speed is not synonymous with productivity

Artificial intelligence also makes it necessary to distinguish between speed and productivity. An organization can produce more documents, analyze more data, respond to emails faster, and generate far more content without necessarily creating more value.

This distinction is particularly important with generative AI, since the marginal cost of producing certain information becomes extremely low. An employee can now produce in a few minutes a report that would have taken several hours. That represents a considerable potential gain if the report is useful to a decision or to an important activity. The gain becomes far less obvious if the report is produced solely because a legacy procedure requires its creation and almost nobody actually uses it.

AI can therefore considerably increase the volume of activity without proportionally improving the organization's performance. It may even help create more noise: more reports, more communications, more data, more recommendations, and more items that other people will then have to read, check, or manage. Productivity does not simply mean producing more. It means creating more value with the resources available.

AI brings the principles of operational excellence back to the forefront

Long before artificial intelligence, operational excellence was already seeking to understand how work actually flows. Why does this step exist? What value does it bring? What happens before and after? Where are the waiting times, the rework, the errors, the unnecessary handoffs, and the non-value-added activities? These questions become even more important with AI.

When a process is properly mapped, the objective should not be to immediately identify the places to put an agent or an artificial intelligence model. It is first necessary to determine whether each activity still deserves to exist in its current form.

Some steps can be eliminated. Others can be combined. Information entered several times can come directly from its source. A systematic approval can be replaced by exception management. A report produced periodically can become information available in real time. A task carried out by several departments can be simplified even before any additional technology is introduced.

It is only after this reflection that AI fully reveals its value. It then intervenes in a process already designed to achieve an objective rather than in a historical accumulation of working methods.

Digitizing a process is not necessarily transforming it

This confusion existed long before artificial intelligence. Many digital transformation projects essentially consisted of reproducing paper processes electronically.

A paper form becomes a digital form. A handwritten signature becomes an electronic approval. A physical file becomes a folder of documents. A spreadsheet becomes an application. The tool changes while the operational logic remains practically identical.

AI risks reproducing the same phenomenon on a far larger scale. A process designed fifteen years ago can now be partially handed over to agents without anyone asking why it still involves twelve steps.

Real transformation begins when the organization agrees to question the process itself. If we were designing this activity today with the technologies, the data, and the knowledge we now have, would we still build it this way?

The answer can sometimes be uncomfortable, because it calls into question procedures, responsibilities, and even certain organizational structures. It does, however, open the way to the most significant gains.

AI agents make a far deeper redesign possible

Traditional systems automate predictable, structured processes particularly well. AI agents can operate in far more variable environments. They can consult different sources of information, interpret a context, prepare a decision, use several tools, and continue a work sequence based on the result obtained at each step.

This capability makes it possible to rethink processes that were once fragmented across several people simply because conventional software could not handle certain transitions.

Take a request received from a customer. Today it may be read by an employee, categorized, forwarded to another department, analyzed, supplemented with information from the CRM, and then routed to a person authorized to make a decision. A properly integrated agent could eventually carry out several of these activities, automatically enrich the file, and pass directly to the human the situation that genuinely requires their expertise.

The gain then comes not only from the fact that the agent works faster. It comes above all from the disappearance of several handoffs, waiting times, and intermediate manipulations. This is precisely where AI meets operational excellence.

Automated bad data becomes a faster problem

The quality of the process also depends on the quality of the information that feeds it. A company whose data is fragmented, contradictory, poorly structured, or hard to find does not automatically fix that problem by adding artificial intelligence. It may even amplify it.

An agent able to act quickly on incorrect information can propagate an error far more effectively than a human. Duplicated data can lead to several interpretations. An obsolete procedure can be used as a reference. Important knowledge can be locked away in employees' personal folders or scattered across documents whose official version nobody really knows.

Preparing for AI therefore also requires preparing the organization itself. Knowledge must be sufficiently structured, sources of information must be identifiable, and responsibilities around data must be clear.

This is where knowledge governance becomes directly linked to operational excellence. An organization that is hard for its own employees to understand will be just as hard for its artificial intelligence systems.

Automation can also accelerate errors

The more autonomous a system becomes, the more speed of execution ceases to be purely an advantage. An employee can make an error in one transaction. An agent capable of processing several thousand can reproduce the same error on a far larger scale before it is detected.

This reality changes the design of controls. Companies must determine which decisions can be executed automatically, which require validation, which thresholds must trigger human intervention, and how to quickly interrupt a process when abnormal behavior appears.

Operational excellence in the age of AI must therefore incorporate resilience. The best process is not only the one that runs quickly when everything goes as planned. It is also the one that quickly detects anomalies, limits their consequences, and allows the organization to regain control. Autonomy must advance in step with the ability to supervise it.

Gains must be measured where value is created

The enthusiasm surrounding AI can also lead to measuring the wrong indicators. The number of users, queries, agents deployed, or tasks automated demonstrates technological activity. It does not necessarily demonstrate an improvement in performance.

An organization should instead observe what is happening in its operations. Has processing time gone down? Has the error rate improved? Are employees spending more time on high-value activities? Is the customer receiving better service? Has the cost of the process genuinely decreased? Has production or processing capacity increased? Are decisions being made faster and with better information?

These indicators make it possible to distinguish adoption from value creation.

A company may have deployed very few artificial intelligence tools and obtained considerable gains on a few critical processes. Another may have given all of its employees access to AI without seeing any significant improvement in its results. AI maturity should therefore not be assessed mainly by the amount of technology used, but by its measurable contribution to the organization's performance.

AI should sometimes lead to automating less

This conclusion may seem paradoxical. A thorough analysis of a process can reveal that certain tasks should not be automated. A customer relationship may hold more value when it remains human. An exceptional decision may require an understanding of context that is difficult to formalize. An infrequent activity may cost more to automate than to keep doing manually. A process may also be simple enough after review that conventional automation is more reliable and less expensive than an artificial intelligence solution. Using AI everywhere it is technically possible is therefore not a sign of maturity. Maturity consists of knowing where it brings enough value to justify its complexity, its costs, and its risks.

Transformation begins with understanding the actual work

To obtain that answer, executives must understand how work is actually carried out in their organization. Official procedures are not always enough. Between what is documented and what happens day to day, there can be many adjustments, parallel files, informal communications, and workarounds.

The employees who do the work then hold essential knowledge. They know where information is missing, which steps needlessly take time, which validations are genuinely useful, and which procedures exist mainly out of habit.

Involving them in the transformation makes it possible to discover opportunities that technology alone will never reveal. It also helps reduce resistance to change, since AI becomes a way to resolve irritants they know rather than a system imposed to change their work from the outside. Artificial intelligence thus becomes an organizational project before being a technology project.

The possibilities offered by artificial intelligence and AI agents are significant enough to profoundly transform the way companies operate. That power, however, makes a rigorous understanding of the processes into which these technologies will be integrated even more necessary.

Automating a useless task does not give it more value. Accelerating a bad decision does not improve it. Generating more information does not necessarily make an organization more intelligent. And adding agents to a complex process does not guarantee that it will become more effective.

The company must start by understanding the work it does, the value it seeks to create, and the obstacles that slow down that value creation. It can then eliminate what is unnecessary, simplify what is too complex, automate what is predictable, and use artificial intelligence where its capabilities bring a real advantage.

This approach directly links AI-driven transformation to operational excellence. At Quantum Beyond, we work with leadership teams, operational managers, and technology teams to analyze processes, structure knowledge, identify improvement opportunities, and integrate artificial intelligence where it can contribute concretely to performance. The objective remains to strengthen the capabilities of the organization and its teams, while maintaining governance suited to the new levels of technological autonomy.

The truly intelligent company will probably not be the one that has deployed the greatest number of AI agents. It will be the one that has understood its own operations well enough to know what it must eliminate, what it must simplify, what it must automate, and what it must continue to entrust to human intelligence.