Companies are increasingly adopting artificial intelligence, but many limit themselves to isolated experiments: automating work in one department, changing tools in another, and reviewing staffing and employee training in a third. The result is not a unified transformation effort but a collection of fragmented initiatives that duplicate one another and often undermine the overall outcome.

Analysts at Bain & Company propose a different approach: start not by looking for tasks for AI, but by completely rethinking how work gets done within the company. In their view, the greatest returns go to organizations that simultaneously redesign core business processes, divide responsibilities between people and AI, and change team structures.

Do not look for uses for AI—redefine how work should be done

In many companies, AI adoption starts with a question: which task can be automated? Yet this approach often does little more than introduce new tools into old processes.

Finance departments review procurement, IT teams focus on automation, operations departments cut costs, and HR handles retraining and workforce optimization. Each department addresses its own task, but the changes are not always coordinated.

This leads to duplicated functions, misaligned actions, and employee resistance. Moreover, a constant succession of localized changes can destabilize the entire organization.

Bain suggests starting with different questions: which activities can be eliminated, what should be simplified, which processes need to change to serve customers better, and where can AI deliver substantially greater results?

This approach pushes companies to move away from numerous small pilot projects in favor of a few critical end-to-end processes. Each should have specific goals and designated leaders accountable for the results.

There is no need to wait until all the company's data is perfect or to develop every technology in-house. In some cases, it makes more sense to use off-the-shelf solutions and technologies developed by partners; in others, to build proprietary tools if they give the business a competitive advantage.

The key is not to try to speed up the existing system at any cost, but first to determine what that system should look like.

Why automating old processes can make problems worse

One of the main problems facing large organizations is accumulated workflow debt. This consists of unnecessary activities that build up over the years around meetings, approvals, handoffs between departments, exceptions to rules, and internal procedures.

As a result, even a simple task can require numerous intermediate steps and participants. Adding AI on top of such a system will not necessarily make it more efficient. On the contrary, automation can scale up existing complexity.

This becomes particularly acute when using autonomous AI agents. A person can make sense of vague instructions, reach an agreement with colleagues, or work around conflicting rules. To operate reliably, an agent needs clear requirements, a stable sequence of actions, and clearly defined authority.

Before handing processes over to AI, companies therefore need to establish which steps are truly necessary, who makes decisions, and where one participant's responsibility ends and another's begins.

Bain recommends designing core processes virtually from scratch. Rather than simply reducing the time needed for individual activities, companies should first establish overarching goals—for example, improving the customer experience, raising service quality, reducing costs, and completing tasks faster.

The appropriate performer is then identified for each stage: a person, an AI-assisted employee, or an autonomous agent.

People and AI must work within a single system

Redesigning processes inevitably requires changes to the workforce structure. In the future, teams will bring together employees using AI as an intelligent assistant, autonomous agents, and, in some fields, robots.

This means the conventional distinction between people and automated systems is no longer sufficient. Companies will have to rethink workforce size and skills, expand training programs, and change their approaches to managing transformation.

Bain identifies several conditions without which large-scale redesign is unlikely to succeed.

First, changes should be designed by cross-functional teams bringing together technology specialists, business process owners, finance professionals, and HR. Leaders rarely combine a deep understanding of operations, technological expertise, and workforce planning experience.

Second, companies need to build employee trust early on. People should understand how their responsibilities will change and what opportunities the new tools will create. Retraining, redeployment, and transparent communication help employees see AI as a way to enhance their capabilities, rather than solely as a threat to jobs.

Finally, new processes must be embedded in corporate systems, task allocation rules, and performance evaluation criteria. Otherwise, even a successful experiment risks remaining a standalone project without changing the company's day-to-day work.

How a British bank cut a complex process to a single day

Bain cites a major UK banking group as an example. Previously, testing or launching a new customer engagement idea took between 60 and 100 days. The process involved more than ten handoffs between departments.

Instead of trying to make each stage slightly faster, management set a more ambitious goal: reduce the entire cycle to a single day.

Over four months, the teams responsible for data and digital channels developed a new workflow. It combined the bank's technology infrastructure with agentic AI capable of complex reasoning and handling multistep tasks.

Technology alone, however, proved insufficient. The bank also had to change how work was organized. Large specialist teams of up to 40 people were replaced by small groups of three or four professionals who could independently handle a broad range of tasks.

This required employees with more versatile skills. Some staff were retrained, while certain specialists had to be hired externally. As Bain notes, employees supported the transformation in part because the bank emphasized training and empowering its workforce rather than reducing headcount.

As a result, the bank's marketing function is gradually becoming a continuous learning system: AI helps identify patterns in customer behavior, while employees analyze the resulting data and determine which approaches truly resonate with their audience.

AI's greatest impact does not depend on the number of deployments

The bank's experience illustrates Bain's central argument: the effectiveness of artificial intelligence is determined not by the number of projects launched or activities automated, but by how fundamentally a company is willing to change the way it organizes work.

If AI is simply embedded in existing processes, it may speed up inefficient procedures, increase the number of automated approvals, and introduce new points of complexity. Achieving more substantial results requires simultaneously rethinking the sequence of activities, the allocation of responsibility, and team structures.

This requires management to take a broader approach than buying new software or launching another pilot project. Specific business objectives must be at the heart of the transformation, with technologies and workforce changes selected to support them.

That is why the next phase of AI adoption in companies may depend less on the emergence of new models than on businesses' willingness to abandon familiar ways of organizing work.