Inside Adecco’s Push to Rebuild Recruiting Around AI Agents

TLDR

• Adecco says AI agents have cut the time to submit its first candidates to clients by about 40%, in part by screening applicants around the clock, says Pierre Matuchet, group senior vice president of IT and digital innovation.
• AI now handles tasks including pre-screening, talent pooling and worker redeployment, while final hiring interviews remain with human recruiters.
• Adecco learned that automating one workflow can create bottlenecks elsewhere – and that AI cannot fix a poorly designed business process.

Global staffing giant Adecco is using AI agents to reshape how it finds and screens workers, moving some of the most time-consuming parts of recruiting from human employees to bots that can work around the clock.

The company has automated tasks including initial candidate screening, building pools of potential workers and contacting people as their assignments end to help place them in new jobs. The goal is not simply to make individual recruiters faster, but to redesign parts of the recruiting process around work that AI can perform automatically.

“Using AI is not an IT project. It’s first of all a business transformation project,” Pierre Matuchet, group senior vice president of IT and digital innovation at Adecco, said in an interview with The AI Innovator.

“We want to harmonize all our processes, the way we are working, in order to be as industrialized as possible for all these kinds of tasks in all the countries,” he added. “The agents are supporting this business strategy, and this business strategy is also to put the human touch, which is the main criterion of Adecco worldwide, at the center.”

One of the biggest changes is happening after recruiters go home.

Adecco branches typically operate during normal business hours, but its AI agents operate 24 hours a day, every day of the week. Many candidate interactions now take place outside regular working hours.

“You are an associate for Adecco. You apply to a job at 11 o’clock in the evening. You will have your pre-screening right now,” Matuchet said. This automation has helped Adecco reduce by about 40% the time it takes to submit its first group of candidates to a client, he said.

Adecco signed a multiyear agreement through 2027 for unlimited global access to Salesforce’s agent platform, and Matuchet said the negotiations took time because both companies had to be confident the economics would work. “It’s not easy for Salesforce to sign an unlimited deal, and it’s not also easy for us to sign an unlimited deal,” he said.

Adecco’s agreement comes as Salesforce experiments with new ways of charging for AI agents. The software company began introducing unlimited Agentforce pricing in May 2025, alongside consumption-based options that charge according to usage.

SaaS vendors that historically charged companies according to the number of employees using their software are increasingly experimenting with usage and outcome-based pricing as autonomous agents perform work without requiring additional human ‘seats,’ according to a Deloitte report.

Screening applicants while the recruiters sleep

Under Adecco’s traditional process, a worker might find a job online, submit an application and wait until the following day or longer for a recruiter to call and begin screening the candidate.

That delay can be particularly problematic because much of Adecco’s business involves placing blue-collar workers who try to work as much as possible, Matuchet said. For instance, someone working when a recruiter calls at 10 a.m. may not be able to answer, creating a cycle of missed calls and callbacks.

AI changes that sequence.

Matuchet described a recruiter at an Adecco branch near Paris who now starts the agents before leaving work at 6 p.m. During the night, the agents can search for candidates, respond to them and conduct initial qualification based on criteria associated with a client’s job order.

By the next morning, rather than beginning with perhaps 200 people to contact, the recruiter could have a group of 20 or 40 candidates already ranked according to predetermined criteria.

Adecco began its pilot in the United Kingdom in May 2025. The initial agent took roughly six to eight weeks to build, according to Matuchet. The company subsequently expanded its deployment and is now rolling agents out across major markets while developing additional ones.

The company’s recruitment processes operate on the Salesforce platform, giving the agents access to an existing tech and data infrastructure rather than requiring Adecco to connect each agent to candidate and job databases.

Deciding what AI should – and shouldn’t – do

Before building agents, Adecco mapped its recruitment process and divided the work into tasks that could be automated and tasks it wanted humans to continue performing.

The company internally calls that map a “snake” of the process, Matuchet said. It then looked for areas where automation could deliver relatively high returns at lower cost.

Besides pre-screening, another use case is talent pooling.

If Adecco knows employers in a particular market are likely to need forklift drivers before Christmas, for example, agents can begin contacting workers months earlier. They can ask what equipment the workers can operate, whether they remain available and whether they still live in the area. When an employer eventually places an order, Adecco already has a pool of prospective candidates.

Another agent focuses on redeployment. When a temporary assignment ends, it can contact the worker to ask about the assignment, what the person learned and whether the worker acquired new hard or soft skills. Adecco can then use that information to identify another potential assignment.

The company is also piloting an associate support agent in the U.K. to answer routine questions from workers without requiring a human employee to pick up the phone. Agents are being explored for back-office and sales functions as well.

The final hiring interview, however, is handled by a person.

“The final interview is human,” Matuchet said, adding that Adecco must comply with the EU AI Act governing the use of AI in employment decisions.

Adecco has also developed what Matuchet called a “responsible AI by design” process to address potential bias. Its responsible AI team becomes involved at the beginning of agent development, rather than reviewing systems only after they have been built.

When better automation created a worse process

Adecco’s rollout also produced an unexpected lesson about AI productivity.

During the U.K. pilot, the company deployed a pre-screening agent broadly. The system became so effective at processing candidates that hundreds passed the initial screening and moved to the next stage.

There was just one problem: Adecco didn’t have enough people available to interview them.

“We learned that an agent worked much more than a human,” Matuchet said.

Instead of reducing work, the agent initially risked creating more of it. Adecco had optimized one section of its recruitment pipeline without increasing the capacity of the human-operated section immediately downstream.

“By over-optimizing this part of the process, we are de-optimizing the process,” Matuchet said.

Adecco responded by putting controls around the system so agents would not generate more candidates than the human portion of the recruiting operation could handle.

The experience points to a broader challenge for companies deploying AI agents. Measuring how much faster an isolated task becomes may say little about whether the entire business process becomes more efficient. Automation can simply move a bottleneck somewhere else.

Adecco learned a second lesson when another agent failed.

The company had tried to automate a sales process that was already poorly designed. Instead of repairing it, AI reproduced the underlying problems.

“We had one agent who failed miserably because we have identified a messy process,” Matuchet said. “If it’s a mess for a human, AI will not fix it.”

He summed up the lesson more simply: “AI is not a magician.”

From AI pilots to AI at scale

Adecco is now trying to avoid another common enterprise AI problem: getting stuck in perpetual experimentation.

Instead of running separate proofs of concept in country after country, the company concentrates much of its experimentation in the U.K. Once an agent has matured and been tested, Adecco uses a standardized methodology to deploy it in other markets.

For Matuchet, that ability to scale – rather than simply demonstrating that an agent works – is becoming the critical test.

“It’s easy to incubate an agent and to say we will gain productivity,” he said. “The complexity is to be able to execute at scale.”

Adecco’s longer-term strategy is to continue developing agents that take over repetitive parts of recruiting while shifting recruiters toward work that depends more heavily on human relationships, including advising candidates about careers and helping them find appropriate jobs.

That makes the company’s AI experiment about more than whether software can screen candidates faster. Adecco is testing what happens when an entire recruiting operation is divided between machines that can work continuously and people who increasingly handle the parts of employment that the company believes should remain human.

As for the cost of AI, it is not an issue for Adecco for now since it has an unlimited contract with Salesforce to use Agentforce. But that may change down the line.

“Do we need the complexity of the model we are using today? I’m not sure,” he said. “If one day there is a concern of pricing with the model, we will go to the cheaper model.”