A common question posed over the last couple of years is around whether traditional Business Process Outsourcing (BPO) firms will fall victim to the advance of AI? Or will the alternative be true – AI will represent a lucrative new revenue opportunity for such providers to re-invent themselves?
BPO refers to organizations contracting with third party providers to take usually routine tasks, processes, and business operations off of their hands, freeing up their internal teams to focus on more lucrative core corporate objectives than, for example, processing payroll.
But why turn to conventional BPO providers if agentic AI tech can do the same thing for you? That’s the threat posed by AI here, as Philippe Salle, CEO of Netherlands-based ATOS, warns:
BPO will suffer a lot with agentic.
AI is not the “easy catch” that many of his peers seem to believe, he suggests:
We see a lot of things, but it’s not agentification of a process or agentification of operations. It takes some time. I have already said that it’s not a problem of technology because building an agent is not that difficult. It’s mainly about data and process first and that’s why we simplify. We have more and more consulting projects on this.
There is, of course, a lot of questions on AI…that it is difficult to escape that kind of conversation. But it’s not an easy catch, so it doesn’t translate immediately, let’s say, to AI projects.
None of this is to say that ATOS does not see new AI-related revenue coming in to its coffers, he adds:
What is important to understand is that AI is, in fact, embedded in the different projects that we do. So we have projects fully on AI data, for example, on data lakes, or making sure that the company has the right data momentum. And we have also some agentification.
On the other hand, at Indian services firm Genpact, CEO Balkrishan Kalra sees the AI shift as a moment to re-think and re-invent the firm:
When such a rare structural shift is presented and a company can sharpen its differentiation and has the courage and discipline to act, the resulting advantage compounds in ways that are difficult to replicate. This is our moment. We are intentionally disrupting ourselves…Genpact is not the company you knew.
He adds:
We are changing what the whole business is and moving to a new category. This is where it all comes together. Clients do not come to us to buy core or Advanced Tech. They come to us with a vision for the future and a reality of where they are today. we are enabling the journey to agentic-led autonomy that enterprises can trust.
Demand driver
At Capgemini, CEO Aiman Ezzat is enthused in his comments on the subject of AI’s potential:
AI has become the leading driver of new demand. To capture this opportunity, we are building AI enterprise hubs around each of our core partners, bringing together our capabilities, assets and expertise to deliver enterprise scale outcome-driven AI transformation. We continue to enrich our portfolio of AI offerings to help our clients accelerate adoption and realize business value faster.
Capgemini identifies five AI value pools that are re-shaping business transformation – enterprise tech modernization around helping clients address years of accumulated technology debt and prepare the environment for AI scale; the re-set of the tech stack, as application data platform and infrastructure are re-designed for AI native work; the agentic control plane, which provides the governance orchestration, security and observability required to deploy agentic AI safely at scale; agentic products and services, enabling new Customer Experience; and the identification of enterprise processes where AI agents augment and increasingly orchestrate end-to-end business offering.
As Ezzat positions it, Capgemini’s conviction is that combining AI industry and domain knowledge as well as data operation expertise is becoming a critical lever for enterprise transformation:
Every organization today wants to become agentic, but before they can become agentic, they must become AI-ready and most are not. Accumulated technical debt has left many enterprises, with fragmented data, legacy systems and complex integration layers, limiting their ability to deploy AI at scale and realize its full value.
What is changing is that agentic AI is transforming the economics of modernization itself. Modernization program, once seem too costly or too complex, are becoming economically viable and deliver faster, and initiatives that were often viewed as optional have become strategic imperatives.
Meanwhile at Indian giant Infosys, AI services currently account for 8.2% of total revenues, but CEO Salil Parekh expects to see this increase rapidly:
I think if we are able to execute on this AI transformation, as we have done in the last few quarters, it’s not that difficult to see that in the coming few quarters it will start to become more and more larger part of our overall revenue and that will drive the growth of the overall company.
Parekh sees precedents for how AI might re-shape the sector from earlier evolutions:
Not that it’s the same thing, but there’s some lessons maybe on digital [transformation]. At one stage, we were at 20% and then over a few years, we then went to 60% of our revenue becoming digital. If that sort of path becomes followed we can see a big transformation and long-term support for the view that what we are doing remains relevant in terms of services for our clients.
Relevancy
Traditional services firms do still have long-term relevancy in the face of the agentic advance, he insists:
What I think works for us is we have over 300,000 employees. We have deep knowledge and context of select clients that we work with, and that becomes the way to really ensure that AI gets leveraged into that environment, which is typically quite complex.
Meanwhile over at UK services provider Capita, CEO Adolfo Hernandez, totally fancies his firm’s chances. Capita does view AI as a transformative technology wave that will re-shape the services industry in its wake, Hernandez says, but adds:
There was nothing wrong with what we were doing; it’s how we were doing it that could be worked on.
He cites data from Boston Consulting Group that suggests that 70% of the value derived from AI is derived through people who understand process:
[People who] are able to nuance the delivery of a particular service, who are not only good to deal with the happy path that can be automated, but they have the skill and the experience to deal with the unhappy path of service provision that needs that human intervention.
Technology is a means to an end, he argues, and a firm like Capita can take an agnostic stance on the AI stack across enterprises:
We are not tied to any particular architecture. We’re not tied to any particular LLM (Large Language Model). We’re not tied to any particular data lake or analytics structure. But we’re sitting on the top with the depth of the process, with the people, with the understanding that is required to deploy technology there and get that ultimate outcome that a regulated industry, a government department, a local administration, our forces really need to deploy.
That’s the secret sauce, he insists:
You cannot buy your way through trillions into long-term expertise and deep understanding of a process. That is the one thing you just cannot do.
Tech is easy, people and process are hard
The tech bit is in many ways the easy part, he suggests:
Most of our employees know how to build an agent, but is the agent relevant? Is the agent going to effect and leverage something in the business process? It’s going to be grounded? Is it going to be safe? Is it going to be governed? So it’s not a silver bullet; it’s like a collection of lead bullets! It needs to be safe, it needs to serve a purpose.
All this impacts on adoption prospects, he goes on:
You need to have the operating team willing to use it. They need to see an immediate benefit, and the benefit that we’re seeing is if it makes their life easier, that is, in operations, for us the real measurement. If it adds value to you because you’re doing something, because it allows you to populate your responses faster, because it creates a better quality, because it checks what you’re going to do before, or because it gives you faster relapse times, most people go, ‘Yes, give it to me!’.
Hernandez pitches Capita as an early mover in all this, but one that is, he says, “de-risked” as adoption realities kick in:
What really started to sink in towards the later half of last year was there’s going to be a lot of orchestration required – orchestration of agents, orchestration of multiple agents. And then ultimately, it’s a realization whereby the whole narrative in the industry has changed back from ‘Humans are irrelevant’, to now, ‘Humans are absolutely critical to work in this workflow’. That’s been the journey we’ve been on and our journey maps that.
The question is no longer whether you can build an agent; the question is no longer whether you can build 10, 20, 30, 100 agents; the question is whether you can manage the agents? The question is whether you know who built the agent for what purpose, whether you know for every single agent, what data are they using?
How do you make sure that are there not three different people, three different departments building the same agent, whether the return on investment of the agent is from this one one, whether the agent should be retired and has it been retired properly? It’s not just the technology, it’s knowing what technology you need to build, you need to know about what operational processes. That skill of understanding, orchestrating, managing the life cycle of the agents will become the #1 priority in deployment of agentic AI in regulated industries, and we’re way ahead of the pack in terms of having built that capability internally for us as customer zero and now very happy to be taking it to market.
Capita wants to think in terms of deployment orchestrators, not the Forward Deployed Engineers talked about by the frontier model firms and other services outfits:
Their thesis is you go in, you understand, you build and you move out as the Forward Deployed Engineer. Our ambition is to leverage what we’ve been doing for 40 years, which is observe a business process, deploy that new business process, orchestrate the business process, and then operate it.
That’s what we’ve been doing for 40 years in the analog world, so we’ve taken that very same logic into the agentic world. We will look at the business process, we will re-imagine the business process, and then we will orchestrate it and we’ll manage it, but we will do that with agents.
So instead of building an agent, deploying, and running away, our proposition to our customers is, ‘We will work with you. We will look at your business processes. We will optimize it. We will build an agent. We will stay. We will run it with you. We will help you train people. We will help you embed it into the operating culture. We will make sure you get the usability. And because these things are never static, if it needs improvement, we will drive the improvement for you’.
There is an alternative approach, he admits:
You’re going to get a tons of agents, you drop them in, no training, no development, no change in the process, they won’t be used, and you will be wasting your money.
My take
There’s certainly no shortage of ambition on the part of the BPO vendor community to milk this lucrative new revenue stream for all it’s worth. The question will surely be whether they are able to adapt their own traditional ‘big ticket’ practices to accommodate the pricing competition from agentic alternatives. While tokenomics is a particularly acute problem today, it’s not unreasonable to expect this to pass in time. The current nascent stage of adoption needs to be treated as a period of adaptation by BPO providers to position themselves as the safe pair of hands, the image so many of them will undoubtedly shape their pitches to present to prospects.