Why McKinsey advises focusing on greenfield AI-first processes to achieve value

These days organizations are being pushed to adopt AI before they fully know how or where it will deliver the most value. As companies speed up AI adoption, many are realizing the shift isn’t happening evenly across service operations.  

This is because the work inside these teams varies widely, some tasks are highly structured and rules-based, while others rely heavily on human judgment. That mix means AI is advancing quickly in some areas and much more slowly in others. 

Martin Rosendahl, McKinsey Senior Partner London, for McKinsey’s Service Operations Practice, argues that  when it comes to consideration of AI adoption patterns for service-based businesses, within each business function there are three main types of activity - rule-based activities, deterministic activities, and probabilistic activities:

These can occur in both the front office and the back office and so the application of AI is hugely mixed, and the big opportunities need to be identified. You need to go all-in on two to three domains – don’t let a thousand flower bloom, pick a couple that will be really meaningful. One of these is customer care and operations, where it is possible to achieve over 50% productivity gains in call-centers.

In the back-office areas of Finance, HR and Procurement Rosendahl says there are also AI hotspots for adopting AI such as recruitment, learning or accountancy operations.

Leading clients are taking an agent-first model using conversational interactions via voice. They are now going beyond a human in the loop. Sectors such as financial services, technology and telecom have many real fast-mover actors. With banks, for example, agents are handling high-intensity processing in the middle office. Back-end processes are where you get real bottom-line impact and the adoption problem here is not to do with the technology, rather it is because the processes are super-complex, with, say 20 hand-offs between different teams and eight approval steps to get things done.

For example, if you want to recruit a sales rep, following that process flow takes between 60 and 80 days in order to get a request approved, a job spec written, invite people to interview and so on. In areas like this, agents have the biggest impact. AI adoption is not one train leaving the station, it’s many trains leaving the station. The nature of the work is different in services and some areas have much higher opportunities. Consequently, it will be implemented unevenly across many domains and industries.

We are not taking old school legacy into the future

Might part of the problem with AI adoption not providing value might be that not enough resource is provided for change management? Rosendahl reckons: 

For every dollar spent on technology clients should really be spending two or three dollars in change management. They need to do it but it is not common for organizations to make that commitment. This is why it is really important to focus on high impact opportunities.

You need to take a greenfield approach, applying AI to look at the processes we have. Now we need to start anew: what does the process look like in the agentic version? Then you need to migrate users in the 20 or so use cases across a process, in a step-by-step implementation. If you take a bottom-up approach, adopting Co-Pilot, this has some benefits by familiarising people with the technology, but it won’t make a material impact.

There are tangible examples available of agentic-enabled business value so clients should go and see them. They now exist, illustrating true reinvention by taking an AI-first model. There is no patience any more for the idea that if you invest in AI and data this will at some point pay off. But if you can deliver on an ambitious business case then organisations will invest.

Other service providers have spoken about the true value of AI lying with brownfield modernization. However, this is not where Rosendahl thinks the main value of the technology lies: 

I am not talking about modernizing legacy applications where you take the traditional model, find improvements and implement via a backlog – let’s not do that. Instead take a white paper and draw your process from scratch leveraging AI-first – we are not taking old school legacy into the future, we are redrawing the process, taking a greenfield approach.

As for the likely impact that AI will have on offshoring and outsourcing models, his view is that: 

AI-first is disruptive to this business, and we are currently working with clients specifically targeting their outsourcing contracts. Clients are bringing an AI automation understanding to the suppliers of outsourcing services. We are advising them to negotiate by saying, ‘Give me 30% productivity increases or we will take to the market for a new tender.’ Clients like this approach because it doesn’t affect their own headcount, so they can move faster without restructuring costs. 

In response outsourcing suppliers can use their expertise in the processing of certain domains to deliver these productivity benefits. These players used to be disruptors and now they are being disrupted. All players now have to understand where the source of their value really is – it can no longer be in repeatable, lower value processes.

Meanwhile, with respect to the impact of agents on the McKinsey business model, Rosendahl’s response is: 

McKinsey itself is a really heavy user of AI and has been for some time now but we are well-positioned because we have a value-based way of looking at it. We get the best possible value for customers by providing them with the best ROI.

My take

If agentic AI is to cross the chasm of enterprise adoption, it will not occur by the proliferation of pilot projects based on different tech vendor promises. McKinsey is saying take time choosing the domain to tackle and really crack it by starting anew using AI. Rosendahl posed this question to me: 

Is this a higher risk approach than continually patching a broken process?

Despite Rosendahl’s words ringing in my ears, when it comes to technology adoption 'rip-and-replace' is never an approach for the faint-hearted. However, it might just provide the value from agentic AI that CEOs are looking for.