Agentic AI and the Future of Global Business Services

Global Business Services (GBS) is being redefined in the age of AI. Agentic technology is enabling GBS to move beyond traditional automation and expand as it becomes the operational backbone of AI transformation in many organizations. So what should organizations think about the future of GBS to come out ahead?

In this episode, Christian Johnson speaks with McKinsey partners Heiko Heimes and Josh Peters. They explain why GBS is strategically placed to own workflow orchestration and the talent implications of AI transformation. They also tackle one of the most pressing strategic questions facing operations leaders today: should you automate first or right-shore first, and how do you decide?

TL;DR

A ~25-minute episode of McKinsey Talks Operations on the McKinsey & Company channel, published 3 August 2026 — host Christian Johnson with partners Heiko Heimes (Cologne) and Josh Peters (Washington DC). Global Business Services (GBS) is defined in the intro as “a centralized organization providing shared services across business functions and geographies.” Load-bearing claims:

1. What agentic AI changes for shared services. Peters: GBS “has been a hotbed for automation for many years”; the difference now is “the ability to act independently, understand context, adapt and make decisions in real time… going beyond very kind of rule-based individualized decisions, starting to apply judgment.” The consequence he draws is a category shift: AI moves from “just an augmentation or an assistant” to “much more of an individual contributor.”
2. The counter-intuitive headline: GBS is not uniformly shrinking. “You might expect that GBS’s would be shrinking. And we do see that in some areas there is a lot of automation being applied. But we also see in other areas it’s really expanding. More people, more scope, more budget, more spend. So AI isn’t replacing GBS yet. It’s actually augmenting it.” Their net forecast of the two countervailing forces: “GBS’s are likely to be 20 or 30% smaller but dramatically more productive than they are today.”
3. Three axes of scope expansion (Heimes). (a) Up the value chain — from transactional/administrative work (procure-to-pay, record-to-report) toward “advisory, strategic, judgment-based” work such as strategic procurement and financial forecasting and planning. (b) Into work the wider organisation never mastered — strategic workforce planning in HR, supply-chain planning: “why shouldn’t GBS be the one holding the tools in their hands and doing exactly that type of work that the rest of the organization hasn’t mastered in the past.” (c) Frequency — internal controls and auditing that previously ran irregularly or on a sample basis can become continuous: “you suddenly can use AI to really do real-time processing… so you continuously check transactions for compliance and irregularities.”
4. The FP&A worked example: from answer to next best action. Reporting consolidation and advanced analytics are the long-standing GBS play. “The interesting twist has been: how do you actually not just think about giving an answer to a question, but also think about a next best action?” A client FP&A co-pilot does not stop at “profit is down because of seasonality or there was a weather event” but adds “here’s what I think are the two or three next questions you should ask or the two or three next actions you should take.”
5. The talent model: pyramid → diamond. Heimes: “we think about it as a diamond-shaped setup… you will have more in the middle layers of the GBS organization, simply because we are seeing less need for managing and running the more transactional type activities where automation — also in the more classical sense, RPA plus AI — will do a significant chunk of the work.” Work shifts “to the analytical, decision, judgment-based” end, requiring different skills, “in addition to being able to manage the AI, especially the agentic AI side… you also need the people who are able to manage the agent force.”
6. The consequence the panel is most worried about: the entry-level training cycle. Peters: “we have to move a lot faster to train and upskill entry-level talent because we just won’t have as long of a cycle for that layer of talent to develop technical expertise and move into the supervisor layer. If a lot of that layer of talent goes away, we’re going to have to come up with new mechanisms to train people… and get them ready to orchestrate and manage digital talent.” Heimes adds the compounding problem: GBS organisations already run high attrition, so constant rehiring and retraining is the baseline, and with less entry-level intake the answer has to be mobility — “you need to have people from the remainder of the organization outside of GBS to come in, perhaps people from GBS go out.”
7. The adoption lighthouse that failed on the human side. Peters’s example, in customer order and account management: the company “developed just an incredible technology” that knew the where and the why of each call and could infer customer intent — “and they just really struggled to get the sales force to adopt it. I think there was a lot of fear.” His two lessons: “the people side of the equation is as big or bigger than the technological side,” and the framing determines adoption — “can you show them how this gets to better outcomes for both them and the client, versus can you just show them oh well we can divert 50% of the calls and therefore we’ve driven a lot of savings.” The augmentation upside he names is status, not speed: reps resolve calls “that normally would have been escalated to somebody sort of above their level,” which is “a more gratifying experience for the employee.”
8. The strategic question the episode is built around: automate first or right-shore first? Not either/or. Right-shore first — migrate work into GBS for talent, scale, standardisation and quick value capture, then apply AI to the now-standardised work; suited to organisations “seeking to accelerate value capture.” Automate first — “completely reimagining the process flow before migrating some of that work into GBS”; suited to organisations “thinking about building a long-term digital capability” in core strategic processes with large enterprise-value unlock. Peters: “it’s not one-size-fits-all. It’s actually a little bit more horses for courses. Some parts of a process may benefit from moving to GBS first. Other parts of a process may benefit from automating first.”
9. The three decision factors (Heimes). (a) Your own experience — current GBS setup, and track record leveraging AI and technology to improve processes. (b) Impact expectation and timing — “when do you need to achieve what,” where impact spans cost efficiency, quality of outputs, and operational stability. (c) Process type — procure-to-pay, order-to-cash and record-to-report differ in how easily AI can be adopted. Observed pattern: green-field situations with little existing shared-services footprint, urgency, and “a lot of low-hanging fruit… processes that are not incredibly standardized and mature today” still choose right-shore first. Automate-first appears where the target is enterprise-value unlock — “reduce the amount of revenue leakage… improve price realization… squeeze out a few more basis points of EBITDA” — rather than headcount or process cost.
10. The three E’s, and why order-to-cash is the flagship. Heimes: “we are moving away from a mere efficiency debate… to a more holistic set of ambitions. I always think about the three: efficiency, effectiveness and experience, as in customer experience.” Order-to-cash is both an efficiency play (fewer resources), an effectiveness play (less revenue leakage — “make sure that people pay on time as originally aligned”) and, because it is customer-facing, an experience play that “can even increase your customer experience levels, thereby create much more bonding between you as an organization and the customer.”
11. State of adoption, and the source-to-pay lighthouse. “Roughly half have started to pilot and work on it.” The lighthouses of the last 9–12 months are concentrated “in the more classical finance process domains.” The worked case: GBS had long owned the back end of source-to-pay (invoice in, checked against purchase order, paid, supplier interaction). The client reimagined the process so AI takes over “a good chunk of the supplier identification, supplier selection, which now is being managed by GBS” — pushing GBS into the front end that “very much determines how smoothly you can run the back end,” giving GBS accountability across the whole flow and, downstream, more satisfied suppliers paid on time.
12. The episode’s sharpest line — set a north star, or automate the wrong work. Peters on a client that asked him to kick the tires on its AI strategy, which was largely “what are the set of 10, 20 use cases in each area that we think might have the most potential”: “I actually think it’s very important to set a north star and not just sort of experiment… if the first step that you’re taking is not do we have the right materiality threshold in place for how many of these things we go after and is there any low-value work that we can just eliminate — you’re still just automating the wrong work.”
13. Closing calls to action. Heimes: GBS “should be an accelerator for the AI journey of any organization” because it is “the one place where you already have a good tech foundation” and already looks at processes end-to-end rather than in functional silos — “don’t start too small, don’t start too large. Find that sweet spot in the middle where you create a lighthouse without overburdening your organization with too much complexity.” Peters closes on the historical rhyme: “I remember 2017, 2018, right after RPA got mature, a lot of people including all the consultancies and a lot of the SIs said GBS is dead. And I just think I hear a lot of that sentiment now… I would just say the death of GBS is greatly overexaggerated.” His prescription is to run both levers — next-gen AI levers that reduce work and classical levers that move more scope into GBS — so the organisation moves “away from just transactions and cost and more towards outcomes and value.”

What was actually ingested

The full auto-generated (ASR) English caption track (609 segments, consistent with duration: 24:45 / length_seconds: 1485). The episode has no chapter markers. Speaker turns are not labelled in the captions and Heimes and Peters hand off frequently mid-topic; attributions above were reconstructed from context and the two partners’ stated remits, and a small number of short interjections are ambiguous between them. The show’s standard open and close are excluded from the substantive summary. The ASR consistently rendered “GBS” as “GPS”; this was corrected at acquire time.

Dynamic-capabilities tagging

• digital-transforming/redesigning-internal-structures — the diamond-shaped talent model is a structural redesign of the shared-services organisation: a thinner transactional base, a thicker analytical/judgment middle, a new role managing the agent force, and — because the entry-level rung that fed the supervisor layer thins out — new mechanisms for training and new GBS-to-non-GBS career mobility. The source-to-pay case is the same redesign at process level, moving accountability for the front end of the flow into GBS.
• digital-seizing/balancing-digital-portfolios — automate-first versus right-shore-first is a portfolio-balancing decision, and the panel treats it as one: decided per process and even per part of a process (“horses for courses”), against three named factors (existing capability and AI experience, expected impact and its timing, process type), with materiality thresholds set before a use-case list is drawn up.
• strategic-renewal/business-model — the three E’s reframing moves GBS’s value proposition from cost arbitrage to a combination of efficiency, effectiveness (revenue leakage, price realisation, EBITDA basis points) and customer experience, and the panel’s closing framing is explicitly a renewal of what the function is for: “away from just transactions and cost and more towards outcomes and value.”
• contextual/internal-barriers — the customer-order lighthouse failed on adoption, not capability: “they just really struggled to get the sales force to adopt it. I think there was a lot of fear.” High baseline attrition in GBS organisations is named as a second structural barrier that AI-driven changes to the entry-level rung will worsen before they improve.

Linked entities and concepts

• McKinsey & Company — publisher; McKinsey Talks Operations strand. Updated in this ingest.
• BBC AI Decoded — published two days earlier; the same workflow-scope-before-tool-deployment argument at broadcast altitude; see this source’s relationships:.
• McKinsey software-development panel — the same firm making the same diagnosis for a different function.
• Catlin — both locate the durable advantage in reimagined end-to-end processes; this episode adds the sequencing decision.
• Emerson, Kropp et al. — the diamond talent model as a function-specific instance of reshape-over-replace.
• micro-productivity-trap — “you’re still just automating the wrong work” and the materiality-threshold-and-elimination-first prescription.
• automation-vs-augmentation — the automate-first versus right-shore-first decision, and the framing lesson that augmentation sells where headline call-diversion savings do not.
• ai-employment-effects — the 20–30%-smaller-but-more-productive forecast and the entry-level training-cycle problem.
• enterprise-ai-adoption — GBS as the organisation best placed to own end-to-end workflow orchestration; “roughly half have started to pilot.”
• ai-agents — the “agent force” that GBS staff are now expected to manage, and agentic AI as “individual contributor” rather than assistant.
• dynamic-capabilities — the four cells tagged above.

Dangling (single-source mention, deferred per author-entity promotion): Christian Johnson, Heiko Heimes, Josh Peters.

Source quality note

Auto-generated (ASR) transcript, no speaker labels; “GBS”→“GPS” and “McKinsey”→“Mackenzie” mistranscriptions corrected at acquire time (see the raw file’s notes:). This is first-party consulting-firm content about the firm’s own service line — the partners describe client engagements they ran, and GBS transformation is a McKinsey Operations offering. Every quantitative claim in the episode is unattributed practitioner estimate: the 20–30%-smaller forecast, “roughly half have started to pilot,” and the 9–12-month lighthouse window carry no sample, method or denominator. All named client cases are anonymised. Treat the framework contributions (the diamond talent model, the automate-first/right-shore-first decision, the three E’s) as the durable content and the numbers as directional.