Beyond the Hype: Status of Gen AI in GBS and Strategy for 2027
October 2026
Overview
The discussion examined where GBS organizations stand with Gen AI (including agentic AI) at the end of 2026, where they expect to be by the end of 2027, and how they plan to close the gap. Panelists rated their maturity on a 5-point scale (1 = exploring, 2 = a few pilots, 3 = several use cases in production, 4 = scaling across multiple towers, 5 = end-to-end AI-enabled processes). Current positions ranged from 2 to 3.5, and all three panelists are targeting 4 or 5 by the end of 2027.
Panelists:
- Business Process Manager, AI Solutions, Company A (diversified global manufacturer)
- Senior Manager, Global Continuous Improvement and Master Data, Company B (global automotive supplier)
- IT Director, GBS Financial Services Digital Products and Delivery, Company C (global consumer products company)
Note that this document is a summary. Peeriosity-100 members have full visibility to company names, presenters, Peercast attendees, and a recording of the Peercast discussion.
Key Themes
- The technology works; the process and data often don’t. All three panelists said AI itself rarely failed. Efforts stalled because of non-standard processes, inconsistent regional practices, messy master data or weak knowledge bases. AI won’t fix these alone, but it can be the catalyst that justifies fixing them.
- Business and IT together beat either alone. Asked whether IT or the process owner should lead, Company C’s answer was both: a “two-in-a-box” pairing of a process owner who can envision the future process with technical builders. Company A arrived at the same model.
- From isolated use cases to end-to-end processes. Early wins are targeted tasks (deduction handling, invoice comparison, contract extraction, status chatbots). The 2027 goal is end-to-end flows combining AI, RPA and other tools, since fixing one step can simply move the bottleneck.
- Governance is becoming the critical layer. As citizen-built agents proliferate, companies are standing up AI offices, functional champions, agent inventories and evaluation infrastructure to avoid duplication, ensure output quality and manage cost.
- From “build muscle” to “show me the money.” Leadership tolerance for exploratory work is giving way to value expectations, accelerated by rising token costs. Efficient engineering choices, such as lighter models, are becoming a cost lever.
- Mindset is the biggest barrier. People expect AI to mimic today’s process. The task is to redesign the process with AI inside, while addressing resistance from staff who fear job loss.
Panelist Company Snapshots
| Company A | Company B | Company C | |
|---|---|---|---|
| Scope | GBS (one of ~20 AI workstream leads across the company) | GBS finance continuous improvement; separate team for large end-to-end projects | GBS financial services, the largest GBS tower |
| Team | 1 full-time AI lead (role created Nov 2025), 1 direct report on governance, plus IT developers and PMs | Functional champions, AI office, IT support for complex builds | ~10 dedicated (mostly IT), ~10 business-side, ~50 including part-time SMEs, in an org of ~2,000 |
| Maturity today | 3 | 2 | 3 to 3.5 |
| Target, end of 2027 | 4+, moving end-to-end | 4 (aspiration: 5) | 5 |
| Use cases in production | 3: customer deduction email parsing; contract and deduction term lookup; P2P invoice status chatbot | 11 (11 more in progress): consignment invoice vs. receipt comparison, multilingual lease extraction, master data chatbot, AR lookups, collections emails | 3 AI products: invoice analytics (4.5M invoices/yr), end-user support chatbots (~1M requests/yr), conversational data analytics |
| Results | ~20% of the time of 4 people saved so far | Mostly personal productivity; savings not yet large | Chatbots resolve 50 to 60% of questions (vs. 5 to 6% before AI); invoice analytics live in 2 to 3 markets, 15 to 20 more by June 2027 |
| Key constraint | Skills to build production-grade solutions; data and process standardization | Data quality; IT capacity; mindset change | Process standardization, data, knowledge, culture |
| 2027 focus | End-to-end flows, data foundations, likely a centralized AI COE | Governance model, functional champions, scaling beyond pilots | AI governance and evaluation infrastructure; evolving human roles |
Notable Takeaways
- Start where there is readiness. Company A abandoned its first use case (blocked invoices) because of political history and resistance, not the technology. Lead with willing teams rather than forcing adoption.
- Keep several use cases in motion. AI development is unpredictable; a portfolio protects against any single effort stalling.
- Not everything needs AI. Company B regularly redirects requests to RPA or simpler bots, which can be better and cheaper.
- Build reusable capabilities, not one-off fixes. Company C frames its work as horizontal “AI products” usable across teams; Company A evaluates each use case for how it builds a foundation for the next.
- Get proactive, not just responsive. With chatbots absorbing volume, Company C is asking why users need to ask at all, for example notifying suppliers of a blocked invoice before they inquire.
- Citizen development needs guardrails. AI makes building so easy that ungoverned solutions proliferate quickly; Company C had to contain and roll back a couple.
- Watch the cost of tokens. People with software engineering backgrounds tend to build more cost-efficient solutions; cost management is the next evolution.
- Large documents remain a challenge. Company B’s user-built agents struggled with 80-page contracts and needed IT help.
Audience Q&A Highlights
- How were Company A’s use cases prioritized, and did you commit to hard savings? Leadership committed to hard-dollar savings upfront but set no fixed timeline. Steady progress and line of sight to value kept support; one use case in development is expected to cover the full commitment.
- Should Gen AI be led by IT or the business process owner? Both. The winning formula pairs a process owner who can envision the future process with technology delivery.
- Is executive leadership OK with fuzzier value creation? Initially yes; Company A’s chief digital officer wanted people to “go to the gym” and build skills. Now the CEO is asking for demonstrated value, partly because token costs are rising.
- What talent should we hire or develop? Curiosity and willingness to learn matter more than deep AI experience, since no one has more than 3 to 4 years in Gen AI. Look for self-starters who can execute and also handle change management and influencing.