Set the Rules for AI in Service Delivery Before You Scale It

You can put AI agents and copilots into shared services and BPO delivery and prove what they are worth. The route is to set four things for each service capability before you scale it: how much an agent may decide on its own, where human oversight sits, what evidence each decision leaves, and which outcome measures replace effort-based ones. You already hold most of what you need: the contracts and SLAs that set today’s commitments, the process descriptions each team keeps, and the people who know which local rules exist for a reason.

This episode follows one IT help desk ticket that an agent resolves in three seconds against a contract that expects fifteen minutes of human effort, and shows how a measure built for labour can read that result as underdelivery. It works through who stays accountable when an agent acts inside a process a provider runs for you, using the systems and rules your retained teams own, and how a dashboard showing 10,000 transactions can look excellent locally while an enterprise control goes unchecked. The episode draws on ISG’s 2026 State of BPO report, which reports AI moving BPO beyond labour-based delivery toward optimisation, stronger governance and measurable business outcomes, and on the NIST AI Risk Management Framework, which sets out a use-case and risk-based approach to governing, mapping, measuring and managing AI risk across the system lifecycle. The answer it reaches is to standardise the agent and set its limits locally: one reusable capability, with the variation a law such as GDPR or a client contract requires recorded beside it.

The practical starting point is one AI-enabled service capability. Write down which decisions the agent may make without approval, who signs off the rest, and how you will measure its value once transaction effort falls. IGX360 Insights shows which of the process descriptions behind that capability you can stand behind, who owns them and how current they are, so the limits you set rest on something you can evidence. Which of your services would you put an agent into first?