PrymeIntelligence

Training teams for AI agents·Article

What outsourcing providers get wrong about AI agent training

What outsourcing providers get wrong about AI agent training: SOPs taught instead of agent bounds, SLA pressure on approvals, no proof of competence.

Outsourcing providers train staff on each client's procedures and assume the agent fits inside them. It does not. Every client's agent runs on its own bounds and limits, staff move between accounts carrying the wrong habits, service levels reward fast approvals, and when a client asks for proof that reviewers are competent, the provider has attendance sheets instead of evidence.

Why isn't the client's procedure enough?

An outsourcing provider's training model is built around the client's standard operating procedure: learn the steps, the systems and the service levels, pass a floor assessment, go live. When the client brings an agent into the process, the provider adds a module on the new screens.

The procedure tells staff what the process should do. The agent's bounds tell them what the agent may do, at which depth, and why it stopped. For one client the agent may issue credits up to the customer's limit at depth 3; for another the same kind of credit is held at depth 2 because that client set a tighter limit. Staff who learn only the procedure cannot read the verdicts. Teach each client's agent bounds explicitly, alongside the procedure, with real held cases from that client's queue.

What goes wrong when staff work across clients?

Providers move people between client accounts to balance load. A reviewer who spent the morning on a client whose agent holds every refund above a low threshold may spend the afternoon on a client whose agent issues larger refunds within its bounds. Habits carry over: the reviewer either over-checks the second client's routine work or, worse, assumes the first client's looser rules apply elsewhere.

Approval rights should be granted per client, per depth, not per person in general. A provider can adopt the practice of tying each right to a current certificate for that client's agent, so a reviewer certified on one account cannot approve held payments on another until assessed there. Keep one register that shows, for each person, which clients and depths they are certified for and when each expires.

How do service levels distort review?

Outsourcing contracts measure turnaround, backlog and handle time, often with penalties. An agent that holds actions for a person adds items to the backlog, and every held item is a clock running.

Under that pressure, reviewers approve quickly. The client's record then shows a high approval rate that may hide poor decisions, and when the client's agent comes up for renewal, its certificate lapsing after 90 days, the evidence may not support the depth the client wanted. Negotiate service levels that separate held-action review from routine handling, measure review quality by sampling decisions, and make it plain to staff that a correct hold beats a fast release.

What evidence will clients ask for?

More than they used to. A client remains accountable to its own regulators and customers for what its agent does, including the decisions the provider's people make on its held actions. Financial clients in the EU manage ICT third-party risk under DORA and will want to see how their providers' staff are kept competent. Clients everywhere will ask who approved a disputed action and whether that person was trained.

Attendance sheets do not answer that. Certificates that name what was assessed, carry an expiry date and can be checked at source with the issuer do. So does a register the client can inspect, and the client's own record showing each approval with its reason. Build the evidence before the client asks for it; it is also a strong point in a bid.

Why does new-client ramp-up need its own plan?

Providers win work and ramp up quickly, often hiring and training a large cohort in weeks. If the new client runs agents, that cohort must be ready to review held actions on day one of go-live, not after a few weeks of learning on live cases.

Before committing to a go-live date, test where the team stands. How to run an AI agent readiness assessment in an outsourcing provider covers the steps: test the shared vocabulary in the client's industry, look at scores by concept, and find the gaps before they reach the client's customers. Then run the agent in shadow at low depth while the cohort practises on its drafts.

What should an outsourcing provider do instead?

  1. Teach each client's agent bounds and limits alongside its procedure.
  2. Grant approval rights per client and per depth, and keep a register.
  3. Separate held-action review from turnaround targets in contracts.
  4. Hold certificates that clients can check at source.
  5. Assess readiness before each new client goes live.

The general order is in the guide to training teams for AI agents. For detail, see what an AI agent training programme for outsourcing providers should include and how much AI agent training costs for outsourcing providers. Providers serving healthcare clients should also read what hospitals get wrong about AI agent training, because many of those mistakes land on outsourced administration.

Where the Academy fits

Pryme Intelligence Academy offers a free readiness assessment for teams of up to 25 people, which suits a provider's new client cohort: the team takes the Foundations test set in the client's industry, and the report shows scores by concept, the gaps, and the first agent to start with. Operator certificates are issued through VerifyDoc.ai, so clients can check them at source by number or QR code.

Questions people ask

Who is accountable when a provider's reviewer approves a bad action?

The client remains accountable to its own regulators and customers, and the provider is accountable to the client under the contract. The record shows who approved and why, so both sides need evidence that the reviewer was competent at the time.

Should a provider use its own trainers or the client's?

Usually both. The client's process owners know what a correct decision looks like; the provider's trainers know how to run training at scale. Have the client sign off the assessment cases.

How do we keep certifications current with high turnover?

Build assessment into onboarding, track expiry dates in one register, and remove approval rights the day someone leaves or changes account. Turnover is a reason to make certification routine, not a reason to skip it.

Train the people who run your agents.

Pryme Intelligence Academy: free courses on the real product, audio lessons, labs in a free workspace, and certificates anyone can check on VerifyDoc.ai. Teams get seats, an admin view and a free readiness assessment.