Now accepting guest editorial from SAP, Oracle, Workday and ServiceNow practitioners.Pitch a guide
The Sign-Off Table

Lesson 6 of 6

How AI changes the work of key users, consultants and testers

9 min

Listen to this lesson

0:00 / 7:55

SAP's direction is clear: users describe outcomes, assistants coordinate, agents act, and people confirm. That moves effort away from clicking through transactions and toward specifying, checking and approving. This lesson turns that into concrete changes for three roles at Contoso, and ends with what a quality-minded team should put in place before the first agent goes live.

Key users and process owners

Maria and Dan will spend less time navigating apps and more time judging proposals. An agent that prepares a collections call or proposes a clearing still needs someone who knows the customer, the contract and the policy. Their new tasks are to define when an agent may act and when it must stop, review the Source and AI Reasoning section of agent answers, own the acceptance criteria for agent behavior, and report wrong outputs as defects rather than shrugging.

Consultants

Consultants gain SAP Joule for Consultants for answers grounded in SAP documentation, and Joule Studio for building agents where SAP has no standard one. Their new responsibilities are to check that a proposed agent really fits the process and the edition, to design prompts, tools and guardrails for custom agents, and to tell the customer clearly which parts of a design depend on announced rather than available features.

Testers and test leads

For Priya's team, AI shows up twice: as a tool and as a test object. As a tool, SAP Cloud ALM is adding agents for test scoping, test preparation and test data. As a test object, every agent Contoso turns on is a non-deterministic component in a signed-off process. The same request can produce slightly different wording, and occasionally a different recommendation. Testing an agent means testing outcomes and controls (did it pick the right documents, did it ask for confirmation, did it respect authorizations), measured across a set of cases rather than once. Course sap-ai-201 covers how to evaluate an agent pilot.

RoleLess time onMore time on
Key user / process ownerNavigating apps, collecting data from several screensSetting limits, reviewing proposals, owning acceptance criteria
ConsultantSearching documentation, writing first draftsFit checks, agent and prompt design, honest status labeling
Tester / test leadManual scoping and test data preparation (as Cloud ALM agents mature)Evaluation sets, control checks, evidence for non-deterministic behavior
PMO leadChasing status updatesAI Unit budgets, AI inventory, adoption and value tracking
Where effort moves when agents arrive, by role.
The Sign-Off Table · thesignofftable.com/academy/sap-ai-101

Program and PMO leads

Grace's job changes too. AI Units are a consumption budget, so somebody has to forecast them, watch the balance in SAP for Me and decide what happens when a pilot burns faster than planned. AI capabilities also need an owner in the application inventory, the same way an interface or a custom report has one. And because SAP's AI portfolio moves monthly, the PMO needs a habit of re-checking statuses each quarter and after SAP events, instead of copying last year's slide.

Start hereHigh value, low risk: summaries and preparation work a person reviews, such as an account analysis before a customer call
Pilot with strong controlsHigh value, higher risk: proposals that change financial data, such as clearing or dispute creation, with confirmation and thresholds
Quick wins onlyLow value, low risk: navigation and lookups that base Joule skills already cover
Avoid for nowLow value, high risk: automating rare, judgment-heavy decisions with little volume
Where to start: pick the first scenario where value is clear and the risk of a wrong answer is contained.
The Sign-Off Table · thesignofftable.com/academy/sap-ai-101

A readiness checklist before the first agent

Contoso's path from interest to a first agent in production.
The Sign-Off Table · thesignofftable.com/academy/sap-ai-101
  1. Name a process owner for each assistant and agent, and record what the agent may do without asking.
  2. Add AI capabilities to Contoso's application inventory (SAP AI Agent Hub can hold SAP and non-SAP agents).
  3. Agree how AI Unit consumption is monitored in SAP for Me and who reacts to an overrun.
  4. Define evidence for sign-off: evaluation cases, authorization tests, confirmation checks and the exported conversations.
  5. Write down what users must never paste into any AI prompt (Course sap-ai-102).

What would you do?

Priya is asked to sign off the Accounts Receivable Assistant pilot. The vendor's demo script ran perfectly five times. What evidence should she insist on first?

Key points
  • Agents shift effort from navigating to specifying, reviewing and approving.
  • Testers test outcomes and controls across a set of cases, because agent outputs vary.
  • Before the first agent: one process, GA for your edition, prerequisites, AI Unit sizing, evidence in the quality system.
  • Tell leaders what is free, what is paid by use and what is still early, in one breath.

Check your understanding · not graded

Lesson check

  1. 1. Why must agent testing use a set of cases rather than one run?

  2. 2. Which SAP Cloud ALM agents listed in the Discovery Center matter most to a test lead?

  3. 3. Which item belongs in the sign-off evidence for a first agent in production?