What AI-native means
Every function has agents and automations handling repetitive work, decisions draw on live cross-system intelligence, and AI is part of how work happens, not a tool people switch to.
The starting point · the execution gap
SAP Runs the Business. It Was not Built to Run the Work.
SAP manages the processes you depend on, and does it exceptionally well. But these systems were built to manage business logic, not to deliver the modern experiences that people, and now AI, need to actually get work done. The logic already exists inside SAP; what is missing are the applications and intelligent agents that turn that logic into work getting done, on any device, connected or not.
That distance between what SAP manages and how work actually happens is the execution gap. Becoming AI-native is how an organization closes it, and this playbook is how you get there, use case by use case, until execution runs itself and people are free for judgement, relationships and decisions. That end state is the autonomous enterprise for SAP.
Executive vision
Not Adding AI Tools. Becoming AI-native.
An AI-native organisation is one where AI is embedded in how every team works, decides and delivers, not bolted on as a separate tool. Agents handle the repetitive work, decisions are powered by live intelligence across SAP and connected systems, and AI is a natural extension of how people do their jobs.
Humans + AI, not humans vs AI
This is about leverage, not headcount. Low-value work moves to agents; human intelligence is reserved for strategy, relationships, creativity and judgement.
The north star
Embed AI so deeply into operations on SAP that it becomes part of the organisation’s DNA: scaling capability without scaling headcount, and closing the execution gap for good.
The enablers
Three products, one path to AI-native.
The programme in this playbook is made real by three Neptune products working together. They are named throughout, because each carries a distinct part of the journey.
The execution platform – Neptune DXP
The SAP-native execution platform. Builds and runs the business applications people work in, with native SAP integration, offline and any-device support, across ECC, S/4HANA and hybrid landscapes.
Agentic AI for business – Naia for Business
Embedded, governed AI agents that execute SAP business processes on the user’s behalf. The MCP Server securely exposes SAP business capabilities as governed tools for enterprise AI agents.
AI for developers – Naia for Developers
Naia Build generates production-ready SAP business applications with AI, using your semantic SAP business knowledge and reusable templates. Hours instead of months.
The through-line
Neptune DXP, Naia for Business and Naia for Developers help an organisation become AI-native, unlock the autonomous enterprise for SAP, and close the execution gap. The platform is the enabler; the business outcome is the value.
The execution gap
The distance between what SAP manages and how work happens.
SAP manages the processes your business depends on, and it does that exceptionally well. But it was built to manage business logic, not to deliver the modern experiences your people, and now AI, need to actually get work done, on the shop floor, in the office, and out in the field, on any device, connected or not.
The logic already exists inside SAP. What is usually missing are the applications and intelligent agents that turn that logic into work getting done. Turning SAP processes into applications people enjoy using, and agents that can act on them, has traditionally taken months of custom development, while your requirements change faster than a development cycle can keep up. That distance is the execution gap. As you adopt AI, it only widens. Closing it is the whole point of this playbook.
What it costs you today
Work happens in spreadsheets, paper and workarounds because the SAP screen does not fit the job. Data is entered late or twice. Decisions wait on reports. Every one of those is the execution gap showing up in your operation, and in your numbers.
Why your SAP is different
Your customisations are the advantage, not the debt.
You have spent years, often decades, shaping SAP to fit how your business actually runs. That is not technical debt; it is your operation, encoded in software. It is also exactly what most AI cannot see.
Generic AI tools can reach SAP data, but they cannot understand what it means, because the meaning lives in the logic and configuration around the data, not in the data itself. An agent that does not understand your pricing rules, your plant setup or your approval flows cannot safely act on them. Neptune runs natively inside SAP and treats your standard and custom logic alike, so applications and agents inherit the way your business really works. That is why an agent built on Neptune can genuinely participate in a process, not just describe it.
How Neptune is different
Activate, abstract, accelerate.
Three things Neptune does that turn your SAP investment into modern applications and AI agents, fast, and without a rip-and-replace.
Activate
Turn your SAP business processes into modern applications, with embedded AI agents that act on them. Native to SAP, backed by 15 years of ECC, S/4HANA, BTP and ABAP expertise, so it works with your logic, not around it.
Adapt
Build an application once and run it across SAP ECC, S/4HANA and hybrid landscapes. Your investment is protected as you modernise, and your core stays clean.
Accelerate
Generate applications with AI, drawing on your own SAP business knowledge and proven templates. Fit-for-purpose apps in hours and days, not months.
What you actually buy
You invest in outcomes. The platform stays behind them.
You are not buying a development platform to evaluate. You are improving receiving, procurement, maintenance or inspections, with the technology as the means, not the message.
Neptune, inside SAP
The layer that activates your SAP logic and lets agents act on it, safely.
A business application
A fit-for-purpose app for your process, with embedded AI agents.
The outcome
Faster receiving, fewer stockouts, less downtime, stronger governance. What the business feels.
With SAP, not against it
Neptune accelerates your SAP strategy. It does not replace it.
Most organisations will run SAP ECC, S/4HANA, or a mix of both for years as they modernise. Neptune meets you where you are today and stays with you as you move, one consistent application layer across your landscape, clean core preserved.
Proven where it matters most
Some of the most demanding organisations in defence, oil and gas, and regulated industries trust Neptune every day for mission-critical operations. The question is not whether Neptune is a better platform; it is whether you can deliver the applications your people need, faster and more economically. You can.
Layer 1 – The applications people use
Intelligent applications
The starting point, and the front door. Fit-for-purpose business applications for your SAP processes, on any device, online or offline, that connect your people to their data and to the agents working alongside them. No SAP screens to wrestle with, they just do the job.
What it looks like on the floor
- A receiving app that scans a delivery and posts the goods receipt to SAP in seconds.
- An inspection app that logs findings and raises the notification for the critical one.
- A maintenance app that shows a technician their work orders and pre-fills the usual parts.
Powered by Neptune DXP & Naia for Developers
Neptune DXP builds and runs these applications with native SAP integration and genuine offline capability, so they work in a dead zone and post the moment there is signal. Naia for Developers (Naia Build) generates them with AI, starting from proven templates that already cover roughly 80% of the app, so a fit-for-purpose application is delivered in a fraction of the usual time and cost.
Layer 2 · The work getting done
Agents and automations
The execution layer, embedded inside the application rather than bolted on as a separate AI product. Automations run a defined workflow on a trigger; agents use AI to read context, decide within the limits you set, and act. Either way, a person approves anything that writes to SAP.
Rule-based Automations
When the conditions are met, the automation fires. Consistent, reliable, every time. Ideal for the well-defined, repeatable steps of an SAP process.
Trigger >>> Action
AI-powered Agents
Agents reason and adapt. Inputs vary, and the agent handles the variation, executing steps of an SAP process on the user’s behalf within clear boundaries.
Reason · adapt · act
Powered by Naia for Business
Naia for Business provides the embedded, governed agents that securely execute your SAP processes, triaging an inspection finding, pre-filling a purchase requisition, or reconciling a goods receipt against open purchase orders. Human-in-the-loop by design: the agent drafts, suggests and prepares; your person reviews and approves before anything is written back to SAP.
Layer 3 · The knowledge behind the answers
The intelligence layer
An agent is only as good as the information it can reach. The intelligence layer turns your SAP data, and the systems around it, into a single source people and agents can query in plain language, drawing on live data rather than yesterday’s export.
Always current
Connected to the live system. Ask about stock, an order or an asset and you get today’s position, not last week’s.
AI-powered Agents
Connected to the live system. Ask about stock, an order or an asset and you get today’s position, not last week’s.
Only what the user can see
Every answer respects the person’s SAP authorisations. Agents never see or do more than the user is allowed to.
Powered by Naia for Business & MCP Server
This is where the MCP Server earns its place. It is how Neptune safely opens SAP to AI: it exposes your SAP business capabilities, reading a purchase order, checking stock, posting a movement, as governed tools that an agent can call, always inside the logged-in user’s permissions. Instead of building a bespoke integration for every request, an agent simply calls the tool it needs, and the MCP Server handles the SAP conversation securely. Naia for Business uses those tools to reason across your systems and act on them.
Layer 4 · Many agents, working as a team
Agent orchestration
The most advanced block, and the one that makes the autonomous enterprise real. Instead of one agent doing everything, an orchestrator coordinates a team of specialists, each handling one part of a process in parallel, and brings their results together for a person to approve.
How it works
An orchestrator takes a complex request, splits it into tasks, and hands each to a specialist agent. They work at the same time, each pulling from the systems it needs, then return their results for the orchestrator to synthesise. Your people guide the orchestrator and stay in control of the decisions.
Powered by Naia for Business
What it can do
A period-end operations view assembled on demand: one agent reconciles goods receipts, another checks inventory positions, a third flags open supplier claims. The orchestrator pulls it into a single brief and surfaces the risks, ready for a manager to act on, with any SAP posting still confirmed by a person.
Where this leads
Applications give people the interface, agents do the work, the intelligence layer gives them context, and orchestration coordinates them into a digital workforce. That is the autonomous enterprise for SAP, reached one layer, and one solution, at a time.
Preparation · readiness & foundations
The groundwork that makes everything after it work.
Readiness is not a gate you complete once. Get ready enough for the first workflow and deepen the foundations as you scale. The work runs on three fronts in parallel, each with a named owner.
Culture
An AI-first mindset before the tools. Frame AI as leverage not threat, train leaders first, and address “will I be replaced?” openly.
Data
Agents are only as good as the data they can reach. In an SAP estate, that means the custom objects and semantic business knowledge that make agents accurate, reachable as a governed surface.
Process
Capture how work is really done and define the gold-standard output. A one-page SOP per workflow becomes the spec for the automation and the training context for the intelligence layer.
Conservative route
Do not boil the ocean. Get one workflow ready, ship it in Phase 1, and let each success fund the next. Readiness deepens workflow by workflow, not company-wide up front.
The phased rollout
From co-pilots to the autonomous enterprise.
Phase 1 – Co-pilots & early agents – Assisted work and first automations
Fit-for-purpose apps and co-pilots connected to live SAP data, plus first agents on well-defined, repetitive tasks. Naia Build accelerates the apps; Naia for Business embeds the first agents. Visible wins for staff and fast, measurable ROI.
Outcome: early ROI and cultural buy-in.
2 Phase – Intelligence & expanded agents – From efficiency to intelligence
The intelligence layer centralises knowledge across SAP and connected systems via the MCP Server, while agents move from drafting and routing into analysis, enrichment and monitoring. Decisions become data-driven by default.
Outcome: faster, smarter decisions across functions.
Phase 3 – Multi-agent systems – Orchestrated, autonomous execution
Orchestrator agents coordinate the specialists within and then across processes. Complex workflows run end to end with a human approving outcomes rather than typing every step. The autonomous enterprise, on SAP.
Outcome: leverage at scale, people on judgement.
Finding the right first use cases
Where the opportunities come from.
A steady flow of good candidates beats a big-bang wishlist. The best ones come from the people doing the work, alongside the proven patterns we bring to the table. Four sources keep the pipeline full.
Front-line friction
The repetitive, low-value tasks your teams do every day. The richest source, because the pain is real and already understood by the people who feel it.
Process capture
As a team documents how work really gets done, the automatable steps fall out of the process naturally.
Proven patterns
The solution library: AI-native patterns for common SAP processes, ready to adapt to how you actually operate.
Each new connection
Every SAP area or system you connect unlocks a fresh batch of use cases that were not possible before.
How we find them with you
From the art of the possible to a business case.
Run team by team, this is how a room full of “AI could help here” becomes a shortlist of use cases worth building, with the numbers to back them.
#
1
Art of the possible
Show the team what AI can do for their work, with real examples from the solution library.
#
2
Surface the pain
Capture the day-to-day friction: the task, who it affects, the KPIs it touches, the outcome wanted.
#
3
Score & shortlist
Rank each idea on value against feasibility, live, and agree the shortlist in the room.
#
4
Business case
Turn the shortlist into a costed case with a target metric, and validate the technical fit alongside.
How we prioritise
Value against feasibility, then sequence.
Every candidate is plotted on the same two axes, so the order of work is deliberate rather than loudest-voice-wins.
High value · high feasibility – Do first
Easy to build on connected, read-only SAP data. The first candidates, they build trust and ROI fast.
High value · lower feasibility – Plan & invest
Higher value but harder, needing write access or new integration. Sequenced deliberately with IT and security.
Lower value · high feasibility – Quick wins
Modest value but easy. Batched together for cheap credibility and adoption momentum.
Lower value · lower feasibility – Skip or defer
Said no to clearly, so the backlog stays honest.
The four roles
A clear owner at each altitude.
Executive sponsor – AI leadership
Owns the strategy, phase sequencing and investment. Sets the roadmap from co-pilot rollout through the autonomy decisions, and makes the build-versus-buy calls.
Technical owner – AI Lead
Owns technical execution: the build log, the agent and tool catalogue, integration to SAP, evaluation and observability, and consistency across teams. The translation layer between strategy and shipping.
Functional owner – Process / LoB Lead
Owns their function’s transformation end to end: capturing the SOP, defining the gold-standard output, and signing off when an agent is good enough to ship to their people.
Grassroots owner – AI Champion
Owns day-to-day adoption inside one function. A front-line power user who becomes a peer trainer and surfaces what is broken or missing. The antibody against shelfware.
Who decides what
The RACI, in plain terms.
Responsible, Accountable, Consulted, Informed. Two accountabilities are deliberately split: the AI Lead is accountable that an agent is built well; the Process Lead is accountable that it is good enough to ship. Neither ships alone.
| Decision | Exec sponsor | AI Lead | Process Lead | Champion | IT / Security |
|---|---|---|---|---|---|
| AI strategy & roadmap | A/R | C | C | I | C |
| Use-case prioritisation | A | R | C | C | I |
| Build an agent / automation | I | A/R | C | C | C |
| Approve agent to ship to a team | I | C | A/R | C | I |
| New write access to SAP | A | R | C | I | A |
| Set agent autonomy level | A | R | C | I | C |
| Drive adoption in a function | I | C | A | R | I |
Meeting rhythm
Rollout dies in the gaps between meetings, not in them.
Weekly backbone
Sponsor and AI Lead on strategy-to-build, an AI Lead delivery stand-up, and the Process Lead and Champions check-in, the single most-skipped meeting and the one whose absence kills adoption fastest.
Monthly review
All four roles. The AI Lead reports velocity, evaluation scores and platform health; each Process Lead brings one win, one blocker, one ask; Champions demo. The heartbeat the rest of the organisation sees.
Quarterly reboot
Rebuild the roadmap, retire agents not earning their keep, re-score phase progression and autonomy levels, and onboard new Champions.
How success is measured
Two scoreboards: programme health and business value.
Productivity is only half the story. Lead with programme-health indicators that move first, and hold every solution accountable to the business KPIs it is meant to improve. Every shipped agent carries a target on both.
Programme health · leading and lagging
Adoption · leading
Weekly active users against eligible seats, per function. The number that predicts everything else.
Time reclaimed · leading
Hours saved per role per week, measured after human rework. The tangible promise and the ROI case.
Agent health · leading
Evaluation score and error rate, tracked continuously. A drop triggers iteration or rollback.
Quality · lagging
Gold-standard conformance on the workflows an agent touches: fewer errors, more consistent outputs.
Speed · lagging
Cycle time on target workflows, from need to response.
Return · lagging
Cost per unit of output and margin. Capacity added without proportional headcount.
Business value · the impact it moves
Time saved is the start, not the finish. Every solution maps to one or more business impact categories, so value is measured where the board feels it, not just in hours reclaimed.
Top line
Revenue captured: fewer stockouts, faster fulfilment, better service levels.
Bottom line
Cost removed: less rework, lower error rates, fewer manual touches.
Working capital
Cash freed: leaner inventory, faster cycle times, fewer write-offs.
Customer experience
Service that shows: accuracy, responsiveness, on-time-in-full.
Resilience
Risk reduced: continuity, compliance, fewer single points of failure.
Sustainability
Footprint improved: waste, energy and materials tracked and cut.
The KPIs an agentic solution can move
A cross-section of the metrics that matter across the main SAP processes, with the sort of performance ranges seen from laggard to top-quartile operators. Filter to your area to see where a solution would earn its keep.
Process area
| Process area | KPI | What it measures | Laggard | Median | Top quartile |
| Supply Chain | OTIF (On-Time In-Full) | Orders delivered in full, on time | <75% | 90% | 95% |
| Supply Chain | Production Schedule Adherence | Plan followed as scheduled | <80% | 85–95% | 95% |
| Supply Chain | Overall Equipment Effectiveness | Availability × performance × quality | <70% | 70–85% | 85% |
| Supply Chain | Scrap Cost Rate | Value lost to scrap | 3% | 1–2% | <1% |
| Process area | KPI | What it measures | Laggard | Median | Top quartile |
| Finance | Cash Conversion Cycle | Days to turn spend back into cash | >80 days | 50–70 | <40 days |
| Finance | Days Sales Outstanding | Days to collect what is owed | >65 days | 45–55 | <35 days |
| Finance | Financial Close Cycle Time | Days to close the books | >12 days | 7–9 | <5 days |
| Finance | Days Payables Outstanding | Days taken to pay suppliers | >45 days | 55–70 | 70 days |
| Process area | KPI | What it measures | Laggard | Median | Top quartile |
| Manufacturing & Inventory | Inventory Accuracy | System vs physical stock | <75% | 90% | 98% |
| Manufacturing & Inventory | Inventory Turnover | Times inventory is used per year | <4x | 6–9x | 10x |
| Manufacturing & Inventory | Backorder Rate | Orders delayed by no stock | 10% | 5–8% | <2% |
| Manufacturing & Inventory | Demand-Supply Match Rate | Supply meeting demand | <75% | 85% | 95% |
| Process area | KPI | What it measures | Laggard | Median | Top quartile |
| Procurement | Spend Under Management | Spend actively controlled | <60% | 75% | >90% |
| Procurement | Savings Realization Rate | Negotiated savings actually banked | <60% | 70–85% | 85% |
| Procurement | Sourcing Cycle Time | Days to run a sourcing event | >25 days | 15 | <10 days |
| Procurement | Contract Cycle Time | Days to execute a contract | >30 days | 20–25 | <15 days |
| Process area | KPI | What it measures | Laggard | Median | Top quartile |
| Sales & Service | Opportunity Win Rate | Opportunities won | <15% | 20–30% | >30% |
| Sales & Service | Sales Cycle Length | Days from lead to close | >60 days | 35–45 | <30 days |
| Sales & Service | Quote Accuracy | Quotes right first time | <85% | 90–95% | >95% |
| Sales & Service | Order Booking Accuracy | Orders booked without error | <90% | 95% | >98% |
| Process area | KPI | What it measures | Laggard | Median | Top quartile |
| People | Time to Fill | Days to fill an open role | >45 days | 30–40 | <25 days |
| People | Offer Acceptance Rate | Offers accepted | <65% | 70–85% | >85% |
| People | First-Year Attrition | New hires leaving within a year | >20% | 10–15% | <10% |
| People | Internal Hire Rate | Roles filled from within | <15% | 20–30% | >30% |
| Process area | KPI | What it measures | Laggard | Median | Top quartile |
| Projects & Assets | Project Gross Margin | Margin delivered on projects | <15% | 20–30% | >30% |
| Projects & Assets | Cost Performance Index | Value earned vs cost spent | <0.9 | 0.9–1.1 | >1.1 |
| Projects & Assets | Schedule Performance Index | Progress vs plan | <0.95 | 0.95–1.05 | >1.05 |
| Projects & Assets | Billable Utilisation | Time that is billable | <70% | 75–85% | >85% |
| Process area | KPI | What it measures | Laggard | Median | Top quartile |
| Sustainability | ESG Data Completeness | ESG data captured and reportable | <80% | 85–95% | >95% |
| Sustainability | GHG Reporting Accuracy | Emissions reported accurately | <80% | 85–95% | >95% |
| Sustainability | Carbon Emission Reduction | Year-on-year emissions cut | <2% | 3–5% | 5% |
| Sustainability | Energy Intensity | Energy used per unit of output | High | Mid | Low |
The rule
Every shipped agent carries a target on both scoreboards: a programme indicator and a business KPI. The ones that do not move their metric are retired at the quarterly reboot. Each solution in the library states the KPIs it is built to improve.
Continuous improvement
The self-healing mechanism.
As the programme matures it accumulates operational memory: KPI history, build logs, integration data, and agent telemetry. An agentic layer monitors that signal, reasons over it, and surfaces structured recommendations, so the programme improves itself rather than waiting for someone to call a meeting.
Signal
Continuous monitoring of KPIs, the build log, evaluation scores and integration telemetry. Sustained deviations are flagged automatically.
Intelligence
Anomalies are cross-referenced with recent changes and history to generate a hypothesis about cause, not just a symptom.
Recommendation
A ready-to-act brief with a named owner and suggested action, not a dump of data. Actions taken feed back so the mechanism learns what works.
A phase 2+ capability
The self-healing mechanism needs a critical mass of operational data and stable integrations to reason over. Design for it from Phase 1; the AI Lead owns it, and the executive sponsor approves any autonomous action it proposes.
The stack
Four layers, one governed path.
Every request follows the same path, whatever the process. Identity, secrets, approval and audit are enforced in one place, on Neptune DXP.
Intelligent apps – authenticated request
Where people work: fit-for-purpose SAP applications on any device, online or offline.
Neptune DXP + Naia – governed MCP / API
Hosts agents, automations and the orchestrator. Authenticates every request, enforces human-approval gates, and logs everything.
Integration layer – reads & writes
The MCP Server and native SAP integration expose SAP capabilities as governed tools.
Systems of record
SAP ECC, S/4HANA and hybrid landscapes, plus connected systems as each use case needs.
The security spine
Governed on every request, by design.
Identity & SSO
Users authenticate with their normal enterprise credentials. Access is controlled by security groups per system, so leavers lose access automatically and agents run within the user’s authorisations.
Secrets in a vault
Credentials live in a managed vault, read by the platform via managed identity. No secrets in code, in config, or on any user machine.
Human approval gates
Read is open; anything that writes to SAP pauses for explicit human approval until an agent has earned a higher autonomy level.
Tracing & audit
Every agent action, tool call and session is logged: who did it, when, and whether it succeeded. Performance and error trends are visible over time.
Clean core
Native SAP integration extends from within and preserves a clean core across ECC, S/4HANA and hybrid landscapes.
Least privilege
Each agent and tool gets the narrowest scope that does the job. Sensitive data stays disconnected unless justified.
Deployment topology
Build, validate, run, with a controlled promotion path.
Isolated environments with their own secrets, identity and connector targets. Only the governance model and infrastructure-as-code are shared, and promotion follows the operating model’s sign-offs.
Development – Build: Where agents and integrations are built
Synthetic or non-production data, connectors pointing at sandbox SAP instances. The AI Lead and engineers iterate here.
Sandbox / UAT – Validate: Prod-like staging and pilot
Masked, prod-like data and UAT endpoints. Champions pilot here, and the Process Lead signs off before production.
Production – Run: The live platform
Live SAP connectors, full governance, high availability. Access for all eligible users via security groups.
Disaster recovery – Resilience: A warm standby
A backup of production in a second region with regular restore and failover testing to meet RPO and RTO targets. Not a place to build.
Platform standards
The consistent bar every solution is built to.
So the platform grows without sprawl, and anything shipped is secure, observable and repeatable. These are the rules every solution in the library follows.
- Read before write.
The first release of any integration is read-only. Write actions are a separate, approved step. - Least privilege.
Each integration gets the narrowest scope that does the job; sensitive data stays disconnected unless justified. - Human-in-the-loop on writes.
The agent proposes, the human confirms, the application executes the SAP write.
- Central, not local.
Integrations run centrally, not on user machines. Onboarding a user is a single group change. - One tool, one operation.
A read-only tool never shares implementation with a write tool. - Observable by default.
Every action is logged and auditable, so you can always answer who did what, and when.
From playbook to build
Each solution’s build blueprint applies these standards to a specific process. See the Goods Receipt blueprint for a worked example of the security model, the write path, and the human-in-the-loop pattern in practice.
Adoption & change
Land AI as leverage, not a threat.
This is mostly done by leadership and Champions, not by technology, and it starts before the first agent ships. Get the message and the momentum right and every later stage gets easier.
Frame augmentation
AI takes the repetitive, low-value work so people spend time on judgement, relationships and creativity. Name “will I be replaced?” openly and early; unspoken, it quietly kills adoption.
Lead through Champions
Front-line power users become peer trainers and the fastest feedback loop. They surface what is broken or missing before it becomes shelfware.
Celebrate the wins
Make early wins visible. Time saved on a real workflow, communicated well, does more for adoption than any mandate.
Managing it day to day
Manage agents like teammates, not projects.
Building AI-native capability is not a one-time implementation. The role shifts to continuous tuning, monitoring and improving, treating each agent like a junior colleague who does the work while a human signs off.
- Daily check-ins.
Review how key agents performed, scan for errors or stuck workflows, and ask teams where AI saved or cost them time. - Feedback loops.
Make it easy to flag a bad answer, collect it weekly, and feed it back into refinement. - An AI satisfaction score.
Track confidence like an internal CSAT, so trend beats anecdote.
- Usage monitoring.
Low usage is an adoption problem to fix in messaging, training or workflow design. High usage is a signal to double down. - Quality spot checks.
Randomly audit generated outputs against brand, compliance and the gold standard. - Retire what does not earn its keep.
Every agent carries a target metric; the ones that do not move it are retired at the quarterly reboot.
The goal
Embed AI so deeply into how work happens on SAP that it becomes part of the organisation’s DNA. That is the autonomous enterprise for SAP, and it is reached one adopted, well-managed workflow at a time.