Intelligent and Automated Billing Workflows in SAP S/4HANA Public Cloud for Professional Services Automation — using AI (SAP Joule) and a rule-based framework to shift billing specialists from reactive firefighting to proactive, guided decisions.
Research Approach
Exploratory design research focusing on experience and workflow automation using a mixed-method UX approach across four iterative phases.
Context
Project Billing is the process of reviewing, validating, and preparing project-related billing elements for customer billing based on contractual agreements, recorded time, expenses, and project activities.
"My job is to bill as much as possible accurately, timely regarding the project situation."
Problem Space
Discovered through day-in-the-life sessions, user interviews, Customer Councils, Qualtrics surveys, and cloud reporting data.
Billing Specialists struggle to determine what to work on first. High-value and overdue items are not clearly highlighted, leading to inefficient task prioritisation and delayed billing.
The billing process requires navigating multiple applications and business objects without a clear starting point. This fragmentation leads to confusion, redundant steps, and higher effort.
Critical issues such as invoice splitting and billing errors are often discovered only at very late stages, forcing users to cancel submissions and repeat the entire process.
Key figures, columns, warnings, and aggregated numbers in billing apps are not clearly explained. Users do not understand how values are calculated or what actions are required next.
Users repeatedly perform the same actions and data inputs across different steps and apps, increasing effort and frustration while providing little added value.
Billing Specialists lack early visibility into billing readiness, including the inability to preview invoices during PBR creation. Approved time and expenses can remain unbilled for long periods.
Workflow Design
The core design thesis: surfacing billing issues at the data-review stage eliminates the costly rejection-rework loop between specialist and clerk, reducing BDR rejection rates and manual back-and-forth.
Key Insight: In the current workflow, issues surface only when the Billing Clerk rejects the BDR — triggering a costly rework cycle. Joule moves issue detection to the data review step, eliminating the rejection loop and lowering BDR rejection rates.
Design Concept
SAP Joule is the AI-powered copilot embedded in SAP applications. It provides contextual insights, prioritised situation cards, and guided actions, directly within the user's workflow, without forcing them to switch between multiple apps.
The design concept uses Joule's Situation Handling framework to proactively surface billing risks before they become costly errors, ranked by urgency and financial impact.
Prototype Demo
A walkthrough of the high-fidelity Joule prototype — from the situation card on the S/4HANA launchpad through risk analysis and billing request creation.
Evaluation Results
7 moderated usability sessions + 21 Customer Council responses using UEQ-S, NASA-TLX, PSAT, and UX Lite instruments.
Both systems scored Excellent. Slight decrease with Joule, users need more practice to fully leverage AI-guided actions.
Biggest gain (+1.14). Joule's conversational, guided experience is perceived as significantly more engaging and stimulating.
Overall UX improved by +0.48. Both systems rated Excellent, Joule's score is driven primarily by its superior hedonic quality.
Key Insight: Hedonic quality gains outpaced pragmatic gains, Joule's improvement is primarily experiential. Users find the AI-driven interface more enjoyable and engaging, even as task-completion efficiency continues to mature.
Pattern: faster triage, but more thinking/verification.
Joule scored higher on ease-of-use and satisfaction; current system led on "does what I need" due to missing billing summary and unclear actions.
"Billing is too complex, guiding & prompting potential issues sooner is better"
"This solution would give the chance to distribute responsibility in the future"
Key Findings
Six findings distilled from qualitative and quantitative analysis across all study streams.
Situation cards reduce searching and highlight what matters first. Users clicked directly on high-priority prompts and valued quick visibility into overdue items and key risks.
NASA-TLX pace/temporal demand dropped by 5.7 points. Users reported feeling less rushed during the triage phase of billing preparation.
More intuitive onboarding. Joule scored higher on ease-of-use. Complex billing cases and newer users benefited the most from guided guidance.
Users still verify AI suggestions in the app. Mental demand rose by 5.7 points. The concept currently serves as an assistive layer, not a replacement for expert verification.
Missing billing summary before submit, unclear "Send email" actions, and lack of deep links to exact items reduce perceived functional sufficiency and trustworthiness.
Joule augments the workflow rather than replacing it. For clean cases, the current system is faster. AI delivers more benefit for complex billing or less-experienced users.
Synthesis
Hypotheses Validation
Three hypotheses were tested across the within-subjects usability study.
Discovery improved; resolution and submission are gated by trust and action clarity. Users found risks faster but needed app verification before committing to billing decisions.
Cognitive load reduced for issue identification. Adding billing summaries and progressive disclosure would fully realise this hypothesis. Mental demand increased slightly due to verification.
Users found billing risks faster and with greater clarity using Joule situation cards. Full confidence to submit without verification still requires improved transparency and data source explainability.
Iteration 2
A second-iteration prototype addressing the key trust, clarity, and actionability gaps surfaced in the usability study. Each improvement directly targets a validated finding from evaluation data.
Joule now explains how it fetched potential billing issue information — surfacing data sources and reasoning inline. A Billing Summary is presented before the Billing Document Request is submitted, giving specialists a consolidated view of totals, groupings, and expected invoice structure.
Action buttons are now placed directly within the expanded issue card, adjacent to the issue description. This eliminates the need to navigate away or search for actions, reducing context-switching and the number of steps to resolution.
Joule entry points are clearer and more consistently labelled throughout the workflow. Users no longer need to discover how to invoke Joule — triggers are visible, contextual, and use plain-language labels aligned with the billing task at hand.
Notes and comments now display the team member or consultant name alongside the content. This restores accountability and traceability — specialists can immediately identify who flagged an issue or added context without leaving Joule.
Joule offers a persistent "View in App" option at any point during the workflow, preserving specialist control and trust. Users can drop into the S/4HANA app to verify data or complete edge-case actions, then return to the Joule flow — without losing context.
Design rationale: Each iteration targets the three root causes identified in the evaluation — lack of trust in AI-generated data, incomplete in-Joule resolution paths, and unclear workflow ownership. The refinements collectively reduce the need for app verification while preserving full specialist control via human-in-the-loop affordances.
Limitations & Dependencies
Results indicate clear direction and value but require validation in a real, integrated environment with larger samples.
Findings are indicative, not statistically generalizable. Larger samples with diverse experience levels are needed.
The Joule prototype was not fully integrated with live S/4HANA apps. Real integration may surface additional usability gaps.
No longitudinal trust or learning effects could be measured. Trust in AI typically builds over weeks of repeated use.
Focus was limited to Project Billing (not the full E2E sales billing chain), which constrains generalizability.
Task efficiency was inferred from self-reported data and think-aloud sessions. Time-on-task logging was not captured.
Varied workflows, billing models, contract types, data quality, and authorization constraints across organisations.
Future Scope
Five strategic directions that build on this thesis toward a fully autonomous, intelligent ERP billing experience.
PSA Team Impact
Beyond the research findings, this thesis delivered concrete, strategic value to the SAP Professional Services & Billing product team, across discovery, design, and product direction.
Provided a clear blueprint for transforming Project Billing so AI can anticipate billing discrepancies and unbilled revenue before they become critical issues, not after.
A direct comparison between the existing solution and the AI-enhanced prototype provided data-backed proof of improvements in billing accuracy, efficiency, and reduced manual effort.
Demonstrated how to build a more intuitive, guided billing process, significantly lowering the learning curve and reducing human error, especially for users new to the system.
Delivered crucial insights into real user needs and pain points, giving the product team greater confidence to make informed roadmap decisions and justify an AI-first strategy.
By intelligently guiding new users through complex billing processes, the solution helps employees become productive faster, reduces training overhead, and fosters greater adoption.
Conclusion