Making conversational AI buildable without code

Summary

As interfaces moved from graphical to mobile and on to conversational, SAP wanted enterprises to keep pace. Built on AI and ML, SAP CoPilot became a cloud-based digital assistant that powered conversational experiences inside S/4HANA, SuccessFactors, Hybris, Ariba, and Concur, and inside communication tools like Slack, Microsoft Teams, and GSuite.

The Skill Builder was the code-free platform behind it, built for enterprise readiness so people across roles, not only developers, could create, configure, and test custom skills without writing a line of code.

SAP CoPilot Skill Builder's Manage Sales Order screen, showing intent creation with business objects, actions, and parameters defined for a sales order

What the Skill Builder did

A no-code platform for conversational UX, built from reusable enterprise patterns:

Code-free creation and management of skills and intents, so business users and designers could contribute without depending on developers.
Pattern-based design that reused enterprise workflows, such as approvals, status tracking, and lookup services, to build reusable skills.
Secure integration with backend systems such as S/4HANA, keeping access control, role-based authorization, and data consistency across applications.

SAP CoPilot running as a conversational interface across a laptop, tablet, and phone, with logos for SAP, SuccessFactors, Ariba, and Concur and integrations with Slack, Microsoft Teams, and GSuite — SAP CoPilot's conversational interfaces across devices and integrations.

Key personas and their tasks

Configurator / admin

Configured skills for SAP CoPilot to use.

Project manager

Defined business goals, use cases, services, and utterances.

UX designer

Created dialogue flows, mapped task flows, and translated user needs into conversation logic.

Developer

Evaluated technical feasibility, and tested and published skills.

Linguist

Defined linguistic rules, synonyms, and dictionaries, and trained the NLP models.

Our UX approach

Set expectations

Held stakeholder alignment meetings to establish design goals, timeline, deliverables, and ownership across teams.

Project plan timeline for release 1811 to 1902, showing usability test and design iteration phases with new features and testing fixes listed for each phase — Release plan: usability testing and design iteration phases.

Conduct workshops

Ran cross-functional workshops to understand technical constraints, map existing workflows, and identify key user needs across personas.

Design workshop with five team members in a meeting room, one presenting at a whiteboard covered in sticky notes next to a screen with early mockups

Create a hero story

Crafted a central narrative and user scenario that showed how the Skill Builder would solve real user problems end to end.

Wall-mounted journey-mapping board organized by persona rows (configurator, product manager, UX designer, developer, linguist) and phase columns (identify need, plan, design, create, test), covered in yellow sticky notes

Idea generation

Used collaborative idea cards to explore conversational UI use cases and map features back to user goals.

Three handwritten idea cards, Skill Validation, Start From The End, and Skill Simulator, each with an idea summary and a rough sketch

Define the user journey

Created journey maps to show the end-to-end experience of each persona, including triggers, tasks, tools, and pain points.

Skill building process diagram running from identifying a need, through planning, designing, building, and testing a skill in the Skill Builder, to publishing and monitoring it in the Skill Store

Map the information architecture

Structured the platform into logical modules and flows for scalable use, informed by user interviews and system capabilities.

Information architecture diagram of the Skill Builder platform, branching from the landing page into Learn, Try, and Build sections, then through skills, skill definitions, intents, entities, and response configuration

Validate with lo-fi wireframes

Designed quick, testable wireframes to validate dialog flow, task breakdown, and usability with stakeholders and potential users.

Four colleagues standing together reviewing wireframes and sketches posted on a whiteboard during a validation session

Design for development

Delivered high-fidelity mockups with component specs, interaction guidelines, and developer annotations to keep implementation accurate.

Five high-fidelity Skill Builder screens laid out in sequence for basic configuration, UI display, utterance, and entities, connected by arrows showing the build flow — Specs from basic configuration through UI display, utterances, and entities, annotated for engineering handoff.

Key screens

Query Product configuration screen listing the filter parameters selected for a Product query intent, category, product, supplier, and name, with mandatory parameters and co-reference options below — Product query configuration: defining parameters and entities for conversational interactions.

Manage Sales Order screen for building an intent, with a sample utterance broken into action, business object, and parameters, and a list of business objects, actions, and parameters to enable for Sales Order — Intent creation: connecting conversational skills to backend business objects and services.

Outcomes and impact

The collaboration between distributed teams earned praise from stakeholders for how seamlessly it came together. The Skill Builder's usability drove a clear upward trend in SAP CoPilot adoption, and let business users build custom assistants of their own.

By the end of Q4 2018:

Skills created using SAP CoPilot Skill Builder: 9K+
Daily active users: 1.3K+
Unique tenants, mostly across S/4HANA and SuccessFactors: 160+
User engagement from skill-based interactions: 87%

Challenges and constraints

Distributed team coordination

The project team was spread across multiple time zones and locations, which took precise coordination to make the most of limited overlapping hours and added complexity to workshops, design reviews, and daily collaboration.

Technical complexity versus usability

Designing for non-technical users in a highly technical enterprise environment meant striking the right balance between powerful, flexible functionality and an intuitive, approachable experience.

Enterprise security compliance

Every feature and workflow had to comply with SAP's strict security standards and integrate cleanly with backend enterprise systems, which added complexity to the design process.

Conversational UI paradigm

Building a no-code platform for conversational experiences meant rethinking traditional UI patterns and developing new mental models for users unfamiliar with conversation design.

Reflections and learnings

Distributed teams as strength

It was amazing to see the entire team come together across geographies. The distributed nature of the team turned into a strength, letting us make overnight progress through staggered work cycles and accelerate idea exploration and feasibility checks.

Research-driven conversations

Research and continuous user testing mattered more here than on most projects. With no visual UI to lean on, we had to rely on understanding user mental models, working environments, and typical tasks to get the right intents and flows.

Validation through usage

The project confirmed that enabling non-technical users to create meaningful conversational experiences was not only possible but scalable. Prioritizing outcome-focused flows and cutting technical barriers drove a sharp rise in skill-based interactions, 87% of usage, a clear validation of the design direction.

Guided flexibility

Thoughtful, guided workflows paired with enterprise-grade flexibility unlocked innovation, even for a complex domain like conversational UI in business systems.