SAP Digital launchpad

Streamlining SAP's Gen-AI Prompting Experience

Streamlining SAP's Gen-AI Prompting Experience

Streamlining SAP's Gen-AI Prompting Experience

Overview
Making ‘garbage in, garbage out’ yesterday’s AI problem

The quality of output from AI models depends directly on the quality of the input data. For SAP customer success teams generating content like emails, scripts, and pitches, incorporating SAP solution details and prospect information is crucial. This design addresses a key challenge: creating an accessible experience for users of all AI expertise levels to construct effective prompts.

The solution combines system prompts, prompt templates, and seamless SAP data integration - ensures teams can easily incorporate relevant context and generate high-quality, consistent outputs regardless of their AI experience level.

Role
Led UX research and design for this feature, creating mockups and interactive prototypes following SAP Fiori Design System. Collaborated closely with engineering teams throughout the development life cycle.
Team

1 Project Manager, 2 Business Architects, 20 Developers

Timeline

May 2024 - Oct 2024 (Launched in Dec 2024)

Achievement

After launch, we saw a 10x increase in contextual prompting

I collaborated with the data analytics team to monitor user behaviors. The recent usage data demonstrated noticeable advancement in personalized content generation.

10x

contextual prompting

Up from the previous prompt repository design, showing increased awareness of SAP context integration in content generation and improved output quality.

40%

time saved in prompting

The average prompt composition time decreased from 2 minutes to 40 seconds, with users finding it easier to refine outputs during conversations.

solution overview

Three integrated prompt configurations scale from novice to expert users

The prompt configuration system features three layers: System Prompt, User Prompt Template, and Optional Context Integration. These layers progress from simple to complex, ensuring accessibility for beginners while offering flexibility for advanced users.

Beginners benefit from a guided, straightforward approach that maximizes output quality with minimal effort. Expert users can incorporate extensive context data, enabling them to generate the highest quality LLM outputs.

1

1

1

System Prompt

Output Quality:

User Effort:

2

2

2

User Prompt Template

Output Quality:

User Effort:

3

3

3

Optional Context Integration

Output Quality:

User Effort:

Design Highlights

Final designs and solutions

The system starts with system prompts for beginners, adds customizable user prompt templates for intermediate users, and offers advanced data integration for experts. This layered approach ensures each user can access the right level of complexity while maintaining consistent output quality across all SAP customer success teams.

System Prompt

System prompts enable high-quality outputs by simply selecting task types, maintaining consistency while minimizing complexity.

Prompt Template

We offer pre-engineered prompt templates for emails, call scripts, video content, etc. Each template features customizable parameters, letting users create tailored content with just a few clicks.

Context Integration

AI-expert users are able to filter and integrate any SAP context into AI input, further enhancing AI outputs.

context

What is Digital Launchpad?

Digital Launchpad is an AI-powered tool designed to streamline the daily tasks of SAP Customer Success teams. Its standout feature is an AI chat feature that produces SAP-compliant content by incorporating essential context—such as prospect information, company policies, product details, and lines of business—ensuring accuracy and relevance every time.

User Input

Prompts and data integrations

Act as an SAP sales executive who...

LLMs

Processes user queries

AI Output

Text and images based on input

SAP S/4HANA, our intelligent ERP...

How do Digital Launchpad enable users to integrate SAP context in prompts?

Context is essential for generating accurate and relevant AI responses. However, users often struggle to determine what context is needed, how to integrate it, and where to find the necessary information.
To address these challenges, we have developed prompt templates that guide users in incorporating the right context seamlessly. These templates standardize the input process, making it easier to include essential details and ensure that the AI performs tasks effectively.
Prompt Template Repository

Pre-built for customer success activities

Variables in Prompt

Customization LLM outputs with SAP context

problem

Prompt templates lack flexibility

Digital Launchpad serves SAP customer success teams, but data shows low adoption of prompt templates. User feedback revealed two key challenges.

1

1

1

1

Feature Undiscoverability

Prompt repository was not easily found within the chat interface, causing users to struggle in locating and accessing prompt templates. This led to underutilization of templates.

2

2

2

2

Non-Scalable Prompt Templates

Prompt templates failed to cover all user scenarios. Not all use cases could be pre-defined by prompt templates. Meanwhile, expert users found prompt templates too rigid to incorporate additional context.

3

3

3

3

Context Loss During Prompt Editing

When users added parameters to prompts, the editing window blocked their view, resulting in loss of chat context. Users couldn’t reference ongoing conversations while editing, leading to workflow disruptions.

4

4

4

4

Lengthy User Flow

Integrating data required too many steps to add mandatory and optional parameters like SAP product and prospect details. This made it difficult to search for and include additional context in prompts, discouraging users from fully utilizing the tool.

How might we design a discoverable and adaptable prompting experience that maintains context and streamlines parameter integration?

How might we design a discoverable and adaptable prompting experience that maintains context and streamlines parameter integration?

How might we design a discoverable and adaptable prompting experience that maintains context and streamlines parameter integration?

Design proposal

Q1: How might we enable prompt configuration within the user’s current context?

When users open Digital Launchpad, they'll find a variety of ready-to-use templates - from emails to call scripts to video content.

Each template is pre-engineered with parameters. Clicking and customizing the parameters creates content that works for their needs.

Option A

Option A

Option A

Option A

Side panel

A unified view shows users prompt, parameters, and allow users to set up parameter in one panel.

+

Least number of clicks needed to configure parameters

-

Free typing and template inputs are separated

-

Overwhelming for AI beginners

Option B

Option B

Option B

Option B

Conversation area

Prior to starting the conversation, the system helps users input necessary parameters, shaping the context of the discussion.

+

Minimal information required to start

-

Templates don't work across different chat topics

Option C

Option C

Option C

Option C

Text input field

Users can click on the items enclosed in the container box to select one or many from the list - without a need to click anything to reveal options.

+

Matches users' expectations of an AI chat interface

+

Shows prompt structure and parameters before sending

+

Keeps writing focused in one input space

Parameters and interactions

The parameter editing interactions offer tailored interaction patterns based on parameter complexity, editing requirements, and user input needs.

  • Single Selection: Quick pop-over menu with dismiss on outside click.

  • Data Integration: Modal window requiring selection and confirmation.

  • Long Text: Full-screen editor for adding and modifying content.

Q2: How might we streamline the integration of contextual data into prompts?

Users can enhance chat responses by adding context such as prospect details and contextual information.

Option A

Option A

Option A

Option A

Pop-over

Add prospect details and context through a compact pop-over.

+

Space-efficient for limited options

-

Hard to manage longer item lists

-

Easy to dismiss accidentally

Option B

Option B

Option B

Option B

Side panel

Expandable panel for viewing, filtering, and adding chat context.

+

Shows complete content overview

+

Streamlines multiple data entry

-

Interface may overwhelm new users (collapsed by default)

Interactions with side panel components

The side panel integrates data from two SAP sources. Outreach provides prospect profile and context, while MXP enables search for SAP-specific data including customer references and business lines.

Q3: How might we enhance the flexibility of prompt customization to accommodate diverse user needs?

Separating prompts into system and user components helps new users access key LLM features while maintaining quality and compliance standards.

Option A

Option A

Option A

Option A

Independent panel

Displayed in a separate editing panel.

+

Offers full system prompt visibility

+

Enables expert prompt customization

-

System prompt and user prompt feel disconnected

Option B

Option B

Option B

Option B

Attach to text input field

Embedded within the text input area, with options to switch or remove.

+

Intuitive for users of all levels

+

Seamlessly combines with user prompt templates

+

Makes system prompt switching effortless

Integration of System Prompts and User Prompt

The design enables flexible prompt management through a set of key interactions. Users can add system prompts to existing conversations, remove them when no longer needed, combine multiple system prompts in a single chat, and access user prompt templates as needed.

OUTCOMES & ITERATION

After launch, we saw a 10x increase in contextual prompting

Looking at 2 month of data -

I collaborated with the data analytics team to monitor user behaviors. The recent usage data demonstrated noticeable advancement in personalized content generation.

10x

contextual prompting

Up from the previous prompt repository design, showing increased awareness of SAP context integration in content generation and improved output quality.

40%

time saved in prompting

The average prompt composition time decreased from 2 minutes to 40 seconds, with users finding it easier to refine outputs during conversations.

Some kind words from users

Real experiences from SAP team members highlight how the new prompt system transforms their daily workflows, from simplifying basic tasks to enabling advanced customization.

"This is exactly what I needed! I don't have to overthink prompt writing anymore. Just pick a template, tweak a few things, and boom - my customer emails are done in seconds."

— Marketing Manager, 6 months with AI

"What a relief! No more copy-pasting from different systems. Everything I need about products and prospects is right there in the side panel. And switching between prompts? Super smooth."

— Senior Account Executive, 2 years with AI

"Being new to all this AI stuff, I was nervous at first. But wow - the basic templates are perfect for getting started, and I'm slowly exploring more features as I get comfortable."

— Junior Sales Rep, 2 months with AI

Learnings + Reflections

Lessons from an agile enterprise product team

A simple user experience is crucial even for business users

Working on Digital Launchpad challenged my assumptions about enterprise software complexity. While business tools often involve complex data integrations and workflows, I learned that simplicity is crucial for user success. It allows users to complete tasks efficiently, reduces frustration, improves productivity, and ultimately leads to better adoption and satisfaction with business tools, even if they are complex in nature; essentially, a clean and intuitive interface can make even intricate business functions easier to manage. 

Design Systems bridge the gap between design and engineering

The design system serves as more than a toolkit—it's an empowering platform enabling consistent, accessible, and visually coherent applications. Working with SAP Fiori Design System and UI5 library demonstrated this value firsthand. By sourcing UI5 components and studying their variables during design, I created a seamless bridge between design and development teams and ensured the designs were developed perfectly.

@2024 Jayce Kong — Thanks for visiting :)

kwngzy@gmail.com

@2024 Jayce Kong — Thanks for visiting :)

kwngzy@gmail.com

@2024 Jayce Kong — Thanks for visiting :)

kwngzy@gmail.com

@2024 Jayce Kong — Thanks for visiting :)

kwngzy@gmail.com