Evaluating the Power of Generative AI: A Side-by-Side Comparison of ServiceNow Now Assist and Salesforce Agentforce

‍

Generative AI has emerged as a pivotal technology for modern enterprises, particularly in customer service and operations. Two leading platforms, Salesforce Agentforce and ServiceNow Now Assist, offer generative AI capabilities tailored for enterprise needs. This whitepaper explores these solutions across the following five below dimensions.

• Data Security
• Use Cases
• Model Support
• Out-of-the-Box Agents
• Integration with Core Platform

‍

Data Security: Protecting Sensitive Information Salesforce:

Comprehensive Security with Einstein Trust Layer Salesforce places data security at the core of its generative AI ecosystem through its Einstein Trust Layer.

• Zero-Data Retention Policy: Ensures data isn’t retained, shared, or used for LLM training by third-party providers like OpenAI and Azure OpenAI.
• Dynamic Grounding: Merges real-time Salesforce data with prompts while preserving role-based access controls to ensure only authorized data is accessed.
• Data Masking: Detects and masks sensitive information, such as personally identifiable information (PII), across multiple regions and languages.
• Prompt Defense: Enforces strict system policies to reduce hallucinations and unintended outputs.
• Toxicity Scoring: Monitors and scores content for potential harm, storing results in Data Cloud for audits and compliance.
• Auditing: Provides detailed logs of prompts, responses, and trust signals, along with pre-built dashboards for analysis. These measures create an environment where enterprises can confidently use AI without compromising on security or compliance.

‍

ServiceNow’s Generative AI: Secure, Private, and Scalable Solutions

• Data Security: Data is securely transmitted using Transport Layer Security (TLS) 1.2, ensuring it’s encrypted during transmission. Once processed, any data used for AI predictions is deleted from the compute hubs, preventing retention of sensitive information.
• No Commingling for Domain-Specific Instances: For domain-separated instances (a method of partitioning data), data from one instance is kept separate from others. This ensures no cross-contamination of data between different customers or use cases.
• Opting Out: Users have the option to mask sensitive data or opt out of sharing their data for model improvement purposes, giving more control over data privacy.
• Third-Party Integration: ServiceNow may use third-party services (like Azure OpenAI Service) to augment the AI capabilities. However, data processed through these third-party endpoints remains secure and is not accessible by the third-party providers themselves. The third-party services are hosted within the ServiceNow network boundary, which maintains strict access controls.
• Capacity Bursting: In cases of high demand, ServiceNow can use Azure-hosted GPUs for capacity bursting, but again, the data is processed securely and without compromising customer privacy.

Use Cases: Generative AI and Beyond

Salesforce: Enabling Generative and Autonomous AI

Salesforce’s Agentforce empowers businesses with a blend of generative and autonomous AI capabilities, going beyond simple text summarization or recommendations:

• Generative AI: Summarizes interactions, generates tailored responses, and creates recommendations for agents in real-time.
• Autonomous AI:
  • Topics and Actions: Identifies conversation topics dynamically and triggers relevant actions based on pre-configured rules.
  • Reasoning Engine: Mimics human-level decision-making to perform complex, multi-step workflows autonomously, such as order processing or claim validation.
  • Task Automation: Handles repetitive processes, allowing agents to focus on higher-value interactions. This holistic approach makes Agentforce suitable for comprehensive customer service and operational excellence.

ServiceNow’s Now Assist concentrates on generative AI functionalities , such as:

• Summarizing knowledge articles.
• Automating playbook generation workflows.
• Assisting developers with code generation and workflow scripting.
• Now Assist panel: Allows agents to ask questions and request summaries using natural language.

Now Assist Admin: Provides a console for admins to set up, configure, and monitor Now Assist applications and features.

ServiceNow Task Intelligence leverages machine learning to improve operational efficiency by automating key aspects of task management, such as:

• Task Creation: Automatically generating tasks based on incoming requests or events.
• Task Triage: Prioritizing and categorizing tasks to ensure they are routed to the right resources or teams.
• Task Investigation: Using AI to assist with troubleshooting and resolving issues more efficiently.

Task Intelligence currently is only available for ITSM and CSM modules.

Key Comparison:

Currently ,ServiceNow’s task intelligence is not integrated with Now Assist features or Now Assist Panel. While these capabilities are impactful, ServiceNow Now Assist lacks the autonomous decision-making and task execution features currently. However it allows creating custom Now Assist skills using NASK and adding the autonomous behavior to some extent using UI Actions.

Salesforce AgentForce  provides generative AI and extensive automation across their customer service and operational workflows with its prebuilt agents, topics and actions.

‍

AR.BOT Use Case: Intelligent Task Automation with Now Assist

In this use case, AR.BOT leverages ServiceNow’s Now Assist capabilities to streamline and automate workflows. The core functionality revolves around summarizing interactions, analyzing sentiment, and suggesting the best next actions based on AI-driven insights.

‍

Key Features and Workflow:

• Summary Generation with Now Assist:
  • Using Now Assist's Generative AI, one can automatically generate a summary of an account based on details like Total Account Receivables, Due Account Receivables, Aging buckets etc. This summary helps contextualize the situation, saving time and ensuring that important details are captured for further analysis.
• Sentiment Analysis:
  • Sentiment analysis is used to assess the emotional tone of the communication (e.g., whether a customer email is positive, negative, or neutral). This analysis is integrated with the Task Intelligence models, which evaluate the sentiment of emails, tickets, or customer interactions.
  • Based on the sentiment analysis, the system can trigger specific actions, such as prioritizing certain tasks or escalating negative feedback.
• Next Action Suggestions:
  • After analyzing the generated summary and sentiment of a received email , focusing on risk factors (e.g., high overdue balances, old invoices) and suggesting action to the collector based on the aging and overdue amounts.If there is an email from a customer regarding some discrepancy in the transactions related to the account, the system might suggest creating a dispute or escalating the issue for further investigation.
  • These suggestions are informed by the AI's understanding of past cases and historical data, which helps reduce human error and ensures a more efficient response.
  • Task Intelligence can analyze patterns across multiple interactions , account summary and suggest changes at the account level. For example, if the system could recommend changing the account’s handling strategy, escalating the account to a higher-tier team.
• Automated Email Generation:
  • Based on the sentiment and the nature of the received email, the system can generate contextual emails to the relevant parties (such as an account owner or customer). For instance, if a customer is requesting a statement from an account , the system will generate and send a statement as a reply without the user needing to stitch it together.
  • If the sentiment is positive, the system might generate a follow-up email offering additional products or thanking the customer for their engagement.
• AR Copilot

‍

The AR.BOT App includes an AR Copilot feature, which empowers agents to request on-demand data analytics and insights using natural language queries. Agents can ask business questions, such as "What is the total overdue amount for Customer X?" or "How many promises to pay are due next week?" The AI system processes these questions and provides detailed answers, eliminating the need for manual searches and analysis. This copilot acts as a virtual assistant, improving decision-making and providing valuable insights to the agents in real-time. In future this co-pilot will also be able to carry out instructions in natural language such as send email reminders to a customer, create a dispute or a ticket based on the context of the current session and many other actions that can be taken via a regular user interface.

• Processing Payment Remittance: Using Document Intelligence, the system can automatically read and process remittance details for cash applications. The system can handle various formats, including:
  • Scanned Images: Extracting text from scanned images using Optical Character Recognition (OCR).
  • PDF Documents: Parsing structured and unstructured data from PDF files.
  • Handwritten Copies: Recognizing and interpreting handwritten information using advanced handwriting recognition.
  • Free form text:  Understanding the free form text written in emails, ACH/Wire instructions and documents to and action accordingly.

‍

Model Support: Flexibility and Adaptability:

Salesforce: Supports an extensive range of third-party and proprietary models to cater to varied enterprise needs:

• Third-Party Models:
  • Amazon Bedrock
  • Azure OpenAI
  • OpenAI
  • Google Vertex AI
• Proprietary Models:
  • CodeGen: Tailored for developer productivity.
  • xGen-Code: Salesforce’s in-house LLM for code generation and assistance.

This diversity allows businesses to choose models based on specific use cases, such as natural language processing or developer support.

‍

ServiceNow: In-House Specialization with BYOLLM Support ServiceNow leverages in-house models designed for targeted tasks:

• Mixtral-8*7B: Text summarization.
• StarCoder: Developer code generation.
• Text-to-Flow Models: Workflow automation.

Additionally, its Bring Your Own LLM (BYOLLM) capability allows integration with third-party models, including OpenAI and Azure OpenAI.

‍

Out-of-the-Box Agents:

Salesforce:  Salesforce’s AgentForce provides configurable out-of-the-box agents designed for dynamic scenarios:

• Agents dynamically adapt to topics and execute multiple predefined actions.
• Deep integration with Salesforce’s Flow and Apex Class enables robust workflow automation.

Some available out -of-the-box agents present today are Service Agent , Sales Development Representative Agent, Sales Coach, Personal Shopper Agent, Campaign Agent etc.

ServiceNow: Skills-Based AI Agents

ServiceNow’s Now Assist Skills focus on enhancing specific workflows like Technology, Employee, Customer, Creator, Platform etc. While effective, these skills lack the adaptability and autonomy offered by Salesforce’s agents.

We can combine Now Assist skills and Task Intelligence to provide an autonomous AI capability.

‍

Integration with Core Platform:

Salesforce: Workflow-Centric Integration integrates deeply with its ecosystem for seamless automation.

Actions can invoke Flows, Apex Classes, and Prompt Templates, allowing seamless task execution within the CRM environment.

ServiceNow: Script-Driven Enhancement

ServiceNow provides the ability to enhance prompts through client-side scripting, allowing you to refine how inputs are handled. Additionally, you can write server-side code or integrate with Flow Designer to add dynamic inputs to these prompts. Currently, Now Assist skills can only be deployed through UI  Actions or the Now Assist panel. However, custom components can be developed to call these skills and automate actions based on the responses received.

‍