Lener Pacania1
ServiceNow Employee
ServiceNow Employee

Since our release of ServiceNow’s Generative AI solution, Now Assist, more and more of the customers I talk to are interested in applying AI to drive Intelligent Automation in the platform. Intelligent Automation combines the automated decision making of AI with automation and integration to automate the remediation process- capabilities that exist in ServiceNow.  

 

In this article I’ll first dive into the current state of AI at ServiceNow, provide examples of combining AI to solve new problems, then I’ll end with an example of where ServiceNow is going with Intelligent Automation.

One common question I hear is if Now Assist (Generative AI) will replace Predictive/Task Intelligence.   The short answer for the next few years is no.  The real interesting question is “how can I combine Predictive/Task Intelligence and Generative AI to create better AI solutions.”

 

AI in ServiceNow is broken out into two areas: Classic AI and the newer Generative AI (Fig 1). Classic AI was introduced into the ServiceNow platform with the Kingston release in 2017 as Predictive Intelligence.  Classic AI is used for categorization, pattern detection, and NLU used in Virtual Agent.  Generative AI was released in May 2023, and is the newest AI addition to ServiceNow’s platform.  Powering ServiceNow’s Generative AI is the Now LLM Service which uses a combination of LLMs built by ServiceNow, Commericial LLMs, and Open Source LLMs.  We remove the guess work and select the optimal LLM for the problem you are trying to solve.

 

1.jpg

(Fig1 - ServiceNow AI categories)

 

You can utilize Classic AI in ServiceNow using Predictive Intelligence, Predictive Intelligence Workbench, and Task Intelligence.  The three core algorithms are classification, similarity, and clustering (Fig 2).  Note: we are re-writing the regression framework in Washington (Safe Harbor).

 

2.jpg

(Fig 2 – classic AI algorithms)

 

Below are some examples of the out of the box use cases addressed by Predictive/Task Intelligence (fig 3a).  Popular use cases (fig 3b) are predicting the assignment group, suggesting solutions, predicting major incident, and identify knowledge gaps.

 

3.jpg

(Fig 3a- Classic AI Use Cases in green delivered with PRO SKU)

 

classicall.jpeg

(Fig 3b - Example Predictive/Task Intelligence OOTB models)

 

Customers can leverage Generative AI in the platform through Now Assist.  Now Assist is ServiceNow’s Generative AI powered assistant with skills that help the user, agent, developer, and administrator do their work better. (Fig 4a).  

 

skills.jpeg

(Fig 4a - Now Assist skills enhance these areas)

 

We add skills on a quarterly basis to help the requestor, agent, and developer/admin in your organization (fig4b).

justskills.jpeg

(Fig 4b - Example Now Assist skills as of July 2024 - does not include ITOM, SPM, and other products)

 

These Generative AI skills cut across all ServiceNow workflows (ITSM, HRSD, CSM, ITOM, SPM, Creator, Industry) and help users get answers faster, improve agent efficiency, enhance chatbot and search, and turbo charges developer productivity.  We are adding new Now Assist skills each month.  Now Assist uses the Now LLM service that runs in ServiceNow’s trusted and secure data centers. For more details see our model cards on how we built, trained, and tested our models.

 

ServiceNow is transforming how customers use Generative AI. In the below picture (Fig 5) on the left, Enterprise AI and DIY Generative AI require a significant investment in time and resources to move a Generative AI solution into production.   On the right, ServiceNow engineers built Now Assist to be a turnkey solution, enabling customers to turn on all the Generative AI capabilities with a click of a button eliminating the need for implementation resources (Fig 5 - right side of image).

transform.jpeg

(Fig 5 – Now Assist is a turnkey solution)

 

Another big change is that Now Assist in VA fundamentally changes how we implement chatbots. If your incident/case can be answered with a catalog item or KB, the Now LLM will dynamically generate the Virtual Agent chatbot conversation – zero development needed.  For those chatbot conversations that can’t be answered with a KB or catalog item, we’ve simplified the Virtual Agent development experience.  The Now LLM can be used for topic discovery instead of NLU, which eliminates the need to create intents and utterances (Fig 6).  See Victor Chen's post to learn more.

 

6.jpg

(Fig 6 - LLM based topic discovery)

 

I’ve not scratched the surface of everything that Now Assist can do in the platform. Stay tuned to the different ServiceNow community hubs for updates on how Now Assist is improving it Virtual Agent, ITOM, SPM, CSM, HRSD, Creator, and Platform Analytics.

 

Combining Classic AI and Generative AI

What does it look like when we bring together Classic AI and Generative AI capabilities? Let’s start with an agent productivity example (Fig 7). In the CSM/FSM Workspace we use Classic AI to route work to the most effective resource by predicting assignment group, priority, and category by using your historical data (step 1).  The agent can then use Now Assist to summarize the case - cutting read time (step 2).  The agent can quickly resolve the incident/case by using Now Assist to reduce write time, by generating a proposed solution (step 3).

 

7.jpg

(Fig 7 – Classic AI + Gen AI CSM Workspace)

 

Generate a KB from an incident/case (Fig 8 ) is an example of the differentiation of Now Assist vs DIY Generative AI. Now Assist knows that if a task is in resolve state to enable the Generate KB Skill.  Now Assist reads through the ticket/case, creates a ServiceNow knowledge template and generates the title, body, and resolution steps in the knowledge article and then publishes it into the Knowledge Management workflow for an approver to review and publish.

 

8.jpg

(Fig 8 – Generate KB from incident/case)

 

Continuing with the theme of agent productivity, we can use ServiceNow’s Generative AI Controller to build solutions unique to your organization. The Generative AI controller is used in Flow Designer and VA Designer to connect to 3rd party LLM’s and Predictive Intelligence, providing the ability to embed AI and Generative AI into your workflows.

 

The below example in Fig 9 (courtesy of my peer- Matt Train) uses Predictive Intelligence to identify the top three similar resolved incidents to the open incident.  We ask the LLM to find a common solution across the three similar resolved incidents.

 

9.jpg

(Fig 9 - Generate potential solution by combining classic and generative AI)

 

The solution in (Fig 9) is made possible because in Flow & VA designer you can access Predictive Intelligence models, Gen AI LLMs, and integration spokes (Fig 10).  Step 1 is to call the similarity model to pull together similar resolved incidents, then in Step 2 we cycle through each similar incident and pass the work notes into the LLM with a prompt asking to find a common solution across the similar resolved incidents. Step 3, we write the generated solution into the resolution notes of the incident.

 

10.jpg

(Fig 10 - Flow Designer allows you mix Classic AI and Generative AI into your workflow)

There are numerous more examples of combining Predictive Intelligence with Generative AI that I don’t have the space to list here, suffice to say the ability to combine Predictive Intelligence + Generative AI to automate decision making can have a tremendous impact on the productivity of your employees, agents, and platform administrators.

 

 The near future.   So, let’s get back to how I started this post - Intelligent Automation.  All future statements fall under safe harbor.  Those who attended the keynote at Knowledge 23 saw a glimpse of how we will make Intelligent Automation a reality.   We have the unique ability to use AI to make a decision and drive resolution without a human getting involved, here are a couple of examples in story form.

 

Imagine an employee tells Now Assist they have a problem with their web conferencing Meet X software.  Now Assist dynamic generates an empathetic response and ask the employee to describe the problem.   The employee says their video is pixelated and laggy.  Now Assist ask the user to upload an image, Now Assist uses the Now LLM to confirm the image is indeed pixelated a laggy (Fig 11).  Now Assist asks for permission to trouble shoot the employee’s laptop and uses ITOM agent client collector runs a series of diagnostics.  Now Assist finds no issues with the laptop, so it runs a parallel check on all resolved incidents.  While doing this Now Assist reasons that it may be worthwhile to check if they employee is due for a new webcam, which indeed they do and Now Assist executes a procurement flow to order and deliver a new webcam to the employee's address.  Now Assist also is done with its analysis and has found that an upgrade of the MeetX software should resolve the pixelation of the video and automatically pushes an upgrade. 

 

11.jpg

(Fig 11 – Intelligent Automation)

 

From an IT operations perspective, Meet X cloud service is sending events about dropped connections to ServiceNow. In Service Operations workspace the agent can see that classic AI has looked through thousands of events and has detected which event is impacting the Meet X service and creates an actionable incident.  Now Assist uses Generative AI to generate a probable root cause by looking at similar incidents, configuration items, changes, and problems and ascertains that a recent Zscalar upgrade is causing the poor MeetX performance. Pretty exciting stuff!

 

To learn more about the latest in Intelligent Automation from ServiceNow reach out to your account team and plan on join us at Knowledge 24 in May to learn more about ServiceNow’s AI and Intelligence Automation Strategy.

 

-Lener

 

 

 

 

 

Comments
sreekanthvy
Tera Contributor

Thank you Lener, for this article, it demystifies the capabilities with NLU powered AI vs the Now Assist GenAI features clearly.

 

This is massive, we are at a stage where we want to enable Now Assist capabilities to our client, we are planning to have demo to be built. Can you please suggest if a similar capability can be built in our partner instance (we have enabled the GenAI entitlements now), if there is a step-by-step demo that we could follow and build it for our client that showcases VA and Flow Designer using the default data that would be great.

 

Thanks,

Sreekanth

PoojaRaghav16
Tera Contributor

Hi @Lener Pacania1 ,

How does the topic 'Similar Resolved Incidents' in the new skill 'Incident Assist' identifies those incidents?

Does it leverage PI or any other specific mechanism?

Also, could you kindly point me to the relevant configuration settings or documentation?


Regards,

Pooja Raghav 

Version history
Last update:
‎07-11-2024 07:58 PM
Updated by:
Contributors