Looking for real-time benefits and use cases of Predictive Intelligence

[Suggy]

Hi All,

We are looking to implement Predictive intelligence(PI). Wanted to know how you are utilizing this today, like

- For which tables are you using this (only for INC or others?)

- Which fields are you populating with PI (like Assignment group or Service or Priority?)

Any other use cases you are using, please share.

Basically looking of PI has helped you in reality. Thanks in advance 🙂 

[Subrahmanyam2]

Hi Suggy,

We have recently implemented predictive intelligence in one our instance.
And here are some of the use cases and the implementation approach we took for your reference:

1) Classification Framework: Predicting Configuration Item based on Short Description field (My recent implementation explained below)

In our instance every configuration item has an associated support group or assignment group. Based on existing rules, if someone is creating an incident and fills the Configuration Item, the assignment group auto-populates on form. Since the logic to auto-populate assignment group is already there we do not need to predict assignment group. But since the short description field gives some context on the issue, we thought like we would use predictive intelligence to auto populate configuration item based on short description.

We used classification framework to trained the system using the most recent data set (last few months) and is able to predict configuration item based on short description.

Service desk wanted to reduce the time they are spending on searching or identifying correct CI and Assignment group.

In the use case I worked on, Service desk works on agent workspace UI and interaction record will be auto created or manually created based on the channel end users are trying to reach out to. During the call/chat interaction, if the agent decides to create an incident to next level teams, agent would click on Create Incident button. And the incident form will be opened with the basic data copied from interaction, such as customer info and short description.

At this point, we are making an asynchronous GlideAjax call to server and on server-side using the ML APIs provided OOB by ServiceNow, we are finding the top 3 results along with the respective confidence levels and sending it over to client side. (Some of these results may be less than 80% confidence level)

On client side, if the script is able to receive results with >80% confidence then system auto picks the CI with top confidence among those results and fills it in Configuration Item field. Along with this it displays the top 3 CI names and their relative confidence levels too below the field as info message just as additional info to agent (Blue color message).

Since the CI is filled, the system will fill assignment group based on our existing rules using another asynchronous ajax call.

If the all the CI's are found, but none of them meet 80% confidence level, we will not autofill CI, but show the CI's predicted and their confidence levels under the field as an information to agent. (Yellow color warning message).

If system is unable to predict CIs, then we show an red color field error message saying that system is unable to predict.

Agents may still edit the CI if they want to modify it.

This helps service desk teams reduce the time they spend on filling out form and transferring ticket to level2 groups. And when new helpdesk team members onboard, it would be helpful to them in identifying correct support or resolver groups with reduced search efforts.

On an additional note here, category and subcategory fields also may be suitable candidates for classification based prediction.

Regarding priority we left that piece to the discretion of the users raising the issue.

2) Similarity Framework: OOB ServiceNow provides some similarity solution definitions in system. In our instance we created and configured below 4 search resources on incident form as per the requirements we received

Similar Open Incidents (Searches the open incidents with similar short description)

Resolver teams or help desk teams can identify similar ongoing issues to see if any existing issue is impacting the current user reporting the issue

Similar Resolved Incidents (Searches the resolved incidents with similar short description)

Resolver teams or help desk teams can check similar resolved or closed incidents to see what were the steps taken by previous analysts or themselves for similar issue in the past

Similar high priority incidents (Searches the on-going high priority issues with similar short description)

Especially helpful for help desk teams to relate dependent issues impacting organization 

Similar change requests (Searches similar change requests with given terms in short description)

Helpful to identify if a change is causing impacts to end users. For example: Lot of end users are reaching out to helpdesk that there is some issue with some network impacting certain locations. Help desk may know that in the organization they have network team doing maintenance regularly. They may use the short description terms of those maintenance changes to see if there is some maintenance underway or check a recently implemented change's role in current issue being reported.

These are some of the issues I came across or implemented in the past.

For most of these to work properly and get advantage, we should have solid process where the process teams or training teams must work in unison and set some guidelines to follow. 

The the clear inputs users give to the system in forms, the more useful PI solutions will become.

I am interested to see what other inputs our fellow community developers would provide.

Thanks and regards,

Subrahmanyam Satti

[Suggy]

Hi @subrahmanyam  @subrahmanyam Thank you so much for the details.

For point 1, when you referred CI, are you referring Service field or Configuration item.

If its Configuration item, may I know how many CIs /types of CIs are there in your environment and approximately how many unique CIs have you tagged for the Inc, say in last 6 months.

Asking this because, in our environment, we have close to 10k CIs and the agents are not tagging the right CIs for all the Incidents. So in this the predictive intelligence will be of no help in predicting the CIs  for this use case right?

Await for your inputs, thanks again.

[Subrahmanyam2 · Accepted solution]

Hi Suggy,

I am referring to Configuration Item field.
There are 2 million + CIs in our instance (30+ types or classes of CI's)

Overall, incidents were created for almost all types of CI's (Some of the CI's have less frequency of incidents and some have more) either by agents or IT teams. For example, we see most of the issues reported are for business applications or computers or servers.

In case of computers or workstations, the predictions may not work properly in certain cases. For example, end users may report issues with their laptop or desktop to help desk teams via call.

In this case agent may take the hostname identify the relevant CI and input it in the configuration item field of that particular incident. And many users will not have repetitive issues for their laptops or desktops frequently.

In case of business application, it may be a good candidate for prediction. Since applications may receive similar issues more frequently like certain site is down or under maintenance or slowness in application. If the agents in the past give the relevant info like a particular site is down or experiencing issues, and provided correct CI related to that issue, then when there is a issue in future and if the agent types in similar short description, it can cross reference and predict the CI which was used most of the times for similar short descriptions in the past.

Predictive intelligence will work best if you have sufficient and fairly good data in subjective fields in the past.

Within your internal teams may want to consider giving training to agents to make them understand the importance of filling the correct data in the fields on incident forms (or any forms for that matter).

Although system predicts the CIs and gives suggestions, the respective agents must make sure that the CI is in alignment with issue being reported. 

You may go through these ServiceNow trainings on predictive intelligence if you have not found these before. ServiceNow explained this with OOB fields and with some use cases.

https://nowlearning.service-now.com/lxp?id=learning_course_prev&course_id=75c32582db4fe0d030c91fdc13...

https://nowlearning.service-now.com/lxp?id=learning_course_prev&course_id=69aa254edb4fe0d030c91fdc13...

https://nowlearning.service-now.com/lxp?id=learning_course_prev&course_id=5eb2d6b91bc83850b9a2cae360...

--

Thanks and regards,

Subrahmanyam Satti

[Suggy]

Hi Subrahmanyam, that makes sense. Thank you so much for your valuable inputs 🙂