Predictive Intelligence for HR - Auto determination of HR Assignment group (Article 1)

[Lener Pacania1]

Overview

Welcome to the first article in a series of five where I will share how to use ServiceNow’s Predictive Intelligence (PI) to improve the employee-agent HR experience. Predictive Intelligence is an in-platform machine learning capability that can drive following outcomes: reduce case hop by predicting the right place to send work, recommend solutions to help agents solve cases faster, find patterns that may be impacting your process, identify gaps in your knowledge base, and prevent profanity in our agent-requestor interactions.

In the five articles I’ll show you how to configure, train, use, and troubleshoot some of the most common Predictive Intelligence use cases for HR. Each article contains a mini lab with steps that you should be able to use in your San Diego instance.

Article 1 - Auto determination of HR Assignment Group
Article 2 - Use similarity to help agents solve issues faster
Article 3 - Find patterns using clustering
Article 4 - Use Knowledge Demand Insights to find gaps in your knowledge base
Article 5 – Watch agent language with Profanity filtering

Article 6- Tuning Predictive Intelligence Models (part 5) – Regression 

DISCLAIMER - This article series is ambitious as I would have preferred to use a controlled Knowledge lab as a teaching platform. In a lab environment all my screen shots would correspond to your results. Because you are using your instance, your results will vary because your data is different than my data. My hope is that you’ll be able to follow along with your instance and replicate these results. If you run into issues, you can (1) post them here (2) open a support case (3) reach out to your ServiceNow HR sales team for assistance.

These articles are not exhaustive. For an overview of Predictive Intelligence for HR please see this video. For additional instructions on other HR PI uses cases please see the San Diego Predictive Intelligence for HR documentation.

REFERENCE - These labs are meant to get you results quickly, so I don’t go into detail explaining each step in the labs. I’ve skipped explaining each step because cover those details in other Knowledge Courses and articles. For example, if you would like to understand what a word corpus is or how clustering works you’ll want to head over to the below NowLearning courses that I’ve listed and invest some time there.

NowLearning Courses:
K21 Accelerate Incident Resolution for Agents using AI/ML and Flow Designer
Predictive Intelligence Advanced Topics
K20 Accelerate Incident Resolution with Predictive Intelligence and Agent Workspace

Before you can get started.

1. The lab assumes you are using a sub-prod San Diego instance cloned from production and you have basic experience using HR Case Management on ServiceNow.
2. Your user id needs to be assigned the hr_admin & ml_admin roles in your sub-prod instance. Ask you administrator to confirm that you have these roles.
3. You need to have a minimum of 10k cases in your sn_hr_core_case table. You can check if you meet this requirement by going to all > application filter > sn_hr_core_case.list
4. The following plugins should be installed by your admin:

• Human Resources Scoped App: Core
• Predictive Intelligence
• Predictive Intelligence Workbench
• Predictive Intelligence Workbench HRSD Content
• Predictive Intelligence for Contextual Search
• Predictive Intelligence for Knowledge Management
• Predictive Intelligence for Universal Request

5. Your admin should set the below system properties to true. These properties enable the Auto Determination HR capabilities. If these properties do not exist create them.

• sn_hr_core.case_auto_assignment set to true
• glide.platform_ml.auto_training.enabled set to true

Lab 1 - Auto determine the HR Assignment group in HR Agent Workspace

We will use Predictive Intelligence to auto determine (aka predict) the assignment group from a case after the employee enters a short description. Predictive Intelligence “learns” from past cases to predict how to route future cases. Using machine learning in this way is useful to reduce the dependency on assignment rules and to reduce the amount of case hop due to sending work to the wrong team (for example a payroll issue getting sent to benefits).

Configure & Train Auto determination

1. All > Predictive Intelligence Homepage
2. Expand the Classification carrot
3. Scroll down to Auto assignment group for HR Cases
4. Click Train

5. Your PI solution may take up to 30minutes to train depending on the size of your data.  When the PI solution is finished training it will have a [Version: #] next to it.

See Auto Determination in Action in HR Agent Workspace

1. Impersonate a HR user (preferably a HR agent). In my instance I impersonate Harry Taylor.
2. Go to All > Agent Workspace for HR Case Management
3. Under All HR Cases click assigned to me.
4. Create a new case, by clicking New

5. Select a new employee, in my case this is Abel Tuter.

6. Click Next
7. Populate the case details with valid information relevant to YOUR SYSTEM.
In my instance I use
   COE = HRIT Total Rewards Case
   HR Service = Dental Benefits Inquiry
   Work Notes = I have a question about my dental benefits

8. Click Create Case

9. If PI was able to make a prediction with high confidence a blue box with the assignment group prediction will show on the case form. At this point the agent can decide if they want to go with the prediction from Predictive Intelligence or stay with the default prediction.

10. In this case let’s change the assignment group and use the prediction. Click into the Assignment group field and change from HR Tier 1 to HR Benefits and hit Save.

Congrats you’ve finished learning about how use machine learning to route work to the right place!

[Lener Pacania1]

Trouble Shooting - Lab 1

If you’re not familiar with Predictive Intelligence I strongly recommend you first take the NowLearning courses that I listed in the overview before attempting the troubleshooting steps.

Issue - I don’t see the blue prediction box in Agent Workspace after I create a case.

Most Probable Cause- the data you are testing may not exist in the training set.

Test Existing OOTB Auto assignment classification model

1. Go to All>Predictive Intelligence>Homepage
2. Expand Classification and scroll down to the “Auto assignment group for HR Cases”
3. Click the test button, then click the test solution tab
4. Click batch test and enter a case number in the filter to test. In my example below I am using an existing case number (HRC0000100) which has the input data I wish to test with. Hit Run Test.

5. Try testing multiple different case numbers. If all your results show “no prediction/unable to predict” for that input record(s). Then you need to create a new classification model to fit your data.

Create a classification model and manually train

6. Since you can’t modify the OOTB classification model, you will need to create and manually train a new classification model by following these steps.
7. The steps at a high level will instruct you to:
     a. Create a new classification model, I called mine “Lener Predict HR Assignment Group”
     b. Define new input & filter criteria to include the right data to train the model. In my case my input  is short description and I cleared all the filters.

 

8. Update & Re-Train your new classification definition..
9. Now we need to swap the OOTB classification solution with this new one.

Swap the OOTB classification model with the new classification model

10. Go to All > HR AI Configuration and select Case Assignment Group.
11. Select your newly created classification solution from step 6 and hit update.

12. Your HR AI Configuration screen should show the updated model for Case Assignment Group.
13. Try going back to step 4 and re-testing. If that works you should be able to see the prediction from HR Agent workspace.

[Dave65]

Lener

Thank you for the time and effort you put into developing these articles.

Do you have a Video demo of each of these labs that show them up and running after they have been configured?

We are just really getting going with HRSD and are currently in the Quebec release. We will be up to San Diego by the end of the summer but we still have a long way to go to get to 10k cases.

My manager was interested and wondered if there is a demo we can all look at to help with future planning.

Thanks again for your help,

Dave Johnson

[Dave65]

I do have 1 other question; Will these lab work on earlier version such as Quebec or are they utilizing functionality that is new to San Diego?

[Lener Pacania1]

Hi Dave, I'm working on a video demo that will show the end result, that should be out before end of this week.  The labs apply to Quebec except that the end user interfaces will look different. In San Diego we introduced HR Agent Workspace.   

[Lener Pacania1]

All the configuration, training, trouble shooting is the same in Quebec.  You'll need to leverage the Quebec docs to implement the predictive intelligence models.  If memory serves me you will need to use the HR AI configuration to do this or business rules to call the PI models if you are using traditional HR forms.  Let me know here if you have any issues with those docs.

[Dave65]

Great news!

I'm looking forward to them.

Will it cover each of the labs?

Thanks again,

Dave

[Dave65]

Would there be enough data in a PDI to do the labs?

Is there an easy way to get enough data if a PDI doesn't have enough? I know that there are some data packages for different modules in addition to the data that you can install when you install a module.

I realize that the more data available, the better the system will be trained. I just don't know if ServiceNow has implemented some limitation that will prevent the training if there isn't enough data to train on.

[Lener Pacania1]

Hi Dave, I doubt the PDI demo data will have the right type of data (meaning lots of variation) nor the right amount of data (min 10k rows) to train the models effectively.  That said you could give it the "old college try".  If you go the PDI route, in the system properties you can reduce the limits down to bare minimum.  Just be warned I've done this in the past with other data and didn't get either was unable to train or go horrible results.  I've been helping customers with PI since Kingston and always recommend a clone from production, it's the only way to see if machine learning will be useful for your organization since the training of the algorithms is highly dependent on the data. HTH-Lener

[Lener Pacania1]

Direct link to Article 5 – Watch agent language with Profanity filtering

[Lener Pacania1]

Another trouble shooting tip.  Try logging in as a different user with admin & hr_admin roles. When I tested recently on an instance my user Harry couldn't see the prediction even though the prediction was working in the classification testing tab.  I tried it with Admin and other users and they could see the prediction, so I suspect that Harry has some special business rules or assignment rules that prevent the blue prediction dialog box from showing.  

[Florencia Lucer]

Hello! This is super helpfull, thank you 🙂

I have a question for you, I do have one HR Service that shouldnt suggest other HR Service to be transfered, how can I exclude one HR Service?

Below you can find current configuration:

The HR Service that shouldnt suggest other one is CAM HR Navigate.

Any ideas?

[Lener Pacania1]

Hi Tera, you're screen shot is correct on how to exclude.  By excluding CAM HR Navigate from the training set the ML model should not have any CAM HR Navigate cases to predict.  Did you already try this and it's  not working as expected? -Lener

[Judy Dong]

Hi Lener! 

Do you know if this is possible to configure with Universal Request? 

Context: 

If a universal request is submitted from the Request Help widget on the portal, depending on the short description, it will create an Incident ticket or an HR ticket. Would I be able to customize a Predictive Intelligence model to look at that newly created Incident or HR Case and assign an assignment group to that departmental ticket based on what the client's data has historically assigned a ticket with that short description content to? 

I don't want a Predictive Intelligence model to look at ALL incidents and HR tickets that are created and assign them to what they have historically been assigned to. 

I want one that will only look at the departmental tickets that have been created from a parent Universal Request record. 

[Lener Pacania1]

Hi Judy,

One implementation pattern is to call the PI model from a business rule attached to a table.  So for example if a new incident/case is created you can fire the PI model to predict the assignment group.  You could add IF..THEN logic in the business rule code to only look at specific department tickets.   HTH-Lener 

PS: at the very bottom of this article there is an example of what you can do with PI and a business rule 

[ge3]

Hi Lener, Saw your input on HR case auto determination of Assignment group. I have some questions. There are now use cases on predicting HR service and AG. I notice that PI only suggests via a blue box? IS it same with email categorization? the case created will also just suggest the HR service, not create the case with the predicted HR service already? Pls advise.

Also I wanted to use a different output field (category on hr case for example) but couldn't get it to work similar in incident categorizations.  I just copied the solution from HR class classification, change the output to the category field and set some filters, What am i missing?

[Siva Kumar Nayk]

@Lener Pacania1 - Does the ML need to be trained based on the historic data only or can the PI be turned on and it can take the live feed of the closed HR cases and train itself on the data and when we are ready we can activate it for the real time usage?

[Anupriya23]

I created my custom solution definition for a custom field called category but I am unable to figure out if it is possible to populate predicted value on load /on change  by script on HR case?

[Szymon Seroka]

@Lener Pacania1 

have you ever seen PI used to predict the Assigned To. 

If you want to keep you Services list short (keep them open to make it easy for EEs to just type in what they want) you end up with having a large variety of cases coming through that service to say a Customer Support team - they they specialize in topics but to assign work to the person specializing in a topic you need someone to read every case and Triage it to the right person. How would you approach assigning it To the person with "best track record of solving similar cases in the past"