TechNow Ep 59 | Work smarter with machine learning-driven automation

[Chuck Tomasi]

Does your organization struggle with an ever-increasing volume of tickets? Are your agents inundated with tickets that are categorized and routed incorrectly? ServiceNow® Agent Intelligence enables you to decrease resolution times and increase customer satisfaction using machine-learning algorithms to set field values during record creation, such as setting the incident category and assignment group based on the short description – letting your agents focus on what they do best. Join the TechNow team, along with special guest Matt Train, ServiceNow Sr. Advisory Solution Consultant, as they explore the depths of Agent Intelligence and how it can be put to work for you.

Quick tip: Related list notation

Originally aired: December 11, 2018 9:00 AM PT

Featured Experts

Stacey Bailey is a Training and Certification Program Manager for ServiceNow with over 20 years of IT experience in various roles. Since 2009 she has been working with the ServiceNow platform customer, partner, and employee. She has earned numerous awards in her career including: ITSM MVP 2017, InfoWorld Top 100 Award, DCI Portal Excellence Award, Financial Times Award for Excellence in E-Learning. Stacey proudly refers to herself a s a "gadget girl". In her spare time she likes trying new gadgets, learning new forms of art, playing in the makerspace, collecting new / unique skills, listening to audio-books, trying new life experiences, and hosting parties to build communities.

Chuck Tomasi is a Sr. TPMM for ServiceNow. He is a computer science major with over 35 years of IT experience. As a former ServiceNow customer, Chuck won the first Innovation of the Year Award at Knowledge 10. Since joining ServiceNow in 2010 as a Technical Consultant, he has done many large scale ITSM implementations and custom applications, acted as an adjunct instructor for Education Services, created and lead the Technical Best Practices program, makes appearances on Live Coding Happy Hour, created dozens of fit for purpose custom applications, and co-hosts the ServiceNow series "TechNow".

Kreg Steppe is a Training and Certification Program Manager within ServiceNow developing and supporting cloud training infrastructure. He specializes in developing integration solutions, automating repeatable processes and Cloud Management in ITOM. Kreg's prior experience includes operating his own ISP, developing web applications in PHP, network integration, managing network support, Application Development on cloud based networks, DNS and email server maintenance. He is a Linux enthusiast and enjoys Photography.

Matt Train is a Solution Consultant for ServiceNow specializing in Platform Intelligence (Performance Analytics, Agent Intelligence and Virtual Agent.) Matt brings over 20 years experience in analytics and data warehousing with products such as Cognos and Birst. He has been with ServiceNow since early 2017 where he started with Performance Analytics and expanding in to the other platform intelligence products more recently. His hobbies include home automation, hiking and wine.

---

LIST ALL EPISODES

Questions and Answers

Can ServiceNow's Machine Learning be used to extract free-form text from a legacy system and put it into a ServiceNow common data model when going from legacy to ServiceNow? -- Richard Fekecs
It sounds like you want to do more of an integration. Agent Intelligence is going to be valuable after you've got your data in to ServiceNow. It won't access information from third party systems. -- Chuck Tomasi

How does this Automated Agent assign or ensure a CI is in the record? -- Roderick Barnett
Within the Solution Definition, you may assign output fields to be populated based on predictions. CI can be defined as an output field. However, if doing this, you must ensure that there is correlation between the CI and whatever fields you select as the input fields. Additionally, when setting the filter to identify records to use for training, make sure to select records where the CI field is not empty. To read more about this, see the Docs article 'Create and train a solution definition' (https://docs.servicenow.com/bundle/london-servicenow-platform/page/administer/agent-intelligence/task/create-solution-definition.html). Agent Intelligence is not going to help to ensure that a CI is populated in the record. For that, you should look to Data Policies or UI Policies. -- Stacey Bailey

Can we train on a sub-prod instance and move that to our Prod instance? The reason I ask -- we JUST archived our old incidents and have a small # of active ones now. We could restore those on a sub-prod instance if this would work. -- Susan Williams
Yes, that is possible. After the training is complete on your sub-prod instance, create an update set. Then, from the related links on the Solution Definition, select 'Add solutions to current update set' Then, you can mark that update set complete and promote it through your instance stack as normal. -- Matt Train

It is available -- Björn Rodemund
Yes. The features being discussed here were first introduced in Kingston. They are available as part of subscription packages. Matt discussed those on slide 16, so be sure to watch the replay for that information. -- Stacey Bailey

Can we train in more than two languages? -- John Feist
Supported languages include Dutch, English, French, German, Japanese, and Spanish. When training solutions, you may select the dominant language of the data set being analyzed. You may use filters to select records for language-specific solution definitions based on criteria like the caller's language. This is mentioned as an approach also in 'Configuration tips for Agent Intelligence' (https://docs.servicenow.com/bundle/london-servicenow-platform/page/administer/agent-intelligence/reference/troubleshooting-agent-intelligence.html) -- Stacey Bailey

After auto assigning to a group can it then auto assign to a tech? and how complicated is that for a round-robin? Can you take into consideration schedules, out sick, workload? -- Wendy Lonstein
As Matt just mentioned, Agent Intelligence currently targets Assignment Groups rather than individuals. You may want to take a look at On-Call Scheduling. This may have the functionality you are looking for to auto-assign to a tech. Per Docs: - On-call scheduling provides a way to determine which member of a user group is available to complete a task, for example to find the right person to assign an incident. - You can regularly rotate on-call positions among some or all members of a group of users. https://docs.servicenow.com/bundle/london-it-service-management/page/administer/user-administration/concept/c_OnCallScheduling.html -- Stacey Bailey

So in order to leverage the AI one would have to move the data into ServiceNow, and then run the AI to clean up and categorize the data? -- Richard Fekecs
Correct. Agent Intelligence works on data within ServiceNow only. Of course, manually setting the filters on the solution definition to select only good data will help ensure that the solutions developed by the AI are good as well. See the Docs article, 'Selecting data records for training your solution' for more recommendations for training. https://docs.servicenow.com/bundle/london-servicenow-platform/page/administer/agent-intelligence/concept/agent-intelligence.html -- Stacey Bailey

So is this not available on self-hosted instances of ServiceNow? -- Sandra Matulis
Correct. It is only available for instances hosted by ServiceNow. Also, currently, it is not available for instances hosted within the US FedRamp data centers. -- Stacey Bailey

So we will have to send our data out to be analyzed? -- Sandra Matulis
Not all data, just certain fields like short description, description, assignment group, assigned to, and priority so it knows "based on this text, this was the result". You're not sending the entire instance data and the data does not leave the datacenter. -- Chuck Tomasi

Can the Agent Intelligence be used on any ticket? such as catalog items? -- Patricia Nichols
If you have a critical mass (>30K) to generate a reliable model, sure. -- Chuck Tomasi

What's the name of the plugin..can't find it -- Kapoor Shalini
Activation instructions may be found at https://docs.servicenow.com/bundle/london-servicenow-platform/page/administer/agent-intelligence/task/activate-agent-intelligence.html Plugins: Agent Intelligence (com.glide.platform_ml) Note that Agent Intelligence is included in the following packages: ITSM Professional, CSM Professional, HR Professional, and HR Enterprise. When you subscribe to any of these packages, you can activate Agent Intelligence on your production instance. -- Stacey Bailey

Thanks! -- Susan Williams
You are very welcome! We hope to see you in January! -- Stacey Bailey

Does the algorithm uses clustering techniques? -- Rupam .
No, it does not. It uses Logistical Regression. -- Matt Train

30K Incident records was recommended for Training and Class Files are created for Assignment Groups to create models, how many records are required per Assignment Group out of that 30K ? -- Paul Ybarra
30K is a minimum - the more data, the more accuracy you'll get. There's no (published) minimum per group. -- Chuck Tomasi

If we've customized Category (to a reference field) , can we use this pretty easily? What updates would we need to make? -- Susan Williams
As long as the field has a finite set of values (like the choices you've defined for Category), you should be okay to use that with no updates or modifications. Per Docs: In general, a good output field has these characteristics. - The field is a choice field or a string field with a finite set of possible values. - The field has some casual connection to the input fields. For more, see: https://docs.servicenow.com/bundle/london-servicenow-platform/page/administer/agent-intelligence/task/create-solution-definition.html -- Stacey Bailey

a model if developed and trained is pretty much a function in my understand. do you provide sort of restful api to call out side of function? -- Connie Cheng
Yes, there is a REST API for Agent Intelligence. The business rule, Default Task Based Prediction, shows a sample call to retrieve predictions for a specific record. var predictor = new MLPredictor(); var solutions = predictor.findActiveSolutionsForRecord(current); predictor.applyPrediction(current, solutions); -- Stacey Bailey

Can you have multiple inputs for the model? -- Katherine Cook
You may define multiple fields as inputs for the model, but all fields must come from the same table. -- Stacey Bailey

I'm not clear on what is used to build the assignment rules? Is it past incidents? Knowledge Articles? -- Brian Barrett
The prediction models would be based on analyzing past incidents. Here is a good overview of the process: https://docs.servicenow.com/bundle/london-servicenow-platform/page/administer/agent-intelligence/task/get-started-agent-intelligence.html -- Stacey Bailey

Sorry, he said that it was not available in Dev instances, but it is available, you just cannot train with the OotB demo data. -- Björn Rodemund
It's not available on the free Personal Developer instances. He didn't mean YOUR dev instance (e.g. dev, test, prod) -- Chuck Tomasi

It is in the free personal Dev instances...at least it is in mine...just saying, not important. -- Björn Rodemund
Actually, Agent Intelligence is not included in personal developer instances. -- Stacey Bailey

We use Incident assignment rules to set the AG on Incident form, and we display the suggested group in an infoMessage -- then we have a UI action if we would like to use the suggested group that would override the group on form and replace with suggested group -- Jamie Nolan
Agent Intelligence includes a business rule called 'Default Task Based Prediction' that is inactive by default. You should be able to look at the logic in this business rule to leverage that from wtihin a UI action instead. (function executeRule(current, previous /*null when async*/) { var predictor = new MLPredictor(); var solutions = predictor.findActiveSolutionsForRecord(current); predictor.applyPrediction(current, solutions); })(current, previous); -- Stacey Bailey

If you spin up new ServiceNow instance for new client, does the 'training' wait to have 30.000 records or so, OR will it do the analysis based on existing data available on ServiceNow datacenter? -- VALON SHEREMETI
While it is possible to train with fewer records, 30K represents a good dataset for more accurate analysis. -- Stacey Bailey

What advantage do we receive from using ML in categorization :: assingment group in INC rather than identifying the categories one time and configure them in a separate table(or a data lookup kind of config) -- Debajyoti Das
It's automatically done by Agent Intelligence. The advantage is faster triage which translates in to faster resolution and happier clients/customers/employees. Those who would normally do triage get to spend their time doing more value-add work. -- Chuck Tomasi

I am seeing Agent Intelligence module on our Instance already. Will this Agent Intelligence works by default or do we need to configure -- Sree GORANTLA
There's some sample data for you to look at. If you want to create a new model, it will require a subscription. -- Chuck Tomasi

Does this use the MID server? -- Susan Williams
No. MID servers are generally used to access resources within a customer network. For Agent Intelligence, the data used for training stays within the ServiceNow data center. -- Stacey Bailey

For those of us that deal with data that is HIPAA regulated etc., sending the raw data to a training server for analysis is not an option. Do you offer any alternatives to analyze the data within our instance? -- Shubha Chakravarthy
We're not sending the entire raw data, only select fields (e.g. short_description/description) needed to understand the context to do the priority/assignment. -- Chuck Tomasi

yes after there is critical mass of data.... so there will be a period of time where you will have to create something via normal routing -- Mathew Xagoraris
Yes, so the machine learning can base it's model no something that was done properly. -- Chuck Tomasi

Could you force a solution training for a smaller data sample? Like when you onboard a new application and want to get it in predictions? -- Björn Rodemund
As Matt mention, it is possible. The key consideration is having enough records for statistical significance. -- Stacey Bailey

So your doing the Assignment group via Description, similar to CI? -- Karen Guerra
It analyzes the history of what keywords were used to drive successful assignment on prior records (incidents, CSM cases, etc.) -- Chuck Tomasi

To get this working in a msp environment with multiple customers... how would you do.. Do we create multiple template? -- Anurag Tiwari
When defining a solution definition, you are able to specify input fields to be used for training. In the case of MSP customers, domain would be an important input field to consider if developing a consolidated solution definition. Creating multiple solution definitions that filter for specific domains or domain paths is another option to consider. However, with that path, remember that we need around 30K records to use for training. -- Stacey Bailey

Is there a plugin for Agent Intelligence...i joined late and missed that part -- Kapoor Shalini
Yes -- Chuck Tomasi

Has there been any discussion about using machine learning to improve the knowledge module, particularly searching? -- Amanda Fullan
That's a very compelling idea! I'd suggest keeping an eye out for this type of capability in future releases. You may also want to log an enhancement request on hi.servicenow.com. -- Stacey Bailey

If you have email parser assigning tickets to the Agent/ServiceDesk can Agent Intelligence be applied to these tickets. -- Patricia Nichols
Yes, you may use Agent Intelligence to assign tickets to a particular group or set the category. -- Stacey Bailey

what if I already have a model developed and trained our side of SNOW, can I use it in the SNOW? -- Connie Cheng
Agent Intelligence only understands the data model created by the training instance. -- Chuck Tomasi

How does the raw data get sent to the training? -- Rita Audi
You define the training solution, identifying the specific records and fields to be included for training. Then, your instance establishes a secure connection with the trainer service within the same data center. It sends the data you identified to the trainer service. The trainer service develops the models and securely sends those back to your instance. It also destroys the data on the training service instance that was used to generate the models. -- Stacey Bailey

what kind of AI skills are expected to leverage the SNow offering on Agent Intelligence and related offerings -- N V L VITAL KANTH JALLURI
No specialised AI skills are required to use this solution. It is a 'purpose-built' AI solution that does not require specialised skills. -- Matt Train

Does it also work from creating an incident via the service portal and directly in ServiceNow from Create Incident? -- Andrea Holmes
Yes. -- Chuck Tomasi

My apologies, I may have missed this. Will it be first available in the London release? -- Steven Shanks
Agent Intelligence was first introduced in Kingston. If you watch the replay, which should be posted later in the week, Matt talked about Availability on Slide 15. -- Stacey Bailey

Is Agent Intellience going to tie into Virtual Agent in the future to choose topics from user entries? -- Björn Rodemund
Keep an eye on the roadmap for Agent Intelligence. It is a rapidly-evolving application. -- Stacey Bailey

can a training instance be a none SNOW server? -- Connie Cheng
Not at this point. -- Chuck Tomasi

What should onpremise customer do to leverage VI, then -- Debajyoti Das
There is no option in the London release to take advantage of Agent Intelligence. -- Matt Train

Thanks people, another good one! -- Björn Rodemund
I'm so glad you found this webinar to be useful, Björn! We hope to see you for the next one also. -- Stacey Bailey

can this help if cmdb is not upto date -- ALEX JOSE
Agent Intelligence probably isn't what you are looking for CMDB compliance. You might want to look at Discovery or Service Mapping to pull in real live information about your CIs. -- Chuck Tomasi

Thanks that answers my question 🙂 -- Anurag Tiwari
You are welcome. -- Chuck Tomasi

What are your long term Agent Intelligence enhancement plans? Are you looking beyond automatic assignments such as determining criticality of incidents and notifying the appropriate users/groups based on the subject matter of the incidents? -- Shubha Chakravarthy
You may be able to train solutions that set criticality based on record content. Then, you may use the notifications engine to trigger notifications to particular groups. -- Stacey Bailey

Thanks All! -- Michael Acampora
You are most welcome, Michael! We hope to see you at next month's webinar. -- Stacey Bailey

so if you go through a reorg..... the model fails right? -- Mathew Xagoraris
Not necessarily. As Matt is showing now, the solutions can be re-trained on a regular basis to keep pace with any major changes in how records are categorized and assigned. Ideally, though, assignment rules and categories should not necessarily be based on organizational structures. -- Stacey Bailey

What happens if there are assignment rules set up already. Who runs first? AI or the rule? -- Björn Rodemund
The default business rule that retrieves the prediction result is called 'Default Task Based Prediction.' It is a 'before' business rule that runs on 'insert' with an order of 100 (though that is configurable). Based on the Docs article listed below, the order of precedence is as follows: 1. Data lookup rules take precedence over assignment rules. 2. All assignment rules run before business rules that run on a record insert with an order value less than 1000. 3. The first assignment rule with the lowest execution order and matching condition. 4. All assignment rules run before business rules that run on a record insert with an order value greater than or equal to 1000. 5. All assignment rules run after business rules that run on record insert. See Docs: 'Precedence between data lookup, assignment, and business rules' (https://docs.servicenow.com/bundle/london-platform-administration/page/administer/task-table/concept/c_PrecBetweenAssignmentAndBusRules.html) -- Stacey Bailey

I think the use of Assignment Groups is implemented pretty consistently across most SN instances, and it's easy to see how AI will automate the process of assigning tickets. However, there is much more diversity in how categorization is implemented often with multiple fields and hierarchies (e.g. Business Service, Technical Service, Category, Sub category, Configuration Item, etc.). How does Agent Intelligence handle this kind of complexity? -- Stephan Thomsen
You may specify multiple input fields to be used as part of the training process. If there is sufficient correlation amongst those fields and your output field (such as category), then Agent Intelligence may based predictive solutions on those fields. -- Stacey Bailey

So ServiceNow is Supervised Learning with Assignment Group as the "Labels" it is trying to predict? -- Paul Ybarra
It is supervised learning. However, it has the ability to use mulitple input fields and mulitple output fields. It is not restricted to just Assignment Group as the output. -- Stacey Bailey

Does the 'Intelligence' auto-update once it it implemented or are regular analysis sessions required to keep the model up-to-date? -- Gerard Dixon
To do a refresh, (e.g. new org structure) you need to re-train the model. This can be done on a schedule. -- Chuck Tomasi

We get a lot of incidents coming in via email. Can the training also consider sender and/or To: ? -- John Feist
The sender is typically mapped to the Caller field, so that should not pose a problem. The baseline Inbound Email Action does not capture all of the recipients of an email as part of the generated Incident. However, if you map that infomration to a field in the generated record, you can certainly use those as well. -- Stacey Bailey

So do have to have Virtual Agent installed or would this work from the Provider service view? -- Helen Scheck
Agent Intelligence works to predict and populate fields on a record, such as category or assignment group. It generally works in the background and the results are visible to providers / agents. Virtual Agent is a conversational interface for users. -- Stacey Bailey

so, this statement was very confusing... is this feature available in personal test instances or not? -- Peter Velinov
It is not available in personal developer instances. -- Stacey Bailey

How will this work when tickets are entered in multiple languages? -- steve vaughan
Hopefully, Matt will address this in his demo. There is an option to select the dominant language when creating a solution definition. https://docs.servicenow.com/bundle/london-servicenow-platform/page/administer/agent-intelligence/task/create-solution-definition.html -- Stacey Bailey

Has ServiceNow considered incorporating a hierarchy structure for Assignment group? Today, I ‘group’ Assignment groups via a naming structure. -- Jim Porter
We already do. There is a 'parent' field on the sys_user_group (Group) table. -- Chuck Tomasi

Does look like the plugins readonly on PDI, doesn't it? -- Munhwan Gim
Yes, Agent Intelligence is not supported on personal developer instances at this time. -- Chuck Tomasi

You said it can handle up to 50 classes. Does that mean for assignment groups that its only going to work for 50 groups? -- Chad Hall
Correct. However, take into consideration that it will train for the most common 50 groups. Most often, there is a set of common groups that represent the majority (~80%) coverage as well as a "long tail" of other groups used less commonly. You may also have multiple solution definitions to help get beyond that limitation. However, there may be conflicts with the predictions. -- Stacey Bailey

thank you! -- Dwayne Whittington
You are very welcome! We hope to see you next month. -- Stacey Bailey

can tickets be auto assigned to correct groups for alerts coming from monitoring tools like dynatrace, sunk, sumo etc -- ALEX JOSE
Yes, that would be done with standard assignment rules. Agent Intelligence is not required for that functionality. -- Chuck Tomasi