ServiceNow Agent Intelligence (Feb 2019)

Benefits of AI

- Agent Intelligence eliminates the bottleneck created by manually triggering tasks and decreasing error rates. -By decreasing error rates, you are improving productivity for IT teams, and end users by getting focus on the work that needs to be done, resulting in less downtime. -Eliminating bottlenecks and improving productivity results in less downtime and greater customer satisfaction. Automating manual tasks will greatly save time and resources.

What are the benefits of Agent Intelligence?

- Improve Agent productivity - Increase customer satisfaction - Develop operational intelligence

What is the Solution Definition Process?

1. Identify the ServiceNow table 2. Identify the training data set 3. Identify input field (Short Description, Location, CI) 4. Identify output fields 5. Determine training frequency

What are the four steps to prepare for agent Intelligence?

1. Improve Data Quality 2. Understand and Communicate your KPIs 3. Define Inputs and Outputs 4. Foster Organizational Change

50% class Distribution for the inquiry class means

50% of records are category inquiry

What does 50% coverage mean?

50% of records will get predictions.

What is a Solution created by?

A Solution Definition creates a Solution, which is an algorithm used to predict a target field value given one or more input field values.

What is a solution Definition?

A Solution Definition is a configuration record that specifies how to train a predictive model.

How many possible outputs can a single solution include?

A solution can only have 1 output.

What is a predictive Model?

A trained Machine-Learning algorithm used to predict field values during record creation

Agent Intelligence

Agent Intelligence (AI) is an artificial intelligence solution embedded in the ServiceNow platform. It delivers predictions based on user input. Using machine-learning algorithms, it can learn from contextual data, understand patterns, and predict intelligent outcomes without manual intervention.

Who can use Agent Intelligence?

Agent Intelligence is not available for on-primes customers and customers who have data sovereignty requirements.

What is the Characteristic of Accurate training data?

All the data points are correctly captured in the record.

What is the Characteristic of Complete training data?

All the relevant information is available in the record to make a decision.

What is a solution?

An algorithm used to predict a target field value given one or more input values.

What is a prediction?

An informed guess based on past observations (historical data)

Where are the training servers located?

Calgary, Toronto, London, Amsterdam for North American and APAC customers. Amsterdam and London ONLY for European customers data centers.

What is the difference between the class and the class distribution?

Class is the output field values for which a model can make predictions. Class distribution is number of records in the table with class.

Does Agent Intelligence support encryption?

Yes, if your data is encrypted by Full disk encryption. If the field that you are using for training is encrypted using platform encryption, the sharedservice.worker user needs to use the correct encryption context role. Agent Intelligence is not supported if you are using Edge Encryption.

Does Agent Intelligence support multiple languages?

Yes, language support is included. Please see the language field on the solution definition.

When a solution definition is scheduled to train on an interval, does the new solution automatically go live when the results are returned?

Yes, older versions are set to inactive. Administrators have the ability to activate an older version.

Can classes be excluded from prediction?

Yes. Under the Class Confidence record (you want to exclude), select the Precision Coverage Lookup combination record where Precision = 100 and Coverage = 0 and select Apply Values from the Actions list.

What is the minimum number of records to make an accurate prediction?

You need a minimum of 30,000 records to make an accurate prediction.

After receiving your solution what should you do before implementation?

You will need to test it and determine if any changes are needed. If so, tune and retrain.

What happens to your data on the training server after your Training Request is completed?

Your data is deleted from the training server.

What happens when training is complete?

The training service exports the solution back to your instance and deletes all customer training data from the training service servers. The solution is uploaded as an attachment.

How do you preserve solutions during a system clone?

There is a system property that can be set to true in order to preserve your solutions. Note, this is available in the London release.

How long can you expect training to take?

Training can take up to a few hours.

What is the Characteristic of Sufficient training data?

Training data contains sufficient number of examples for the individual outcome

What is Training Data?

Training data is a set of example records for the AI Solution to "learn" the right way of ticket assignment, classification, and prioritization.

True or False: One or more input fields can be used to make a prediction.

True

Where can you go to see if your solution has sufficient Precision and Coverage for each class?

Use the Solution Statistics dashboard to determine if a solution has sufficient precision and coverage for each class.

Is AI available on tables other than Incident?

Yes, Agent Intelligence can be expanded to other tables outside of Incident, Case, and HR Cases.

During training, an input field is used to generate

Predictions

Which types of fields can be used as outputs?

Similar to inputs, you can select any field available on the table. However, ServiceNow recommends using a string or a choice field with a finite set of possible values.

Coverage and precision are part of a ___________________________.

Solution

How many records are needed to train a viable solution?

The recommended number of records for training a good solution is from 30,000 through 300,000. If you submit more than 300,000 records, the most recent 300,000 records are used to train the solution.

What is the process to train a solution?

The solution definition and the associated records are exported to the nearest training server at a ServiceNow data center.

What is the maximum number of prediction classes that a solution can support?

The system uses the top 50 classes with the highest number of records when it builds a solution. Note that the number of classes predicted may be less than 50, and can skip a class if there is not enough historical data to build a solution that can predict the class confidently. For Kingston, the system uses the top 25 classes.

Which types of fields (reference, list, etc.,) can be used as inputs for training?

The system will allow you to select any field available on the table. However, ServiceNow recommends using choice, string, or reference fields.

What does 50% Percise mean?

Halve of the predictions are correct

ServiceNow offers out-of-the box solutions are for what applications?

ITSM, CSM, and HR