You can see anomalies if your topics are in fetching condition. Anomalies will not work if the topics are not pulling in any data because they –

• have been manually set to a non-fetching state

• have reached their threshold

• have crossed the end date for pulling data

You can start seeing the anomalies in 2 weeks once the model gets trained on the data. Once the data starts coming into the dashboards, however, it will take time for it to adjust to the new changes. After that, you will be able to detect alerts reliably. But if you are creating alerts using Alert Manager, then the alerts are live, i.e. you do not need to wait for 2 weeks for the model to get trained. The model has trained automatically, and you will get notifications as soon as an anomaly is detected.

When you edit a topic, the backend data is reset and the 'training' of data starts from the beginning. It will take some time for the data to adjust to the new changes. After that, you will be able to detect alerts reliably. We recommend that you wait for some time (say 2 weeks) before the model gets trained on the data coming into the topic. But if you are creating alerts using Alert Manager, then the alerts are live, i.e. you don’t need to wait for 2 weeks for the model to get trained. The model has trained automatically, and you will get notifications as soon as an anomaly is detected.

You can do one of the following while creating an inbound rule, to filter alerts if you are getting lots of alerts:

• You can choose to have Alert Type = High (cases when we see a crest, not trough) while creating an inbound rule under "The properties of the Message".

• You can define Alert Longevity where you can define to get an alert only if there are more than a specific number of alerts on the same topic or topic group.

• You can also choose to have limited types of Alert categories set up for email notifications such as only Negative Sentiment and not all of them which are: Mentions, Positive Sentiment, Negative Sentiment, Language, Country, and Engagement.

• We have introduced the Alert Severity measure which will let you subscribe to alert notifications as per the severity defined on a scale of 1-10 ( non-linear) from the rule engine.

• You can set up hierarchies or distinct POCs for alert management through queues and user groups: such that only particular users get email notifications for a certain type of alert.

Alerts will stop working once the topic reaches a non-fetching state. After that, anomalies will start from the point where things went to a non-fetching state. However, this is applicable only if the topic is unedited. If you edit the Topic, it would be treated as a new topic and would need some time to train the data to start detecting alerts reliably.

The supported sources for Smart Alerts are Twitter and Reddit.

Renaming a topic does not reset the backend data as the Topic ID remains the same. Hence, the alert detection keeps continuing the same way as before.

For Smart Alert email notifications, you need to create Inbound rules with Channel = Automatic Alerts for a particular topic/topic group, etc. The configuration of this rule is different from that of a volumetric rule. For more information, see – Smart Alert By Email.

Ensure you have set up Smart Alerts by Email and configured to send an email notification to relevant users & user groups. Also, ensure the Rule configured above is enabled, and the topics are fetching. Moreover, ensure that the Alerts email is not marked to go into Spam/Others folders.

• You can create ‘crisis issue/subject’ wise topics such as for HR & Policies, Legal & Claims, Political Associations, Finance & Governance, Campaigns, Safety & Hazard, Death or Crime, Sustainability, Data privacy & security, IT & Ops, etc along with Brand terms(broad topic).

• Ensure that the topics have retweets enabled (for Engagement alerts) and are in a fetching state.

• Rules and queues can be set up to capture and triage these alerts according to the crisis issue/subject. Filters & Tagging can also be employed to triage & classify actionability.

• Engagement columns can be created for distinct crisis issues (topics), and alert categories according to the priority/current pressing issues to the crisis management team. Appropriate agent assignment will ensue.