All topics
9 / 55
OBIEEintermediate

iBots / Agents & Alerts

Explain how scheduled agents (formerly iBots) deliver proactive, condition-based alerts and content distribution in OBIEE.

An agent turns OBIEE from something users visit into something that tells them when to look. They were called iBots.

It's the difference between a smoke detector (alerts only when there's actually smoke) and a nightly PA announcement reminding everyone to check for fires themselves — Agents are the smoke detector.

Key Concepts

1
What an agent does.
    schedule or event
        |
    run an analysis
        |
    evaluate a condition -- does it return any rows?
        |
    if met, deliver: email, dashboard alert, a web service call,
                     a downstream action
2
The condition is what makes it an alert rather than a scheduled report.
    analysis: orders on credit hold for more than 3 days
    condition: returns at least one row
    action: email the credit team
3
Nothing arrives on a day with no held orders. That is the difference between an alert and a daily email people learn to ignore.
4
Delivery can be personalised. An agent can run once and deliver a different result to each recipient, by using session variables in the analysis:
5
    the analysis filters on VALUEOF(NQ_SESSION.USER_REGION)
    -> each recipient receives their own region's rows
6
One agent, forty recipients, forty different reports. Without this you would build forty agents.
7
Conditional delivery scales further. An agent's recipient list can itself come from an analysis — email whoever owns each exception, derived from the data rather than from a fixed list.
8
Chaining. An agent can trigger another agent or invoke a web service, which is how BI drives an action rather than just reporting: detect a stock shortage, and call a service that raises a replenishment request.
9
The practical cautions.Scheduling. Agents run on the scheduler, which competes with interactive users. Twenty agents at 07:00 produces a spike exactly when people are logging in. Stagger them.
10
Delivery failures are quiet. An agent whose email delivery fails usually reports success for the run. People stop receiving alerts and assume there is nothing to report, which is worse than no alert at all.
11
Data-driven conditions need data. An agent whose analysis fails — because the ETL did not finish — returns no rows, so the condition is not met and nothing is sent. Silence means "nothing wrong" and "nothing ran" equally, which is why a heartbeat agent that always sends is worth having.
12
Where agents belong. Exceptions and thresholds, where somebody needs to act. A full report emailed on a schedule is a scheduled report, and dressing it as an alert trains people to ignore the alerts that matter.
13
What the interviewer is probing.1. "What is an agent?" Probing: the vocabulary. Stalls: "An email." Moves up: a scheduled job that evaluates a condition, and on success delivers content or triggers an action — not merely a scheduled report.
14
2. "What makes a conditional agent valuable?" Probing: the design point. Stalls: "Nothing." Moves up: it only notifies when something needs attention, so people read it — a daily report that is usually fine gets ignored.
15
3. "What is personalised delivery?" Probing: the capability. Stalls: "Nothing." Moves up: running the agent in each recipient's security context so each sees only their data from one definition.
16
4. "What is the operational concern?" Probing: the practical angle. Stalls: "None." *Moves up:* many agents firing at the same time compete with interactive users, so scheduling needs spreading out.
Use cases:agentsalertingproactive-bi