AI agents for enterprise portfolio management

AI Initiative Review: an interactive demonstration

A worked example of Kiplot's AI Initiative Review, running against a sample initiative called Reduce customer churn. It is a demonstration on a marketing page: the data is illustrative and pressing Execute writes nothing anywhere. Nothing runs until you press the Explore interactive demo button.

The review runs four checks over the initiative. Each one finishes on its own, and you pick a check to see what it proposes. Most proposals are a change to one field: the field, its current value, and the value the review suggests instead. The Actions check is different, because an action that has gone quiet cannot be fixed by editing it, so it proposes a message in Microsoft Teams to the person responsible, already written and ready to read. You select the ones you want, review the set, then execute them. Nothing is changed and nothing is sent until you execute.

The four checks

• Plan Coherence: compares outcomes, risks and key results against the delivery plan.
• RAG: reconciles the reported status against its own roll-up.
• Actions: checks the age and ownership of open actions, and drafts the chase message.
• Data Quality: reads planning and financial fields for values that contradict each other.

Examples of the changes it proposes

• Extend an initiative end date when committed spend falls outside the window.
• Close an outcome whose end date has passed and whose tasks are all complete.
• Assign an owner to a high-severity risk that has been unowned for weeks.
• Move an initiative's RAG to Amber when its own outcomes disagree with it.

Examples of the messages it drafts

• Asking the owner of a long-overdue mitigating action for an update and a revised date.
• Asking someone with four past-due actions whether they are still live.
• Asking a colleague to take three actions assigned to a deactivated account.

Every proposal carries the reasoning behind it and the record it was drawn from, so the evidence can be checked before it is accepted. The agent finds the problem, proposes the edit or writes the message, and names the person; a human still decides whether it happens.