The AI does not read your playbook once and forget about it. Every time something happens on a task, it checks the playbook again and decides what to do. This article explains what makes it check, what it can see while checking, how it reads your rules, and what it cannot do yet.
Where to find it
Your playbook instructions and rules live in AI Control Centre → Task Manager, under each task category's Playbook tab, plus one company-wide set that applies to every category.
What makes the AI check a task
The AI looks at a task's playbook again whenever one of these happens:
Something new lands on the task. An email, a call, a text message, an internal note or a promise to pay. This is the most common trigger and the way most playbooks are written: "when something happens, do X."
The task is created.
A promise to pay changes. It is recorded, kept, missed, cancelled or rejected.
A person changes the task's status. A status change made by the AI itself does not count, so it cannot react to its own actions in a loop.
A payment comes in or a linked invoice balance changes.
A reminder set on a note comes due.
The task's due date passes.
Someone clicks "rerun AI" on the task.
Nothing else wakes the AI. In particular, editing a customer's details or custom fields does not by itself trigger a check. Those details are read the next time one of the events above happens.
What the AI can see when it checks
Everything linked to the task: its emails, calls, notes, promises to pay and invoices, plus why the task was created and what the AI decided on earlier checks.
The customer's details: country, currency, account owner, tags, portfolios, whether they have an email address or phone number on file, and their custom field values.
This means rules can key on customer details, for example
"disputes from US customers go to the US team" or "if the custom field 'Key account'
is Yes, set priority to high."
What the AI cannot see: anything not linked to the task, changes made in your accounting system until they sync into Kolleno, and other customers' history.
How the AI reads your rules
You can have two kinds of rules: rules for one task category, and company-wide rules that apply to every category. The AI reads both together on every check. Neither replaces the other.
Every rule is judged on its own. For each one the AI records one of three outcomes:
The rule applied and its condition was true, so the AI acted on it.
The rule applied but its condition was false, so it did not.
The rule did not concern this task or this event.
There is no ordering and no "first match wins". Several rules can act on the same task in one check, and a rule is never skipped because another one matched first.
Kolleno ships a few built-in rules for promises to pay, for example moving a task back to In progress when a promise is missed. Your own rules always outrank them: if yours covers the same situation and says something different, yours wins.
When the AI skips a task without checking
Checking a full playbook costs time, so the AI avoids it when there is clearly nothing to check. Each rule is tagged with the kinds of event it cares about, worked out automatically from how the rule is written.
If a promise to pay changes or a task's status changes, and none of your rules, category or company-wide, say anything about that kind of event, the AI skips the task rather than checking and finding nothing.
This shortcut applies only to those two events. New activity landing on the task, task creation, balance changes, reminders, due dates and manual reruns are always checked in full. A rule that has not been tagged yet is treated as relevant, so nothing is quietly ignored while it waits.
What the AI remembers between checks
If a task is checked more than once, the AI remembers why the task was created and what it decided last time. If it already tried something that did not work, it does not blindly try the same thing again, and it does not lose track of why the task exists.
Things to note
The AI never contacts your customer on its own. Anything that reaches the customer is drafted for a person to review and send.
Checks run in the background. Expect the task to update within a few minutes of the event, not instantly.
You cannot yet see a rule-by-rule breakdown. The AI records which rules applied and why, but the task view shows only its overall summary. A detailed view is planned.
A skipped task is not "the AI looked and found nothing." If no rule mentions promise or status changes, the AI does not run for that event at all.
Category rules and company-wide rules cannot ignore each other. Both always apply. A company-wide rule cannot exempt one category, and a category rule cannot override a company-wide one.
You do not choose which events a rule reacts to. The AI works that out from how the rule is written.