How an AI employee decides when to run: timetables, changes and messages
An AI employee starts on a timetable, when something changes near your places, when a row is added, or when someone writes to it. How to pick the start that fits.
Part of the guide AI employees
An AI employee in xMap Flows does not decide by itself when to work. You give it one start, called a trigger: a timetable, a change near your places, a new row in a table, a message in its chat, an event in a connected app, or another AI employee handing it work. It runs every time that start fires and at no other time. Choosing the right start is most of the difference between an employee that sends useful news and one that sends the same list every morning or misses the thing you cared about.
This is for people setting up their first few AI employees and for teams whose inboxes are filling with messages nobody reads. It covers each kind of start, which work suits it, a made-up example of changing one, the mistakes, and how to keep a timetable from becoming noise.
The starts an AI employee can have
| Start | It runs when | Suits |
|---|---|---|
| Schedule | The time comes: every 5, 15 or 30 minutes, hourly, daily, several times a day, weekly, monthly, a cron line, or once | Morning lists, weekly reports, monthly reviews |
| When something changes | A check finds a new place of a kind nearby, a rating falling under a line, a place closing, or a road closed near a site | Alerts that matter only when something happens |
| New rows | Rows are added to a table in My data or a sheet in Google Sheets | Screening addresses, enriching leads as they arrive |
| Chat | Someone writes in its chat window or shared chat page | Answering questions on demand |
| App events | Something happens in a connected app, such as a new message | Replying and acting inside your tools |
| Run another flow (in the other employee) | Another AI employee hands it items | Two-step work split between employees |
Each start fires the whole flow; the blocks after it decide what to do with what it brings.
A timetable: when it is the right start
A timetable suits work that is useful even when nothing dramatic happened: a ranked list of new prospects every weekday morning, a weekly brief, a monthly check of every site. The Schedule block runs in the time zone you pick, can skip public holidays of 51 countries and dates you list, and for runs every few minutes or hourly can keep to working hours on chosen days. Before you deploy it shows the next three runs, so you can see exactly when it will wake up.
A change: when waiting is better than checking
When something changes is for work that should happen only when the world moves: a rival opens within 1 km of one of your shops, a nearby business's rating drops below 4, a road near your depot is closed. xMap checks on the timetable you choose, every 15 minutes, every hour, every 6 hours or once a day, and runs the employee only when a check finds a change, with the changed places as its items. The first check only looks around; changes show from the next one. Pausing the employee stops the checks.
The practical difference: a daily timetable runs the whole flow every day, while a change start runs the cheap check every day and the flow only on the days something happened.
A worked example (made up)
Say a bakery group in Montreal set up an AI employee on a daily timetable at 07:00: search bakeries within 1 km of its six shops, score each with xMap LLM, email the list. After two weeks the manager stops opening the email, because it is the same 40 bakeries every morning.
- First fix: add Only new after the search. Now the email lists only bakeries not seen before, and most mornings it is empty and not sent.
- Second fix: move to When something changes, watching for a new bakery within 1 km of the six shops, checked once a day.
- Result: the employee runs on the two or three days a month a bakery appears, and the manager reads every message because each one is news.
The numbers are made up; the pattern is common. Most noisy employees have the wrong start, not the wrong blocks.
Mistakes in choosing a start
- Every 5 minutes for daily news. Frequent runs suit ships near a berth or road closures in the morning rush, not new shops.
- A timetable without Only new. The same results every run teach people to ignore the employee.
- The wrong time zone. A 07:00 list in another zone arrives in the middle of the night.
- A change start for work that needs a regular report. If the board wants a Monday summary even when nothing changed, use a weekly timetable.
- Forgetting holidays. Skip public holidays so nobody is messaged on a day off.
Setting it up
- Ask what the person who reads the result needs: news when it happens, or a regular summary.
- For news, start with When something changes. For a summary, start with Schedule and add Only new where repeats would be noise.
- For lists that people add to, start with New rows; for questions, with Chat.
- Run it once from the editor and read the message as its reader would.
- Deploy. The employee's card on the Flows page says when it works and when it runs next.
What lands in the inbox then follows the start: a daily or weekly email at the time you picked, an alert only on the day something changed, or an answer the moment someone asks.
Questions
Can an AI employee have more than one start?
A flow can hold more than one trigger, and each run comes in through the one that fired. For clarity, most teams give each employee one start and make a second employee for a second kind of work.
Does a change start cost anything on quiet days?
Each check is a small search around your places. The rest of the flow runs, and is paid for, only when a check finds a change.
What happens on public holidays?
If you tick skip, a scheduled run due on a holiday is skipped and the next one runs as usual.
Why did my employee not run?
Check that it is deployed and not paused, the time zone of its schedule, and for a change start whether anything changed: the first check only looks.
Templates that do this
Read next
More on ai employees and automation · Every article · Templates