What to do when an AI employee finds nothing

When an AI employee finds nothing, first look at which block the items stopped at. Every run shows how many items each block passed on, so you can see whether the search itself came back empty, or whether a later block such as Only new or a Filter let nothing through. An empty search usually means the area, the type of place or the data's coverage is wrong. An empty result after Only new usually means nothing new happened, which can be the right answer.

This is for anyone whose flow ran, passed, and sent an empty email or no email at all. It explains the common reasons, how to read a run to find which one it is, a made-up example, the fixes that cause new problems, and how to set up an AI employee so that an empty run is either expected or flagged.

Empty is not the same as failed

A failed run stops with an error and says what went wrong; that is covered in when an AI employee fails. A run that finds nothing passes. Every block did its job; there was simply nothing to pass on at some point. Sometimes that is good news: no road closures near your depots this morning, no new competitor near your stores this week. The question is whether the empty result is true or the flow is not looking properly.

How to read the run

Open the run from the employee's card or the flow's history. On the canvas each block carries a count of the items it passed on, and the same number sits on the link it leaves by. Read them from left to right:

• The data block says 0. The search found nothing. Check where and what it searched.
• The data block found items, a later block says 0. Something after it removed everything. Click that block to see what came in and what it kept.
• Nothing reached a block at all. A block with no link from a source does not run. The canvas marks it with Nothing reaches it.

Clicking a block in the run panel shows the fields that came out of it and the first items, so you can check what an item actually holds before blaming the filter.

The common reasons, and what to change

The usual causes in order of how often they come up in practice.

Blocks that watch for change behave differently on their first run. Web page watch and When something changes only remember what they see the first time; the next run compares against it. What changed passes everything on the first look and marks it as a first look. So a watcher that sends nothing the morning after you set it up is working as designed.

A worked example

A made-up case: an agency in Cape Town set up an AI employee to find new yoga studios in three suburbs every morning and email them to a client. For six days in a row the email said there was nothing new.

1. The manager opens the last run. Nearby places found 41 studios; Only new passed on 0.
2. That means the search works and every studio had been sent before. The question is whether six empty days is believable for three suburbs.
3. She opens the Area of interest: the three suburbs are drawn as small circles of about 1 km each. New studios just outside them are never searched.
4. She widens the circles to cover the whole suburbs, and adds the group of fitness types rather than one type.
5. The next morning's run finds 63 places; Only new passes on the 22 it had never seen, because the wider area brought in places outside the old circles. From then on the daily count settles at zero to three.

The fix was in the area, not in Only new. Removing Only new would have filled the email, but with the same 41 studios every day.

Fixes that cause new problems

• Removing Only new. The email fills up again with yesterday's results. Keep it and widen the search instead.
• Loosening every filter at once. Change one rule at a time and rerun, or you will not know which one mattered.
• Running it more often. If the market is quiet, a run every hour finds nothing more than a run every morning.
• Switching to a block that has sample data where live data is missing. Sample figures are marked as sample; they are not a fix.

How to set up an AI employee for empty days

Decide in advance what an empty day should look like for this employee, then tell it:

1. For an employee that should rarely be empty (new places in a busy city), set an alert under the flow's alerts and exports for Found nothing new, so you hear about it.
2. For an employee that is usually empty (road closures near five depots), leave the alert off. Put a Filter or Branch before the Email so a message goes out only when there is something to say.
3. Read the employee's daily note in the Team inbox: it says how many runs it made and what it found, so a run of empty days is visible at a glance.
4. When you are unsure why it is empty, ask Ask xMap in the editor's chat: it can read the last run and suggest what to change.

Templates such as New places every morning, only new expect a few results most days; Road closures in my city and Rain alert for my city are built to be quiet most of the time.