Data hygiene: why AI agents fail on messy data
Data hygiene means keeping your business records complete, up to date and free of double entries. AI agents, programs that carry out tasks for you, fail on messy data because they take every entry as true. A person who sees an odd entry asks a colleague. An agent accepts it and carries on.
Why people cope with mess and agents do not
Every team knows the weak spots in its own records. Everyone knows that the dates on old offers mean nothing. Everyone knows that a certain customer is in the list twice.
Think of a new temp on the first day. Nobody has told them any of this. They read the customer list and believe every line. An agent is in the same position, every day.
It reads the old date and one of the two entries. Then it builds a reminder, a report or a reply on top. The result looks tidy, which makes the mistake hard to spot.
Four kinds of mess, and what an agent makes of each
| The mess | Example | What the agent does |
|---|---|---|
| Empty field | A task with nobody responsible | Leaves it out of every summary. The work becomes invisible. |
| Old value | An offer still marked "decision in March" in June | Counts it as expected income |
| Double entry | The same customer entered twice | Reads half the history and reports "no contact for months" |
| Free wording | Status written as "almost done", "90%" and "nearly" | Treats them as three different things |
Why forcing people to fill in everything does not solve it
The obvious fix is to make every important field compulsory. This tends to backfire. People who must fill in ten fields before they can save will type anything to get past the form.
A wrong value is worse than an empty one. An empty field can be found. An invented one looks like real information.
It works better to let people save quickly and follow up later. An entry may be incomplete for a few days. After that it shows up as an open point for a named person.
A routine in four steps
- Choose the fields that matter. List the ten or so details that reports, reminders and agents really use. Examples are the responsible person, the due date, the status, the amount and the customer.
- Write one short rule for each. "Every open task has a due date." "Every open offer has an expected decision date."
- Decide who fixes what. For a missing due date, that is the person doing the task. For a task with nobody responsible, it is the person who created it.
- Check regularly, and show each person their own gaps. A company-wide list of four hundred problems gets ignored. Seven entries under your own name get fixed.
If the same rule is broken every week, look for the cause. Perhaps the form asks at the wrong moment.
Clean up before or after you start with agents?
Both. Before an agent starts, clean the fields from step one for the records still in use. Leave old, closed records alone. Then let the routine keep things clean.
An agent can help with the clean-up too. It should only suggest a value where the record supports it, and a person should confirm each one.
How Worknano handles this
Worknano has this routine built in. Admins write data hygiene rules that say what must be filled in, how important it is, how many days of grace a record gets and who is accountable for the fix. New workspaces start with 13 default rules for tasks, projects, OKRs and CRM. Each user sees their own gaps and a hygiene score from 0 to 100 on their My page, and a daily summary tells them how many records need attention. The Data Hygiene Assistant then works through one rule at a time. It suggests a value only when the record supports it, and saves nothing until you confirm.
- Data hygiene: rules that keep important fields filled
- Data Hygiene Assistant: finds missing data and helps you fix it
Create a free workspace and see your hygiene score before your first agent runs.