A.I. Data Writer

What is A.I. Data Writer?

A.I. Data Writer lets an A.I. Agent fill in one Custom Field for you. You describe in plain words what the value should be, choose which fields the Agent may read, and choose the one field it writes. Each time it runs, the Agent reads the current values of those fields and saves a new value in that field.

The value can be a number, a short line of text or a paragraph. A few examples:

  • A fit score from 0 to 100 for every new prospect, based on their budget, timing and the service they asked about.
  • A one-line next step for a Support ticket, based on what the client wrote.
  • A short project summary paragraph on a Target, written from their Kickoff answers.

A Data Writer is a kind of Skill. You build the Skill once, give it to an Agent, and then run it from an Automation or a Kickoff Form. If you are new to Agents and Skills, start with A.I. Agents & Skills.

IMPORTANT: A Data Writer only runs when your account has an active A.I. Service Subscription or a verified connection of your own under A.I. Connections - BYOK. Each run and each preview that reaches the A.I. model uses tokens.


How do I create a Data Writer Skill?

  1. Open Organization Settings from your Flyout Menu, then the A.I. Connections tab (or the A.I. Service tab if your account has the A.I. Service Subscription).
  2. In the Skills panel, click Create Skill.
  3. Choose Write a Custom Field, the card marked A.I. Data Writer, and continue.

The Write a Custom Field card, marked A.I. Data Writer, on the Create Skill screen

The editor has a few sections. Work through them top to bottom, then test the Skill before you publish it.

Title and instructions

Give the Skill a title you will recognize later, for example Risk Tolerance Score. Under Where It Can Run you can see that a Data Writer runs in two places: the A.I. Custom Field automation action and a Kickoff Form's Intelligent Automation.

In Data Writer Instructions, write a Prompt that describes the value you want, the way you would explain it to a colleague. Include any scoring weights here too: which answers count most, and what the low and high ends of a score mean. Say what to do when the information is missing, so the Agent reports that it cannot assess instead of guessing.

Title and Type and Data Writer Instructions sections of a Data Writer Skill, with a Prompt that scores risk tolerance from 1 to 10

Use Insert Field to point at a field by name instead of guessing what the A.I. might call it. The list only offers fields you have turned on in Data Sharing (below). Every run also knows the current date and time in your account's timezone, so a prompt such as "how many days until their launch date" works.

Destination: the field it writes

  • Record Context: the kind of record the Skill works on, for example a CRM Target, a Company, a Project, a Support ticket, a Work Request or an Invoice.
  • Field to Update: the one Custom Field the Skill writes. It must be a number, text or paragraph field on that kind of record.
  • Minimum Value and Maximum Value (number fields): the range a result must fall in. Turn Allow Decimals off to accept whole numbers only.
  • Maximum Characters (text and paragraph fields): the longest answer the Skill may save.
  • Overwrite Existing Value: on by default, so each run refreshes the field. Turn it off and the Skill only fills the field when it is empty, and leaves any value already there alone.

Destination section with Record Context, Field to Update, Minimum Value, Maximum Value, Allow Decimals and Overwrite Existing Value

A result that breaks these rules is never saved, and the field keeps its current value. A number outside your range is not rounded or nudged into it.

Data Sharing: the fields it may read

In Data Sharing, turn on each field the Skill may read. These are the only facts the Agent sees, and each run reads their values as they are at that moment. Each section names the record that owns its fields, so alongside the record's own fields you can also share fields from a linked record, such as the Company a Contact belongs to. A Skill can read up to 100 fields.

Preview Data Writer panel above the Data Sharing section, with four Target Custom Fields turned on

TIP: Share only the fields your Prompt actually needs. Fewer, clearer facts give steadier results, and they keep information the Agent has no use for out of the request.

A.I. Execution

A.I. Execution chooses which provider and model run this Skill. Leave it on your default provider, or pin the Skill to a specific provider and model.


How do I test a Skill before I publish it?

Use Preview Data Writer, at the top of the editor's right-hand side. For a CRM record context, pick a Sample CRM Target; for other record contexts, such as a Project or an Invoice, enter a Sample record ID. Then click Run Test.

The preview reads that record's shared fields, runs your Prompt and shows the result. It never changes the field and does not start any automation. It shows one of four outcomes:

  • Proposed Value: the value the Skill would save, with the reason it gave.
  • Unable to Assess: the shared fields did not hold enough information. Check that the right fields are turned on and that the sample record has values in them.
  • Invalid Result: the answer broke one of your Destination rules, for example a number outside your range or text that is too long.
  • A.I. Provider Unavailable: the A.I. provider could not be reached. Try again, or check your A.I. Connections.

When you are happy, click Publish. Save as Draft keeps your work without making the Skill available. Only published Skills can be given to an Agent and chosen in an automation or a form.


How do I put a Data Writer to work?

First give the published Skill to an Agent: open the Agent from your Staff list and assign the Skill under its Skills. See A.I. Agents & Skills. Then choose where it runs.

From an Automation: the A.I. Custom Field action

In any automation, add the A.I. Custom Field action from Automation Actions, choose the Agent, then the Skill. Each action runs exactly one Data Writer Skill. The automation's trigger must identify the kind of record the Skill writes, so a Skill that writes a Target field is offered where the trigger is about a Target. See AUTOMATIONS: No-Code Automations for how to add actions.

An A.I. Custom Field step in an automation, with an Agent selected and one Data Writer Skill selected

An account owner or administrator must save an automation that holds this action. Each run works with that person's access, so it can only read and write records they are allowed to see.

The A.I. takes a few moments, so the automation does not stop to wait for it. Its other actions carry on straight away. When a later action needs the new value, for example an email that quotes the score, turn on Wait for the A.I. Custom Field result on that later action. It then runs only after the A.I. Custom Field action saves a new value, and it is skipped if no value is saved. Actions without the switch carry on as normal. The switch needs the A.I. Custom Field action to come earlier in the same automation.

TIP: You can also add an A.I. Custom Field step to an Auto-Template, and it runs whenever the template is applied. Inside a template, though, Wait for the A.I. Custom Field result has no effect: the template's later steps run straight away, before the new value is saved. When a step must use the new value, put that step in an automation instead.

From a Kickoff Form: Intelligent Automation

In a Kickoff Form's settings, open Intelligent Automation, pick the Agent, then tick the Data Writer under Skills. Every submission then runs it for the record the form creates or updates. A form runs at most one Data Writer Skill, and it only offers Skills whose record the form can identify. See FORMS: Kickoff Forms.

A Kickoff Form's Intelligent Automation settings, with an Agent selected and one Data Writer Skill ticked under Skills

The person filling in the form never sees any of this. Their submission always goes through as normal, even if the Agent cannot produce a value.


What happens during a run?

A Data Writer is built to leave your data alone whenever it is not sure:

  • It writes one field on the record that started the run, and nothing else. Nothing in a form answer or in the A.I.'s reply can point it at a different record or field.
  • It only saves a value that passes your Destination rules. When it cannot assess, gets an unusable answer, or cannot reach the A.I. provider, the field keeps its current value.
  • If a staff member edits the field while the Skill is working, the staff edit wins and the Skill's older answer is not saved. The same happens when the shared fields, the Skill or the automation change during the run.
  • It runs only when an automation or a form asks for it. Changing a field does not start a run on its own, and a value the Skill saves cannot set off the same Skill again in a loop.
  • Each account has a daily safety limit on Data Writer runs. A run past the limit is held back and says so in Run History.

Saving a value is an ordinary field change, so any automation that watches that field runs as usual.


How do I see what a Data Writer did?

Click the Run History button on the Skill's row to open its Data Writer runs. Each run shows the record, when it ran and how it ended:

  • Saved: a new value was saved.
  • Unable to assess: the shared fields did not hold enough information. The field was left as it was.
  • Invalid answer or Provider failed: no usable answer came back. The field was left as it was.
  • Withheld: something changed during the run, such as a newer staff edit or a change to the Skill, so the older answer was not saved. The run says why.

Open a run to see the value and the reason the Agent gave, plus Selected evidence at assessment time: the field values it read for that run. A run started by an automation also lists the Automation Actions that were waiting for it and whether each one completed or was skipped. Run details are cleared after 90 days; the outcome itself stays in the list.

An opened Data Writer run showing the saved paragraph, the reason, and Selected evidence at assessment time


Why does my Skill ask me to Confirm Evidence Review?

When a field your Skill reads has changed or is no longer available, the editor asks you to check the Data Sharing list and click Confirm Evidence Review before you can publish again. Some older Skills read answers saved with a form submission instead of current fields. When you open one of those, a notice explains that saving replaces those inputs with the current fields you choose in Data Sharing. Turn on the fields you need, confirm your review, and publish.


Related articles