Reviewing uploaded files with OnRamp AI

Last updated: October 30, 2026

Describe what a good file looks like on a file request, and OnRamp AI checks every file a customer uploads against it. The customer sees right away what looks good and what is missing, and your team sees the same review next to the file.


How file review works

When you ask a customer for a file, you usually have something specific in mind: a certificate naming your company as holder, a user list with certain columns, a contract with a signature. File review lets you write that down once, on the request itself.

Each time a customer uploads a file to that request, OnRamp AI reads it and compares it with your description. Within a few seconds the customer sees a short review under the file: what matches, what is missing, and anything else worth adding. The review is a guide. Customers can always continue, and your team reviews the file as usual.


Set up a file request for review

File review is set on the Request Files element in a task's form, in the playbook's task builder or on a project's task.

  1. Open the task and select the Request Files element.

  2. Under Properties, fill in What does a good file look like?

  3. Optionally, add an Example of a good file. OnRamp AI compares each upload with it as well as with your description. Up to two example files are used.

  4. If you want customers to see the example, turn on Let customers download the example. It is off by default.

  5. Save the task.

A request with a description, an example, or both is reviewed. A request with neither works exactly as before.

Writing a description that works

  • List what must be there. "Our company as certificate holder, at least $1M general liability, and your legal entity as named insured" gives OnRamp AI three things to look for.

  • Name the columns or fields you need. For a spreadsheet, say which columns matter, for example "one row per user with Email, First name, Last name and Role."

  • Say what makes a file current. If an expiry date or a signature matters, include it.

  • Keep it to what you would check yourself. OnRamp AI judges only against what you wrote, so anything you leave out isn't checked.

Keeping the example private

An example helps even when customers never see it: OnRamp AI uses it to understand the shape of a good file. While Let customers download the example is off, customers don't receive the example, and OnRamp AI never quotes or describes it in what it tells them. If you do share it, customers see See an example of a good file above the upload area, so use a sample with no real customer data.

Drafting the description from your example

In a playbook's task builder, once you attach an example, Draft with Aero appears under it. Choose Draft description and Aero reads the example and suggests a description of what any good file of that kind must contain. It describes the requirements, not your example's own names, numbers or dates.

  • Choose Use this description to fill in the description, or Replace the description if you already wrote one.

  • Discard leaves your description as it is. Nothing changes until you choose.

  • Edit the draft like any description before you save. Each person can draft up to 20 descriptions an hour.


Try it before customers do

In a playbook's task builder, the Request Files element has Try it with a file. Choose Try with a file, pick a sample, and you see exactly what a customer would see. The check uses your description and examples as they are on screen, so you can try changes before saving.

  • The sample file isn't saved, and nothing is shown to customers.

  • If you edit the description or the example after a check, the result is marked as possibly out of date, and you can check the same file again.

  • Each person can try up to 20 files an hour.


What customers see

The review appears under the uploaded file, with a heading that says how the file did.

Heading

What it means

This looks like what we need

The file has what your description asks for.

Almost there. One thing is missing

The file is close, and the review lists what to add or fix. With more than one gap it says how many.

This doesn't look like the file we asked for

The file seems to be a different document altogether.

OnRamp AI couldn't review this file

The file type can't be read, the file is too large, or the content is unreadable, such as a blurry scan. The customer is told nothing more is needed.

OnRamp AI couldn't check this file

The check didn't run this time. The customer is told nothing is needed and that your team will review the file.

Under the heading, the review groups its findings as What's missing, What looks good and Worth adding, with short quotes from the file where they help. A line underneath reminds the customer that the review is a guide and that they can upload a corrected file to check it again.

Before you continue

If a customer goes to complete a task while one of its files is missing something, they see Before you continue, saying OnRamp AI found something missing. They can choose Go back to upload a corrected file, or Continue to complete the task anyway.

Asking OnRamp AI about a file

If your customer portal includes OnRamp AI chat, a review that found something missing also offers Ask OnRamp AI about this. It opens the chat with the customer's question about that file, and OnRamp AI answers using what the review found, for example explaining a missing item or suggesting where to find it.


What your team sees

On the task

In the project's task, each reviewed file shows the review under it, labelled What the customer was told with the date it was reviewed. It is the same review the customer saw.

On the Files and Responses tabs

On the project's Files tab, a File review column shows a status for each file on a reviewed request. The same status appears next to each file on the Responses tab. Select a status to see the full review.

Status

What it means

Looks good

The file matched the description.

Missing 1 thing

The review found gaps, and the number says how many.

Not what was asked

The file seems to be a different document.

Couldn't review

The file type, size or content couldn't be read.

Couldn't check

The check didn't run for this upload.

Not reviewed

The file has no review, for example because it was uploaded before the request had a description.


Files OnRamp AI can read

File type

What OnRamp AI reads

PDF

The text of each page, plus images of the first pages, so scanned forms and signatures are seen too.

Images

PNG, JPG, GIF and WebP pictures, such as a photo of a document.

Excel (.xlsx, .xlsm)

Each sheet's columns, how many rows it has, and its first rows. Macros in a workbook never run.

CSV

The columns, the row count and the first rows.

Text (.txt, .md)

The full text.

Files up to 20 MB are reviewed. Any other file uploads as usual, and the customer sees that OnRamp AI couldn't review it.


Common questions

Can a customer complete a task when their file is missing something?

Yes. A review is always a guide. Customers see Before you continue first, and can continue or go back to upload a corrected file.

What happens if a customer uploads the same file again?

They see the same review straight away, without a new check. A different file, or the same file after you change the description or example, is checked fresh.

I changed the description. Do earlier uploads get checked again?

Earlier reviews stay as they were. New uploads are checked against the new description. When you update a playbook, projects where the customer hasn't started that task pick up the new description.

How do I review a Word document or another file type?

Customers can still upload it, and your team reviews it as usual. To get an automatic review, ask for the file as a PDF or, for tables, as Excel or CSV.

How often can a file request be checked?

Each request can be checked up to 25 times a day. After that, uploads still go through and show Couldn't check until the next day.

Can customers see my description?

Not as written. Customers see the review of their file, which says what matched and what is missing. Use the task's own instructions for anything you want customers to read before they upload.