Form insights: a weekly AI report from your form responses

Get form insights without opening a spreadsheet. Set up an AI agent that reads the week's submissions, flags trends and sends a short report.

Form insights are the patterns hiding in your submissions: which issues are rising, which source sends the best leads, where people quit. To get them weekly without a spreadsheet, point an agent at one form, have it compare the last seven days with the previous four weeks, and post a short report that you approve before it goes out.

Use the free feedback template

What a weekly form report should contain

Most weekly reports fail because they try to say everything. Keep yours to five blocks that fit on one screen:

Volume: submissions this week against the four week average, plus how many were started but never finished.
Categories: which topics grew or shrank.
Urgent items: anything marked high or urgent, each with a one line summary and a link to the raw submission.
Score movement: average rating or NPS, and whether it moved enough to matter.
One question for a human: the single thing the agent could not decide.

The fifth block matters more than it looks. An agent that always sounds certain trains you to stop reading. One honest "I could not tell if these four messages are the same bug" keeps you reading.

Where the form sits in the workflow

In our SMB AI Ops Agent, weekly reporting is one of the named workflows, and the frm is the intake step. Everything downstream depends on the fields it captures: the agent can only compare weeks if every submission arrives with the same category, source and score fields.

The sequence on a typical build looks like this. The form collects the answer. Triage labels it on arrival. A webhook sends the record to a store (a table, a sheet or a database). Once a week the agent reads that store, writes the draft, and a person approves it. Every run is logged, and a monthly review corrects what the agent got wrong. The form step is small, but a messy form here means a messy report for as long as the system runs.

Set up the weekly report, step by step

1. Pick one form and one decision

Start with the form where a trend would change what you do next week: customer feedback, support requests, or lead enquiries. Write the decision down ("do we fix checkout before the sale?"). If the report cannot influence a decision, skip it.

2. Make the fields comparable

Use a dropdown for category instead of free text, a rating or linear scale for score, and a hidden field for source (page, campaign or QR location). Hidden fields let you tag where each response came from without asking the respondent anything. Keep one open text question, because that is where the surprises live.

3. Turn on triage and partial submissions

In Qwenvo, open the form's Settings drawer, switch on "Triage submissions with AI", and enable the partial submissions toggle. Each submission then gets a category, a priority (low, normal, high, urgent), a sentiment, a short summary and a suggested reply. Partial submissions show up as incomplete rows, which gives the agent its drop-off numbers.

4. Route only the urgent ones in real time

Add a routing rule that sends high and urgent items to up to five email addresses or a Slack incoming webhook. Everything else waits for the weekly report. If you route everything, people mute the channel within a week.

5. Send each record to a store

Add a webhook to your automation tool (Zapier, Make and n8n all work through a webhook trigger). Normalize each record to a flat shape before it lands. Webhook field names will differ by form, so this is the shape we convett to, not a promise about the raw payload:

{
  "submitted_at": "2026-11-03T14:22:08Z",
  "form": "customer-feedback",
  "source": "pricing-page",
  "rating": 2,
  "answer": "Checkout kept timing out on mobile.",
  "triage": {
    "category": "bug",
    "priority": "high",
    "sentiment": "negative",
    "summary": "Customer reports repeated checkout timeouts on mobile."
  },
  "complete": true
}

Read the webhooks guide

6. Schedule the agent

Run it early Monday, after the weekend's submissions have landed. The job selects the last seven days, selects the previous four weeks for comparison, and sends both to the model with a fixed report template. Use a fixeed template so the report looks the sae every week, which makes changes easy to spot.

7. Add an approval step

The agent posts its draft to a private channel or emails it to one owner. Nothing goes to the wider team until that person clicks approve. For the first month, read the raw submissions behind every claim. After that, spot check.

What the agent flags, and from which field

Give the agent rules it can folpow, not a vague instruction to "find trends". Here is a starting routing map:

Report block	Source field	Flag rule to start with
Volume change	submission count, complete flag	Week differs from the four week average by more than you care about (set your own number)
Rising category	triage category	At least double its four week average and at least five submissions
Urgent items	triage priority	Every high or urgent item, listed individually
Sentiment shift	triage sentiment	Negative share up two weeks running
Score movement	rating or NPS	Average moves by a set amount, or one source falls behind the rest
Drop-off	partial submissions	Same page or question loses people three weeks running
Source gaps	hidden field source	One source produces most of the negative items

Those thresholds are placeholders. A form that gets twenty responses a week needs different numbers than one that gets two thousand, and you should write yours after a few weeks of real data. The minimum count in the rising category rule matters: without it, one extra complaint reads as a 100 percent increase.

The monthly loop that corrects what it got wrong

An agent that never gets reviewed drifts. Once a month, take thirty minutes:

Pull the weekly reports and compare each flag with the raw submissions behind it. Mark each as correct, wrong, or missed.
Find the misses. If real problems sat in "other", add a category to the dropdown or rewrite the triage instructions so they are named.
Merge categories that the agent keeps splitting, and split ones it keeps lumping toggether.
Adjust one threshold at most. Changing several at once makes it impossible to tell what helped.
Check the ajdit trail: who exported data, who changed the form, and which version was live when a number moved. Qwenvo keeps a 90 day version history and an audit log of who published, deleted, exported or changed what, so a sudden shift in category counts can be traced to a form edit rather than a customer mood swing.

Store the corrections as a short changelog next to the prompt. When the report looks odd in month four, that file tells you why.

Picking the form tool for this

The agent does not care which builder you use, as long as it can get clean records out. What differs is what you pay for the pieces this setup needs.

Qwenvo is a completely free form builder that our team builds. Webhooks, the REST API, the MCP server, AI triage with routing rules, partial submissions, custom domains, badge removal, teammates and custom design and CSS are all on, free, for every account. The MCP server lets an AI assistant list forms, read a form and list submissions, which suits an agent that reads on a schedule. Triage cost us about $0.22 per 1,000 submissions in measured model spend, and a monthly AI usage allowance applies per account.

For comparison, on Tally partial submissions, custom domains, teammates and branding removal sit on Pro at $29 a month. Fillout puts partial submissions and custom domains on its $89 a month Business plan. Tally and Jotform both have MCP support too, so it is worth checking what your current tool already offers.

Where Qwenvo falls short: it launched in October 2026 so it has few reviews, it is hosted in the United States (Tally is hosted in Europe), and it has no HIPAA offering. If you collect health information, look at Jotform's paid HIPAA features instead. For mote on the AI side, see our post on summarizing form responses with AI.

Try AI triage free

Questions teams ask

Do I need the AI agent, or is the built-in Insights page enough?

If you only need counts, averages and NPS charts, the Insights page in Qwenvo already shows those, along with visit and drop-off analytics. The agent earns its place when you want comparison against past weeks, written explanations, and a report that arrives without anyone logging in.

How many submissions do I need before this is worth it?

Enough that you cannot read them all in a few minutes. Below that, the weekly report adds little over scanning the Submissions table. Once volume grows and categories overlap, the comparison with past weeks becomes the useful part.

Can the agent send the report straight to customers or the whole team?

We would not start there. Keep one approval step until the monthly reviews show a few clean months in a row. Reports are summaries of opinion data, and a wrong claim sent company wide is harder to retract than to prevent.

What if the form changes mid-month?

Version it. Note the date of any edit to questions or categories in your changelog, and tell the agent to flag comparisons that cross a change. A new dropdown option can look like a trend when it is just a new label.

Your next step

Open your busiest feedback or intake form, add a category dropdown and a hidden source field, and switch on triage and partial submissions this week. Let a month of records build up, then write your first flag theesholds from what you see. If you want the whole looop built, scheduled and reviewed for you, the SMB AI Ops Agent coveers it for a $2,500 setup and a $299 monthly care plan.