# How it learns — Formwink

> The whole loop on one worked example: a business profile you confirm, a form built from it, predicted fit that a human corrects, and a suggestion that is labelled a suggestion until it is tested.

Source: https://formwink.ai/how-it-learns

How it learns

# Your form should learn what matters to your business.

Here is how that works for one sample business: Loomwork, a small AI and automation studio that helps growing businesses automate onboarding, support triage, and internal handoffs.

Understand

## It starts from your business.

Loomwork pastes its website. Formwink drafts a profile: what they automate, who they do it for, and what they explicitly do not take on. Every line is marked as confirmed, suggested from the website, or unknown — and a suggestion is not used until someone confirms it.

Business profile · loomwork.example Profile v1

✓

Services · Workflows we automate

Customer onboarding, support requests, internal handoffs

Website suggestion · Confirmed

✓

Delivery capabilities · Tools we work with

HubSpot, Notion, Slack, Zendesk, Airtable, Google Workspace

Customer-provided · Confirmed

✓

Commercial fit · Repeatable workflow

Happens weekly or more

Customer-provided · Confirmed

?

Buyer role · Operations or CS owner involved

Useful, may stay unknown

Customer-provided · Unknown allowed

?

Commercial fit · Budget

Optional scoping question; unknown allowed

Customer-provided · Unknown allowed

✓

Exclusions · Not a fit

Standalone consumer apps; custom software with no existing workflow

Customer-provided · Confirmed

Recommended form Form v1 · Draft

1

What would you like to take off your team’s plate?

Customer onboarding · Support requests · Internal handoffs · Something else · Not sure yet

2

A follow-up that fits the answer

Onboarding asks what happens after a customer signs. Support asks where requests arrive. Handoffs ask who approves.

3

Which tools are involved? How often does this happen?

Checked against the tools Loomwork delivers with. This describes the prospect’s stack — not a Formwink integration.

?

Budget (optional)

Unknown stays unknown. “Would like help scoping a first project” is a fine answer and does not disqualify anyone.

Ask

## Problem first, then the follow-up that fits.

The opening question is about the prospect’s problem, not your service menu. Each answer opens one branch rather than every field at once, and any field can explain why it is being asked.

- Required fields are the fit checks. Everything else is for the brief.
- Budget is optional scoping, never the qualification story.
- Editing a question creates a new version; the old one can be rolled back.

Recognize

## Predicted fit is where a human review starts.

Four inquiries, four different shapes. Note that the prediction is wrong in the last one — and that the correction is the most valuable event on this page.

✓

Harbor Analytics

Maya Chen

Onboarding handoff across HubSpot → Notion → Slack, about 60 new customers a month.

- Predicted: Strong service fit
- Reviewed: Qualified

Budget and access still to clarify. Service fit is not commercial qualification.

~

Fieldstone Advisors

Sam Okafor

“Too much manual work, but we don’t know where to start.”

- Predicted: Needs clarification
- Reviewed: Not reviewed

An unclear request is not an unqualified customer. One useful follow-up.

✕

Park & Co

Devin Park

Wants a standalone consumer mobile app.

- Predicted: Outside current services
- Reviewed: Not qualified

Kept for human review rather than discarded.

!

Weber Logistics

Jonas Weber

“Internal handoffs” that is really a dispatch-system rebuild.

- Predicted: Strong service fit
- Reviewed: Corrected: Not qualified

The owner corrected the prediction with one tap. That correction is what the learning uses.

Service fit and commercial qualification are separate questions. A strong service fit with an unknown budget is still a strong service fit.

Observe

## One denominator, held constant.

Last 90 days · Form v1 · Profile v1 · Sample data

200

### Saw the form

Unique eligible visitors — the denominator

100

### Started

50% of those who saw it

50

### Submitted

50% of those who started

30

### Reviewed qualified

15% of visitors — the number that counts

### The people who couldn’t name their problem

9 of the 50 people who submitted answered “Not sure yet.” A form that scored them down would have buried all 9.

After a conversation, Loomwork called 4 of them qualified — nearly half. Not knowing how to describe the problem turned out to say nothing about whether they were a good customer.

On the 32 we can’t see. Only 18 of the 50 have a recorded outcome, and 11 of those became meetings. The rest count as unknown, never as failures — a system that reads silence as “no” would learn the wrong lesson from every lead nobody got round to updating.

Improve

## A suggestion stays a suggestion until it is tested.

Formwink noticed the “Not sure yet” pattern in the step above. Here is what it does with it, and what it refuses to do.

What Formwink suggests ICP-informed suggestion · not tested

### Keep the option. Ask one more question.

Rather than dropping “Not sure yet” for being vague, follow it with a single question — on that path only:

“Which part of your process takes the most manual effort?”

Nothing else about the form changes.

What it does not do

✕

Change your live form

Approving this drafts a test. Your published form is untouched until you publish a new version yourself.

✕

Call itself a finding

Four out of nine is four out of nine. It may not hold. Until a test says otherwise this stays a suggestion, and the label says so.

✕

Declare a winner on its own

No experiment results yet. A test is judged against one number — qualified inquiries per visitor — and you make the call.

Four out of nine is a real observation and a weak basis for a claim. Both things are true, and the page says both.

## Keep exploring

[Product AI form builder Build forms around your ideal customer and learn from reviewed outcomes.](/ai-form-builder) [Product Lead qualification Understand which inquiries deserve attention, and why.](/lead-qualification-form-builder) [Solutions · For your workflow Client intake form Structure the first conversation so the right clients get a fast, informed reply.](/client-intake-form)

## What we hold ourselves to

Reviewed

You confirm or correct every fit assessment. No inquiry is rejected on a label.

Separated

Service fit and commercial qualification are different questions, kept apart.

Labelled

A suggestion says it is a suggestion. A test with no results says it has none.

Measured

Qualified conversion rate, not submissions. Improvement is measured, never claimed.

[Watch the 30 seconds tour](/#tour) [Join the beta](/beta)

&times;
