Product · Beta
Ambient Qual: interviews at the moment of decision
Ambient Qual drops Sofi into your site with a single script tag. When a shopper tries to leave checkout, lands from a paid ad, or finishes a purchase, Sofi can interview them right there, no calendar invite, no panel recruit. You hear the reasoning at the exact moment it happens, not reconstructed weeks later.
What does in-flow qualitative actually mean?
Traditional qualitative programmes are project-shaped: scope, recruit, schedule, interview, analyse. That rhythm is powerful for big questions, but it leaves quiet weeks while shopper behaviour keeps moving.
Ambient Qual means interviews fire from what shoppers do on your site, so learning continues between those larger projects. Growth and merchandising teams hear language and reasoning in context; research teams still own methodology, consent, and study quality.
How does implementation work?
- Add the site in Fieldwork and copy the publishable key.
- Paste the tag into your page head or your tag manager, with a trigger (
exit-intent,time, orcustom). The site confirms itself once the tag loads, and tells you which addresses it appeared on. - Open the study you want answered, find Where this runs, and name the moment that should open it.
- Optionally set
window.FieldworkTraitsto pass the shopper id and traits. - Transcripts, themes, and topic coverage land on that study in your Fieldwork workspace, alongside anything you fielded over a link.
How does a moment know which study to open?
Your product declares that something happened, by name. Fieldwork decides whether that is worth an interview: the study has to be live, the request has to come from a site you have named, and the person has to be someone you have not spoken to recently.
The name is declared on the study that answers it, so it cannot quietly point at two. One person is interviewed at most once every seven days per site, enforced on our side against a hashed id rather than only in the browser.
Nothing fires inside a cancellation or refund flow. Asking someone for their time at the moment they are trying to leave is how a research tool turns into a dark pattern.
What is Ambient Qual not meant to replace?
It does not replace moderated sessions when you need deep exploration, sensitive facilitation, or relationship-heavy contexts. It complements analytics by adding why, not by substituting for behavioural measurement.
It is strongest for cart abandonment, post-purchase depth, ad-landing expectation checks, and adoption mysteries, places where timing and context matter as much as the questions you ask.
What does an in-flow interview look like in practice?
A DTC cookware brand wants to understand checkout abandonment: what stopped the shopper, whether shipping framing was the issue, and what would have made the order worth finishing. The loader fires an interview when exit-intent hits the cart page.
The brief
"Understand what shoppers were trying to accomplish at checkout, what stopped them from completing, and what framing would have tipped them into the order."
Example conversation · not live data
In-flow prompt
Looks like this cart is not quite ready, got 90 seconds to tell us what is making you hesitate?
What were you hoping to pull off with this order, and what stopped you short of finishing?
I wanted the full set, but shipping bumped it past what I planned to spend tonight.
What would shipping need to look like for you to finish this order right now?
Sofi · following the thread
Transcripts and topic coverage land in the workspace automatically, on the same study you would read if you had fielded it over a link.
Frequently asked questions
What does "ambient" mean in Ambient Qual?
Ambient means interviews surface in the flow of normal site use, triggered by behaviour rather than scheduled by a researcher. A shopper who just abandoned checkout sees a short invitation to talk through what stopped them. Someone bouncing from a paid landing page can be asked what they expected. The research programme keeps running between major study cycles without coordinating each session.
Do we need to replace our existing analytics tools?
No. Ambient Qual adds qualitative depth, the language and reasons behind behaviour, alongside what analytics already measures. It tells you why shoppers did or did not convert, not only that they did or did not. Most teams run it next to GA4, Mixpanel, Amplitude, PostHog, or a warehouse pipeline.
Is this for growth teams or research teams?
Both. Growth and merchandising teams get in-context signal that helps interpret funnels and campaign performance. Researchers keep control of study design, consent, topic coverage standards, and how interviews are routed. Execution scales without turning experiments into an unmanaged mess of one-off conversations.
How is participant consent handled for in-flow interviews?
Before anything is asked, the participant is told that Sofi is an AI rather than a person, what is kept, that their answers are not used to train AI models, and how to leave. No interview turn runs until they accept. Fieldwork records that acceptance per session with a timestamp, study and locale, so the audit trail reflects what a specific person agreed to rather than a one-time banner click. The tag also holds off storing anything on the visitor device until your consent platform says it may, which it picks up automatically if you use the IAB framework.
What does integration look like?
A single script tag with a publishable key and a trigger, which today is exit-intent, time on page, or a custom event your own code fires. It goes in the page head or into your tag manager as a Custom HTML tag. The script iframes a hosted interview surface, so nothing in your DOM changes. There is no package to install. Native app support and a server-side trigger are on the roadmap.
How does a moment know which study to open?
You name a moment on the study that should answer it, for example abandoned-cart. Your product decides when that moment happens and reports the name; Fieldwork decides whether to interview, checking that the study is live, that this person has not been interviewed recently, and that the request came from a site you have named. Because a name is declared on one study, it cannot quietly point at two.
How often can the same person be interviewed?
Once a week at most. A seven-day cooldown per person per site is enforced on our side against a hashed id, not just in the browser, so clearing cookies or switching tabs does not get around it. Anyone who declines is not asked again in that window either.
Related: Fieldwork Interviews, ResearchOps, and pricing.