AI voice and multi-channel survey capture for research study data collection, proven as a proof of concept

The challenge

Research studies depend on the data participants give back. Collection normally runs through a form, a portal, or a scheduled call with a coordinator. Every one of those routes is a single fixed channel, and coordinator time puts a hard ceiling on how many calls get made.

Visual observation is a separate problem. Where a study wants a photograph of the subject, the colour a phone camera records shifts with lighting, handset and camera settings. Two photographs of the same subject taken in different rooms are not comparable, which limits what can be done with the image afterwards.

The question was whether both could be addressed in one system, before anyone committed to a production build. The work was scoped as a proof of concept: prove the mechanism, not ship a product.

What we built

A working proof of concept covering three capture channels and one calibration technique.

• AI voice calling. An automated voice agent runs the survey conversationally over a phone call and captures the participant’s answers as survey data, with no coordinator on the line.
• Web survey interface. A browser route into the survey for participants who prefer to type.
• Mobile application. A phone application for survey completion away from a desk.
• Photographic capture with a colour reference card. The mobile application guides the user to photograph the subject with a printed colour reference card in frame. The card carries patches of known colour, which gives the software a fixed reference to correct the captured image for the lighting and the camera that produced it. That correction is what makes colour measurement from a phone camera usable at all.

Participants could be reached through whichever channel suited them rather than the one the study happened to offer.

The proof of concept makes no diagnostic, clinical or regulatory claim. It was not built or submitted as a medical device. It demonstrated the capture and calibration mechanism, nothing further.

How we worked

Scope was set by what had to be proven and stopped there. That is why the build spans three capture channels and a calibrated image path at demonstration depth, rather than taking any one of them to production depth. Each path was taken to the point where it could be run end to end in front of an audience.

Outcomes

• Demonstrated that a research survey can be administered by an automated voice call and returned as structured data, without a coordinator running the call.
• Demonstrated the same survey delivered across voice, web and mobile, so channel choice sits with the participant rather than with the study.
• Demonstrated calibrated image capture on ordinary phone hardware, using a printed colour reference card in frame in place of uncontrolled photography.
• Established internal capability in AI voice interaction, cross-platform survey delivery and camera colour calibration, now available to client work.
• No measured results are claimed. This was a feasibility demonstration. No pilot metrics, participant numbers or accuracy figures were produced, and none are asserted here.