Interview Protocol
Semi-structured interview protocol for expert practitioner interviews. 15 questions across four sections. Mapped to RQ1, RQ2, and RQ3. Used with the voice interview tool at predictwhy.com/interview.
Opening - context and system overview
01
Could you briefly describe your current role and the AI system or product you work on?
Probe: What domain is the system in? What is the primary interaction it mediates between AI and user?
02
At a high level, how does your system currently track or respond to user behaviour?
Probe: What data does it capture? What does it do with that data in real time?
Signal recognition - RQ1
03
When you observe users interacting with your system, what behavioural signals do you find most informative for understanding their experience?
RQ1
Probe: Are these signals you actively track, or things you notice when reviewing session data manually?
04
Which behavioural signals do you currently track or observe - such as hesitation patterns, navigation switching, re-engagement after drop-off, help-seeking behaviour, or changes in interaction pace? Are there signals available in your data that your team does not currently use for personalisation or intervention?
RQ1
Probe: What prevents you from using those signals? Is it technical, organisational, or conceptual?
Signal interpretation - RQ2
05
When you see a user hesitate or pause during a task, what do you typically infer about their experience? How confident are you in that inference?
RQ2
Probe: Does your confidence change based on context - new vs returning user, task complexity, time of day?
06
Looking at the four ABSD layers - Action, Behaviour, State, Drive - does this structure map to how you think about user understanding in your work?
RQ2
Probe: Where does your system currently operate? Where does it stop? Is that a deliberate design decision or a technical limitation?
07
Can you give an example from your domain where a behavioural pattern clearly indicated an underlying psychological state? And an example where the same signal was ambiguous - where the same pattern could have meant different things?
RQ2
08
What contextual factors - domain, task complexity, user history - most affect how you interpret behavioural signals in your system?
RQ2
Probe: Is this contextual weighting built into your system, or does it happen in the analyst's head when reviewing data?
09
When your system makes an assumption about a user's state - such as inferring they are frustrated or confused - how do you currently validate whether that assumption was correct? Do you use any form of explicit user feedback?
RQ2
10
Moving beyond momentary emotional states, does your team ever attempt to design for or measure a user's underlying psychological drives - such as their need to feel competent, their need for autonomy over choices, or their need for connection?
RQ2
Probe: If not, is this something you have thought about? What would it take to make it feasible?
Framework utility - RQ3
11
If your system could reliably infer user states, what is the primary way you would alter its response?
RQ3
Probe: Would this require new infrastructure, or could it be layered onto what you already have?
12
What are the biggest practical barriers to implementing state-aware features in your current system?
RQ3
13
Does the ABSD framework provide vocabulary or structure that would be useful in your product or analytics discussions - even before any implementation?
RQ3
Probe: Is there language here for something you have been trying to describe to your team?
14
What is missing from the framework? What would you add, remove, or restructure to make it more applicable to your domain?
RQ3
Closing
15
Is there a specific use case in your domain where understanding user motivation - not just behaviour - would have the highest impact on outcomes?
Want to walk through this protocol yourself? The guided voice interview tool at predictwhy.com/interview reads each question aloud and captures your answers by voice.