What it is

Short structured conversations in the lobby, same day, while it is fresh: a fixed set of 3–4 questions asked of everyone for week-to-week comparability, then — for anyone willing to stay fifteen minutes — a deeper conversation about what happened before they came, what they expected, and what they will do next.

What it tells you and what it cannot

Tells you	Cannot tell you
Same-day client voice at low cost, with fresh recall	Representative population estimates — the lobby sample is who was there
A comparable weekly trend (the fixed core) and the why behind it (the depth strand)	Anything about people who never showed up, or left angry before the exit
Which surprises are walking out the door unasked	The full truth from clients who depend on the service — candor needs engineering

How it runs

Who does the work	Participants	Elapsed time	Cost
Trained interceptors who write up the same day	Fixed core accumulates to hundreds; depth ~12–15 per segment reaches saturation	1–2 weeks per wave; can run ongoing	$ — one trained interviewer and a quiet corner

Strengths, watch-outs and quality signals

Equity & consent

The consent script is load-bearing: state plainly that the interviewer is not their caseworker and answers cannot affect their case or benefits. Offer languages, find a private-enough corner, and stay alert to retaliation fear — the clients with the most at stake are the most likely to say everything is fine.

When to choose it

Method & sampling

One researcher pair worked the main benefits-lobby exit across five weekdays, rotating morning, lunch-hour, and late-afternoon shifts. We approached every third adult crossing the exit line, kept the pitch under 30 seconds, and made clear we were not selling anything (real background: GOV.UK Service Manual on pop-up research; Yocco, Smashing Magazine 2016). Every consent script included, verbatim: "I'm not your caseworker, and your answers won't affect your case or benefits."

Across the week we approached 57 people; 41 completed the fixed core, a 72% response rate. Everyone who agreed answered the same four questions, in the same words and the same order — the comparability core that will repeat every week:

1. "Did you get what you came for today?"
2. "Is anything you needed still unresolved?"
3. "Do you know what happens next?"
4. "Would you tell a friend in your situation to come here?"

Completers could opt into a roughly 15-minute depth conversation in a quieter corner of the lobby, with a small token incentive; nine did. Interviewers wrote up each depth conversation the same day. Intercept content spans journey stages 2–4 and 6.

Fixed-question results

The fixed core is the WHAT: counts, no explanations, no causal claims. This is week one, so the tallies below are a baseline; the week-over-week trend begins when the same four questions repeat.

31 of 41 said yes; the tally records the answer, not what "what I came for" contained.

13 named something still unresolved and 4 were not sure — and 9 of those 13 had also answered yes to Q1.

19 of 41 were less than fully certain what happens next.

29 of 41 would send a friend; 8 made it conditional.

Depth findings

The depth strand is the WHY. Nine conversations is a small n — themes below are directional, not stable — and they spanned several segments; two voices are featured here, not the only two. Each conversation walked the same four frames: what happened before you came, what you expected, what you will do next, and what you would tell a friend. Theme counts are conversations touching the theme, of nine.

First featured voice — crisis-entry segment, several programs open at once, intercepted during her service-connection and delivery period.

On the fixed core she had answered yes to "Did you get what you came for today?" — the window accepted her paperwork. The depth conversation held the rest. Her history has now been retold nine times across seven doors, today's visit among them, and next week two of her program appointments sit in the same hour because each office books on its own clock. "I can do the appointments. I cannot do them all at once." On what she would tell a friend: come, but — "If housing is the thing everything depends on, why is it the last thing moving?" She asked why every door needs the whole story again; the interviewer recorded the question and did not answer it.

Second featured voice — came for one document, on a lunch hour.

She was on site 52 minutes on an unpaid lunch hour, and the document that triggered the visit got fixed. Nobody at the counter asked a question beyond the document mismatch. Only when the interviewer asked directly whether anything else was going on did she mention a utility bill she is behind on and a waitlisted concern about her son — adding that she would not have raised either herself, least of all at an open counter. On the friend question she answered "it depends": "If the portal had worked, I would never have needed to come." Whether she would take a referral from someone she met once is more than one intercept can establish.

What the tally alone would have missed

Read alone, the fixed core calls the week a success: 31 of 41 got what they came for and 29 would send a friend. The depth strand complicates that reading without contradicting it: 6 of 9 depth participants left with an unasked need or an unclear next step, and 4 of those 6 had just answered yes to "Did you get what you came for today?"

• Fixed datapoint: "Yes, I got what I came for" (31 of 41). Depth complication: the yes measures the presenting task. The lunch-hour voice left in 52 minutes with the document fixed and two needs unspoken until a researcher asked.
• Fixed datapoint: "Yes, I know what happens next" (22 of 41). Depth complication: next steps were known one program at a time and held in memory; for the crisis-entry voice, two of them now occupy the same hour.
• Fixed datapoint: "Yes, I would tell a friend" (29 of 41). Depth complication: the yes arrived with a survival manual — come at opening, bring every paper twice, budget the whole morning.

Bias & equity caveats

• Acquiescence and social desirability. Respondents depend on this building for benefits; a yes to a stranger with a clipboard is the cheap, safe answer. The not-your-caseworker line was read to every respondent, yet three still asked whether their answers would reach their worker — the separation is stated, not automatically believed. Treat Q1 and Q4 shares as ceilings.
• Selection and length bias. The sample is people who came in person and crossed this exit during our shifts; 9 of the 16 non-completions fell in lunch-hour shifts, so the strand most under-represented is precisely the in-and-out segment — one reason a voice from it is featured above.
• Language coverage. Intercepts ran in English and Spanish; 3 of the 16 non-completions ended at the language line. Exits in other languages are effectively out of sample this week.

Findings & recommendations

Owners are roles, not names. On saturation: nine depth conversations sits below the point where themes typically stabilize — often around twelve interviews (real background: Guest, Bunce & Johnson 2006) — so the themes stay provisional, and the standing recommendation to ourselves is to keep intercepting toward 12–15 depth conversations per segment.

Limitations & quality self-check

• The sample is lobby-present by construction: people who reached the building and walked out the door. Whoever gave up at the portal, the phone queue, or the parking garage never crossed our line and is invisible here.
• One week, one exit. Fixed-question counts are a baseline, not a trend; comparability begins when week two asks the same four questions in the same words.
• Small depth n (9): themes are directional. All tallies, quotes, and respondent details are synthetic composites for demonstration.
• Unknowns preserved. Depth participants raised each of these as lived experience; this report leaves all three unresolved:
  • "Which retellings are legally required and which are habit"
  • "What information can travel with consent today"
  • "Which missed appointments could be prevented by sequencing"

Self-check: WHAT (fixed core) and WHY (depth) kept separate; response rate reported; both featured voices segment-attributed with continuity intact; no causal claims made from the fixed tallies.