Your second brain
for closing deals.
Speak after a showing. Forward an email. Pull up a client. Lumi captures the soft signals, fills the brief, and feeds Claude — automatically.
- Voice → CRM, auto. No forms.
- Works offline. Syncs when you're back.
- Free for agents in EU · LatAm · MENA.
10-min read · Updated April 2026
Lumi · Wednesday
Good morning, Niki.
Two showings · three leads need a nudge.
Showing · Passeig de Gràcia 84
The first 3 minutes
after a showing.
The window between closing the front door and starting the engine is the most valuable three minutes in your week. Most agents waste it checking messages. The agents who close more spend it on a 90-second voice memo that turns into the brief their AI lives off for the next 30 days.
door closes → walk slowly
phone up → speak full sentences
10 prompts → any order, voice
3 minutes → stop, get in car
red light → AI has parsed it
CRM updated.Door-to-CRM in under 5 minutes. The data is captured before it decays.
Why this window matters more than the one before it.
Agents spend hours preparing for showings — comp data, opener lines, curated route through the property. Then they walk out, hand the buyer off to their car, and drive away with everything they just learned slowly evaporating. The asymmetry is bizarre: 4 hours of prep, 30 seconds of capture, and then the brief is whatever you remember at 9 PM when you sit down to write notes — which is, on average, about a third of what was actually said.
The 3-minute protocol fixes the asymmetry by treating the post-showing window as a deliberate part of the workflow, not a transition. The output is a structured brief that compounds — every future AI-drafted message, every showing prep, every weekly review pulls from the data captured in this window.
Memory decay starts at minute 5.
Across studies of professional recall (medical residents, expert witnesses, real-estate agents), specific verbatim detail decays sharply between minute 5 and minute 30 after an event. The exact phrasing of what someone said — the soft signals that make follow-ups land — is the first to go. By the time you're back at the office, you remember 'the kitchen' but not 'this is where I'd make coffee'. The 3-minute window is the only place those quotes still exist intact.
The walk back is the only true private moment.
Once you're in the car, you're checking messages. Once you're at the office, you're answering them. Between the property door and the car door is 90 seconds of nobody asking anything of you. That's when the memo gets recorded. Outside that window the protocol breaks because attention is gone.
Voice beats notes by a factor of 3.
Typed notes capture maybe 40% of what was said. Voice notes capture 95%+ — and they capture the agent's emotional read alongside the facts. Tone of voice in the memo ('she said it sharply') becomes a signal in the AI's output. You cannot type that signal in.
AI extraction means you don't have to structure it.
The reason agents who try to write structured notes give up after 2 weeks is that structuring while you're processing what just happened is cognitively expensive. Voice → AI extraction means the agent's job is to ramble; the AI's job is to organise. The agent gets back the structured 7-field brief, edits anything wrong, and the CRM is updated.
“The 3-minute window is the only place the verbatim still exists. Everywhere else it's already paraphrased.”
Five steps. Three minutes.
The whole loop fits between the front door of the property and the first red light on your drive back.
- 01
Door closes. Phone in hand. Walk slowly.
The walk back is intentional. Take 90 seconds even if your car is 30 seconds away. The point is to give yourself enough physical space between the showing and the next thing for memory to settle. Walking helps.
- 02
Hold the phone like a recorder. Speak in full sentences.
Whisper transcribes well from any iPhone or Android. Speak as if you're telling the story to a colleague — full sentences, names, addresses, quotes. Don't try to be efficient. Efficiency hurts the brief.
- 03
Cover the 10 prompts in the template — in any order.
The template above is the checklist. You don't have to follow it in order — speak naturally and the AI will sort. But if you're missing a field by minute 2 of the memo, the template is the prompt to remember it. Most agents internalise the 10 prompts within a week and never look at the template again.
- 04
Stop at minute 3. Get in the car.
If you're still talking at minute 3, you're rambling. Past minute 3, the signal-to-noise ratio drops. Better to capture the cleanest 90 seconds than to over-record. The unspoken-thing prompt at #10 is the natural stopping cue.
- 05
AI extracts. You review at the next red light.
Lumi (or your equivalent setup: Whisper + Claude) processes the memo within 30-60 seconds and pushes the structured YAML to your CRM as a draft. At the first red light or before you start the engine, you tap once to approve, or tap to edit one field. The whole loop — door closes to CRM updated — is under 5 minutes.
The 10 prompts to cover.
Print this. Tape it to your dashboard for the first month. Within two weeks the prompts are internal — but the scaffold is what gets you there.
# ── 3-minute voice memo · spoken on the walk back ─────
# Speak this naturally — Claude parses it into the
# 7-field brief automatically. Don't worry about order.
1. Who I just showed it to (full name).
2. Property address. One line.
3. Their reaction when we walked in.
- First room they paused in.
- Anything they said unprompted.
4. The thing they asked twice.
(Twice = it matters.)
5. Body-language between spouses / decision group.
- Did they look at each other? When?
- Who got quiet? When?
6. What they listed as the dealbreaker
(and whether they said it nicely or sharply).
7. My read on intent stage AFTER this showing
(browse / serious / urgent — and what shifted it).
8. Next step I committed to, with a date and time.
9. Anything I told them I'd find out
(HOA fees, school catchment, neighbour kids).
10. One thing I noticed they didn't mention —
that I'm going to bring up next time.
What to feed Claude.
The system prompt that turns your free-form ramble into a structured 7-field brief. Tested against Claude Haiku — fast enough to run in your car before the first red light.
You are a real-estate agent's post-showing
note-extractor.
INPUT
A voice-memo transcript recorded by the agent
within 3 minutes of leaving a showing. Free-form.
May ramble. May skip fields. Always in the agent's
own voice.
OUTPUT
A YAML block with exactly these 7 fields, populated
from the transcript. Each field is required —
if the transcript omits it, write [unknown] (with
brackets) so the agent sees the gap.
client: string (full name)
property: string (address, one line)
intent_stage: "browse" | "serious" | "urgent"
intent_shift: string (what changed from before)
soft_signals: string[] (verbatim quotes &
observations, not paraphrased)
hard_constraints: string[] (only ones surfaced
in this showing — not the full
list)
decision_group_obs: string (body language between
partners; who got quiet, when)
dealbreaker: string | null
next_step: { what: string; due: ISO date }
agent_followups: string[] (things to find out)
unspoken: string (what they didn't say
that the agent flagged)
RULES
1. Soft signals must be verbatim. If the agent said
"she paused at the kitchen window and said this
is where I'd make coffee", capture that phrase
in quotes. Do NOT paraphrase to "client liked
the kitchen".
2. If intent_stage shifted, name the shift
("browse → serious") and the trigger.
3. Next step must have an actionable due date.
If the agent said "I'll send some options",
ask for a date — do not invent one. Output
{ what: "send options", due: "[needs date]" }.
4. dealbreaker captures the one thing that, if
not solved, kills the deal. Tone matters —
"she said it sharply" is signal; preserve it.
5. unspoken is the most powerful field. It captures
what the agent noticed they didn't bring up. This
is the open loop for next contact.
Voice: clinical, structured, no commentary.
The agent edits — you don't editorialise.Same showing. Voice in. YAML out.
A real-feeling 90-second transcript on the left, the AI's first-pass extraction on the right. No editing applied — this is the raw output.
OK, just left the Rua da Prata showing. Sofia Ferreira, husband Carlos. They walked in, Sofia went straight to the kitchen window and stopped — literally said "this is where I'd make coffee." Asked twice if the neighbours had kids, both times when Carlos was looking at something else, which tells me she's the one driving on lifestyle. Carlos asked about HOA fees, asked about commute to Canary Wharf, never asked about anything emotional. Standard. The stairs to the second-floor bedrooms — Sofia paused, said "the kids" out loud, Carlos didn't react but I caught his eyebrow. That's the dealbreaker. Stairs. They'd love this place if it was ground floor. I told them I'd send two ground-floor + balcony options by Thursday at 14:00. I need to find out HOA fees on the building because Carlos asked twice. They didn't ask about the school catchment, which is interesting — they have twins entering primary, that's going to come up next time, I want to bring it up before they do. Intent stage: serious. They were browse before this. They left talking about timing.
client: "Sofia & Carlos Ferreira"
property: "Rua da Prata · 2-bed, second floor"
intent_stage: "serious"
intent_shift: "browse → serious — they left
talking about timing for the
first time"
soft_signals:
- "Sofia at kitchen window:
'this is where I'd make coffee'"
- "Sofia asked twice if neighbours had kids
(both times when Carlos was looking away)"
- "Sofia paused on the second-floor stairs,
said 'the kids' out loud"
- "Carlos's eyebrow lifted at the stairs —
he didn't comment"
hard_constraints:
- "ground floor (twins, stairs concern)"
- "balcony"
decision_group_obs:
"Sofia drives lifestyle decisions. Carlos
handles financial/logistical (HOA, commute).
Carlos approves silently — eyebrow on stairs
was the tell. Address Sofia primarily; cc
Carlos on financial details."
dealbreaker:
"second-floor stairs (twins)"
next_step:
what: "send 2 ground-floor + balcony options"
due: 2026-04-25 14:00
agent_followups:
- "HOA fees on Rua da Prata building
(Carlos asked twice)"
unspoken:
"school catchment / primary school —
they have twins entering, didn't ask.
Bring it up at next contact."
Notice the verbatim soft signals — “this is where I'd make coffee” survives intact. Notice the unspoken field — the school catchment thing the buyer didn't bring up — captured because the agent noticed and named it. That's the open loop for next contact. That's the message that lands.
The failure modes — and the fix for each.
If you try the protocol for two weeks and abandon it, it's almost always one of these four. Each one has a one-line fix.
Waiting until you're back at the office.
By the time you're at your desk, 60% of the soft-signal verbatim is gone. You'll write 'liked the kitchen' instead of 'this is where I'd make coffee', and your AI follow-up will read as generic. The 3-minute rule exists because the data is perishable.
Trying to type instead of speak.
Typing forces structure, which forces editing in your head, which means you skip the quotes. Voice memo is non-negotiable for this protocol. If you hate the sound of your own voice, get over it — nobody listens but the AI.
Skipping the 'unspoken' field.
It feels weird to capture something the client didn't say. But the unspoken field is what differentiates a competent CRM from a sales-leading one. The fact that Sofia didn't mention school catchment despite having twins is the single most useful piece of information from that showing — it's the open loop for next contact.
Letting the memo go straight to the CRM without review.
AI extraction is ~85% accurate on first pass. The 15% that needs editing is usually a name spelling, a date, or a misclassified intent stage. 30 seconds of review on each memo, before it lands in CRM, is the difference between a brief that AI can trust and one that compounds errors.
One showing. Six future moments.
The 3-minute memo doesn't pay off in the next hour. It pays off across the next 30-60 days, in places that look unrelated:
- Same eveningAuto-drafted thank-you message references the kitchen window. Buyer reads it as personal, not template.
- + 2 daysPre-call brief for the second showing surfaces the stairs concern as the dealbreaker — agent leads with ground-floor properties.
- + 7 daysReactivation message (if buyer goes silent) cites the kitchen window and the Estoril mother — re-opens the conversation specifically.
- + 14 daysSpouse-aware update gets sent in two versions: lifestyle frame to Sofia, HOA-and-commute frame to Carlos. Same listing. Same hour. Different inboxes.
- + 30 daysThe 'unspoken' field — school catchment — surfaces as a prompt at the next meeting. Agent brings it up before the buyer does. Trust signal.
- + 60 daysQuarterly client review uses soft signals to identify which 4 of the agent's 30 active clients are highest-fit for a new listing that hits two of their flex conditions. The brief drives prioritisation.
None of those six future moments work without the brief. The brief doesn't exist without the 3-minute memo. The whole pipeline of AI-leverage is downstream of those 90 seconds on the walk back to your car.
Speaking it is step one.
Letting AI structure it is step two.
Lumi is the app that runs this workflow for you. You speak after a showing — Lumi captures the soft signals. You forward an email — Lumi updates the constraints. You open the app at 8am — the brief is already there, ready to feed Claude.
- Voice → structured CRM, automatically
- No forms. No data entry. No copy-paste.
- Free for agents in EU · LatAm · MENA
Lumi · Wednesday
Good morning, Niki.
Two showings · three leads need a nudge.
Showing · Passeig de Gràcia 84
Pipeline
Active
8
Warm
4
Cold
2
Clara Ruiz
Active€1.8M · 3BR
Passeig de Gràcia showing · 11:30
Andreas Moreno
Active€2.4M · 4BR
Send comps by 18:00
Dimitri Schneider
Warm€900K · 2BR
Contract review today
Silent 3d · last 3 days ago
Sarah Mitchell
Cold€1.2M · 3BR
Draft re-engagement
Silent 9d · last 9 days ago
A real-estate adaptation of the voice-first capture pattern from operator and vibe-marketing communities — unstructured voice in, structured AI out beats forms every time. Our slice: the 3-minute window after a showing, where verbatim recall has the shortest half-life in the agent's week.
More guides like this on @lumi.estate. Follow if any of this was useful — it's how we know to keep writing.