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.
9-min read · Updated April 2026
Lumi · Wednesday
Good morning, Niki.
Two showings · three leads need a nudge.
Showing · Passeig de Gràcia 84
60 seconds.
Used to take four hours.
A defensible 3-comp deck used to be the agent's Sunday-afternoon project — pulling MLS records, calculating price-per-sqm, formatting in PowerPoint. The protocol that replaces this isn't a generic CMA tool. It's a tightly-constrained Claude prompt with strict comp-selection rules and mandatory source citations.
Five rules. Three comps.
The discipline that produces a defensible deck — defensible meaning every number has a source, every comp meets the hard limits, every recommendation is a band rather than a number.
- 01
Three comps. Not five, not seven, not whatever the MLS dumps.
The CMA discipline that separates good agents from data-dumpers: the analysis is on three carefully chosen comps, not on the random 12 the MLS returns. Three is the sweet spot — enough for a defensible band, few enough that each one can be specifically discussed in the conversation. The protocol's first job is choosing the three; everything downstream depends on that choice.
- 02
Within 500m, within ±15% sqm, within 90 days.
These three constraints are the hard limits. Stretch any one and the comp's relevance falls off a cliff. A 3-bed 200m² in the same neighbourhood is not comparable to a 3-bed 130m². A property 700m away is not comparable to one 300m away — even in the same neighbourhood. A sale from 18 months ago is not comparable in 2026 markets. The protocol enforces all three.
- 03
Cite every source. No exceptions.
Every number on the deck has a citation: MLS listing ID, cadastre reference, portal URL with a date-stamp screenshot. This is what separates the protocol from 'an agent's hunch presented as data'. The citation makes the deck defensible in any subsequent conversation — with the seller, with a mortgage lender, with a buyer's lawyer. Without citations, the deck is just an opinion in a fancier wrapper.
- 04
Recommended price is a band, never a single number.
The deck's recommendation is always a range — 'List between €820k and €865k'. A single number is false precision and forces the seller to either accept or reject. A band gives the seller (and the agent) decision space: where in the band to start, when to consider the lower end, what would justify pushing toward the higher. The default band is ±5% on the comp median; widen it only with explicit reasoning.
- 05
Three use cases. Same deck. Different conversation.
The same 3-page deck supports three different conversations: CMA (with a seller deciding listing price), listing pitch (with a seller deciding which agent to hire), and buyer education (with a buyer evaluating an offer). The deck doesn't change; the conversation around it does. The protocol generates one deck and the agent runs three different meetings with it.
Three deck failures that lose credibility.
Each one looks competent on first glance and collapses under scrutiny. The protocol's strictness — hard comp limits, mandatory citations, banded prices — is what prevents these failure modes.
“[Subject: 3-bed 145m² Cascais. Comps cited: 3-bed 200m² Cascais sold 12 months ago at €1.3M, 4-bed 165m² Estoril sold last month at €1.1M, 2-bed 110m² Cascais listed at €820k.]”
All three comps fail the protocol's hard limits — wrong sqm segment, wrong neighbourhood, wrong recency. Conclusions drawn from these will be wrong. The protocol exists to enforce the discipline; bypassing it produces decks that look authoritative and aren't.
“Recommended listing price: €847,500 (matches comp median).”
False precision. The seller now thinks the price is exact and will not consider €830k or €860k. A range gives the seller (and the agent) decision space, calibrated to comparables but not pinned to a false certainty. The protocol's default is ±5% on the median.
“Three comps in the area sold for €820, €855, and €870 per sqm in the last 90 days. Recommended price band: €820-865k.”
No MLS IDs, no portal URLs, no date stamps. The seller (or any subsequent reviewer) cannot verify these numbers. The protocol's whole credibility — and the seller's trust — is in the citations. Without them, the deck is unsupported.
What to feed Claude.
Sonnet recommended — the comp-selection logic and the price-band reasoning need careful inference, and Haiku tends to over-include marginal comps. PDF rendering happens downstream of Claude's structured output.
You are a senior real-estate agent's
comparable-property analyst.
INPUT
You receive: the subject property
(address, beds, baths, sqm, year built,
key features, asking price), 5-15 nearby
recent sales + active listings within
500m and ±15% sqm, the agent's
target use case (CMA / listing pitch /
buyer education).
OUTPUT
A 3-comp deck — three pages of
structured analysis:
PAGE 1 — RECOMMENDED PRICE BAND
The trio of comps chosen, why these
three (closest match on geometry,
age, sqm, recency). The
recommended-asking band ±5%, with
the reasoning chain visible.
PAGE 2 — COMP COMPARISON TABLE
Side-by-side: subject vs comp 1
vs comp 2 vs comp 3. Rows: address,
beds, baths, sqm, sold/asking price,
€/sqm, days on market, key
differentiators.
PAGE 3 — MARKET CONTEXT
Trend: are similar properties in
this segment moving up, down, or
flat over the last 90 days?
Inventory: how many comparable
listings are active right now?
Velocity: median days-on-market
for the segment.
RULES (non-negotiable)
1. Always cite the source for every
number. MLS listing ID, cadastre
reference, or portal URL.
2. Comps must be within 500m AND ±15%
sqm AND sold/listed within 90 days.
Stretching any of these dimensions
weakens the analysis.
3. The recommended price band is a
range, never a single number.
±5% on the comp median is the
default unless explicit reasoning
for a wider band.
4. Include the AGENT-FACING analysis
on page 1 (why these three) and
the SELLER-FACING summary on page 2
(the comparison table that makes
the conversation easy).
5. Date-stamp every comp. Comp data
ages — a 90-day-old comp is
borderline; 6-month-old is invalid.
ANTI-PATTERNS (never produce these)
- Stretched comps (different sqm
segment, different neighbourhood,
different feature class)
- Single-number recommended price
- Missing source citations
- Inferring trends from <5 data
points
- Padding to multiple pages with
generic market commentaryPull comp candidates from MLS API, feed to Claude with the subject property. Pipe Claude's structured output into a PDF template (3 pages) for delivery.
Generating the deck is step one.
Citing every number 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 data-grounded advisory pattern from financial services — strict citations, banded recommendations, comparability rules. Our slice: a 3-comp deck in 60 seconds vs the 4-hour CMA tradition.
More guides like this on @lumi.estate. Follow if any of this was useful — it's how we know to keep writing.