Situs Group tours: current state and new state

From the raw transcripts of the Sep 1 whiteboard session, the Oct 13 follow-up and the Oct 16 onboarding call with Miriam and the team. Every line is the speaker's own words; Oct 13 speakers are inferred because that transcript only says "Them" and "Me".

Not said by the customer: tour length, buffer or tour hours. The only numbers in transcripts are Eileen's "15 minute break" and Fede's own "15… 30, maybe 45". Our settings for Situs hold 30-minute tours with a 30-minute buffer.

Is Miriam's property-manager role a problem? Yes, in today's mode. With no explicit tour list, only people with the leasing-agent role are picked. The moment a company tour list is set, any staff role can be on it, property manager included, as long as she has a calendar connected. Recommendation: set the list to Lee, Deiby, Miriam once their calendars are connected, and take Courtney off it. What looks like a leftover test account ("tester") also sits on the Situs team and should be removed.

We classified every message on 1,563 Situs guest cards (Jan 2025 to Sep 2026) by language with a word-count rule, no paid model. Spanish prospects already write in Spanish on the card, so the conversation language can decide who gets the tour (inferred from the data), re-checked after the first real message. Asking is the fallback, not the default.

Verified in the code at today's main, not from the design pages. The design pages describe fairness ranking, drive time and automatic re-hosting; none of that exists.

Proposed cut: two small PRs, the tool plus the synced-lead lift, then the notices and the PMS note.

Superseded by the full design: When staff move, cancel or hand off a tour (Nov 1), which covers reassigning, moving with the prospect's yes, and cancelling.

Share of cards where the prospect wrote in Spanish or mixed, among cards with any prospect text. Spanish arrives by word of mouth, Facebook ads and walk-ins. The listing sites are almost entirely English.

Rows do not sum to the 742 cards with text; the remainder is "unknown" (one-word replies) or minor sources. By building: Lampliter and Cottonwoods 30% Spanish, Barcelona 23%, the DU houses 0%. Spanish share fell from 22% of 2025 cards to 10% of 2026 cards. Source is a hint, not a rule: use the conversation language.

From the AppFolio property calendar and Showings report we captured on Sep 1 (held showings, cancellations removed), plus guest-card activity for who booked and when.

Carr Street Flats (182), Barcelona (113) and Lampliter (88) are 48% of all tours; Carr Street sits under 1 km from Barcelona, Mckenzie and Zephyr, which is where most of the under-1-km hops are. Who books: Miriam 256, Courtney 223, Michelle 139, a shared "Assistant Lead" login 124, Deiby 98, Lee 66. Since late March, 79% of bookings were made by someone other than the giver. The AppFolio on-call calendar is the maintenance rota, not a leasing calendar; no tour giver is on it.

In this section: What they do today · The 11 short pairs · The rule · Replay · Offer or confirm · Build and prove · Willows run · Measuring · Sources

What each customer does today, measured

"they schedule a showing for somebody in Denver, and then 15 minutes later, they schedule a showing for something in Lakewood. Would Clara kind of help us with that?"

Miriam, Situs, Slack, Oct 16

Staff practice is the baseline to beat, not the target. On how tours get booked: "there's no Rhyme or Reason this" (Sep 1 meeting transcript, canvas F0BU63PCQ5T, paragraph 536; the speaker label, Eileen, is the transcript tool's inference).

The 11 pairs every rule must refuse

Two different prospects, one host, two homes, and less time between the tours than the drive by over 5 minutes. 8 leave in the weekday rush (22–38 min drives), 5 of them with exactly Miriam's 15-minute gap. 3 are next-door homes booked at the same time.

The rule, and two open choices

Same home: the flat buffer. Different homes: the larger of the flat buffer and the drive, rounded up to 5 minutes; weekday departures from 3 to 6 PM use the rush number, fixed in code. Checked at offer and at confirm. Leaving from the office is out of scope: a slot is checked only against the host's tours before and after it that day, so the first tour of the day has no drive constraint.

"is it 15 minutes without traffic or 30 with traffic? And then adjust it that way."

Fede to Situs, Oct 20. Two stored numbers per pair, picked by the time of day, do this; it is not live traffic.

Shortening first is not an option. 15-minute tours with a flat 15 and no drive check let 5 of the 11 through. Length and gap change only after the check is proven.

Replay: every real tour under each rule

"Pushed later" counts real two-people pairs, besides the 11, that a rule moves to a later quarter hour. Situs rows replay staff tours (Clara has booked none there); today's 30/30 refuses the 11 only by pushing 137 others.

* 46 of these 137 overlap only because today's settings count Situs's 15-minute tours as 30.

Fede's rule pushes 47 Situs pairs, each by one quarter hour: 34 at the same home, which now keeps a flat 15, and 13 between homes, 12 of them hops that left the host up to 5 minutes late. Western Slope staff: 6 (2 at the same home, 4 with no stored drive). Section 9 replays a year of Situs tours under it.

Offer or confirm: times go stale

Clara offers times; the prospect answers minutes or a day later, and meanwhile another tour can land next to the chosen time.

Earlier rule. Medians of 20 runs; ranges 6–9 (confirm only), 0–2 (2 h), 0–6 (24 h). Western Slope staff, 24 h: 2 extra replies with both checks, simulated.

Build, then prove it end to end

Build notes. The drive check ships as a fix with no switch: it changes nothing where no drive numbers are stored, and staff bookings skip it, as they skip the buffer today. One drive-time engine in one place: the two-homes-in-one-visit path uses it and loses its own drive code. The cases the code must pass before the replay: 15 real Situs pairs, the 11 short ones among them. The "that time was just taken" reply already works on all six booking paths (product pull request #21729 corrects the old code comment); step 3 of the Willows run checks it.

Proof, in order:

1. Read-only replay of both customers' real tours on the built code at 15-minute tours with a flat 15: all 11 short pairs refused, and the same pairs pushed as in the replay above (47 at Situs, 6 at Western Slope).
2. One end-to-end run at Willows, below.
3. On Fede's go, Situs and Western Slope switch together from 30/30 to 15-minute tours with a flat buffer of 15 (V1) or 0 (V2), whichever he picks.

The end-to-end run at Willows

Willows at 15-minute tours with a flat 15, with test properties already measured into the drive-time map. From property A: B is over 15 minutes away, C is next door, and D's rush number flips the answer at a 4:30 PM departure. The host has tours at A ending 1:30 and 4:30 PM on a weekday. Clara must offer (example numbers):

1. Ask. A prospect asks Clara for times at B, once by text and once by a call, then at C and D. Pass: the times offered match the table, in the text thread and in the call transcript, and no Google call is made.
2. Book. The prospect confirms 5:00 at B. Pass: the tour row, the host's calendar event and the confirmation message exist. Evidence: their record ids and the message text.
3. Taken at confirm. Clara offers 5:00 at B; before the prospect answers, another booking puts the host at A at 5:15. Pass: when the prospect confirms, Clara says that time was just taken and offers the next clear one; nothing is booked at 5:00.

Measuring: once per enabled property

• What is stored: a drive-time map between the company's properties, kept as one record per direction (A to B and B to A are two; the way back can take longer). Each record holds the minutes with no traffic, the minutes leaving at 4:30 PM on a weekday, and when it was measured. The properties are untouched; nothing goes in the knowledge base.
• Which properties: every enabled one except archived ones, the stand-in for the company's phone line, and any without a street address. No self-guided or distance cut.
• When: the sync that already runs every 15 minutes measures up to 50 new pairs per company each run. A newly enabled property is picked up on the next run; existing customers fill in on their own.
• Before a pair is measured: the area's longest drive applies, so a half-filled map is safe.

About $24 once for all three: one Google traffic call per one-way drive, about 7,400 in all; 5,000 a month are free, then $10 per 1,000 (price pages read Oct 19 and Oct 21). A new property adds two drives per other enabled property. This pass made 0 Google calls.

Later, not now: offering times next to a host's other tours. Clara would offer first the open times closest by drive to the host's tours before and after, so a host's day bunches by area. It needs the same stored numbers, a ranking step in the offer list and its own replay; Fede, Oct 21, in session: for later, not part of decision 5.

30-day replay before the settings switch (Oct 23)

Every tour in our records from Sep 23 to Oct 22, re-offered through the live booking engine at the moment it was booked, at today's 30/30 and at 15/0. Read-only.

Verdict: 15/0 is safe at Western Slope: nothing gets worse, two back-to-back staff bookings now fit, and nearly three times as many times are offered, mostly from the finer 15-minute grid. The 5 still refused: three staff picks inside the 60-minute lead time, one staff pick over a calendar block, one Saturday booked after the weekend cutoff.

Verdict: no replay possible: we hold no Situs tours. At 15/0, staff showings get no gap until Clara books there.

Staff and live run: 7 of 7 staff bookings and moves still book at 15/0. Live on the test line at 15/0: two homes back to back, two tours 15 minutes apart at one home, and a tour 5 minutes after a manual calendar block: all booked. Sloppy staff texts got a question, never a booking. Two prospects racing for one slot: no double booking.

Broke: the company-wide line checks times against its catch-all property, not the home: the test company booked 8:15 AM at a home whose tours start at 9:00 (#21995). At Western Slope hours match, but main-line calls get the longest-drive padding. Offers abut Kat's calendar blocks at 15/0 (60 of 71, #21971). The gap after a two-home visit is measured from the first home (#21974; none at Western Slope).

Verdict: safe; the switch itself broke nothing. Flip after #21971 (it removes the pad around staff blocks). #21995 is live now at either setting.

Files: research/…/drive-policy-v2/replay-30d/. The planted 5-minute pair was refused, the far one offered.

Sources and caveats

Held Situs showings, Jan 2025 to Sep 4 2026 (20 months, all used), as booked vs re-booked under tonight's rule, same order (simulated). Customer page for Situs: Clara update (their tours, what we built, the same tours booked the new way; V2 numbers, same home no gap).

Where the lost tours come from (simulated). Same home / hop gap: V1 15 / max(15, drive); V2 0 / max(15, drive); V3 0 / max(5, drive).

The same-home 15 costs 16 of the 36 lost tours and 34 of the 35 on-time refusals; the other 20 are the price of fixing the 23 late hops. V3 equals V2: tours start on quarter hours, so the 12 hops late by 5 min or less still move to the next quarter hour. No late hop returns; mutation control red in all three.

Verdict: safer, a little slower. The rule removes all 23 late hops (13 at rush), but refuses 35 on-time pairs, mostly same-home back-to-back, and fits 476 of 512 tours in the same hours.

By host: Miriam 10, Deiby 5, Eileen 4, Lee 4. Caveats: their order kept; 6 rush numbers and 8 pairs estimated; no work orders or admin in the export; midday trades not measurable. Files: year-report/.

How they assign today, reverse-engineered

No owner by home, area or day: it is "whoever is already out", with Spanish as a soft first cut. Eileen: "Honestly, it like fluctuates" (Oct 13). Her stated rule (Oct 16, raw transcript): "if they speak Spanish, only Miriam and Davey can do the tour"; English split evenly between Lee and Deiby.

What PropFlow does today

As in section 4: the first person on the list with a free slot owns the day. Nothing reads language or location to pick who. The drive map (all 34 Situs homes measured) sets only the gap between one person's tours.

Would a nearer host have beaten the drive?

Production drive map (24 late hops on it, 23 on earlier numbers). "Free" = no AppFolio showing; their own calendars and midday trades are not in the export, so upper bounds. Work orders carry no staff name. Mutation control: 5 planted nearer hosts found, 5 busy twins rejected.

Options

Gera's model: company-level calendars, an explicit rotation per function (Fede, Sep 18). Every option is an order on that one list. Scripts and outputs sit in this page's research folder.

Pick one per row. The recommendation is first.

PMS-in-adapters check: the decision path names no PMS. The tour's calendar-event field is named after Outlook even for Google; debt, not a blocker. Working notes: the five research reports and scripts are in the docs repo under research/situs-tours-2026-10-19/.