Investor Update Numbers — October 2026

What Clara has done for real customers this year, how fast it is growing, and the other things worth saying to investors. Recounted from scratch on Oct 2, 2026 and cross-checked against Twilio and ElevenLabs.

Jan 1 – Oct 2, 2026 · Central Time · real customers only (Camellia / JP&Co and Western Slope) · every test property, team phone and our own lines removed · Proposed — pending Fede's review

HeadlineGrowthWhat else to highlightWhere this goes nextEngineering velocityRecount vs the first countMethod and caveats

Headline, year to date

Clara has handled 945 conversations, 689 phone calls and booked 101 tours for two customers since April. Real activity starts in April 2026, so this is six months of work. The last 30 days are up 31–79% on the 30 before depending on the metric, with Western Slope's launch inside that window.

945
conversations handled
689
phone calls
101
tours booked
1,207
text messages
1,002
emails
589
people Clara talked to
Year to date, by customer
MetricJP&Co (Camellia, Yale 25)Western Slope (first real activity Sep 18)Total YTDDetail
Tours booked812010185 confirmed, 15 cancelled, 1 other. 96 distinct prospects.
Conversations handled803142945719 of them had a person write in and Clara reply (the rest are Clara-initiated: reminders, outreach).
People Clara talked to49099589Distinct prospects, residents and vendors.
Text messages9702371,207305 in, 902 sent by Clara: 257 conversational replies, 566 automated reminders and outreach, 66 notices to client staff, 13 manual.
Emails8401621,002350 in, 652 sent by Clara: 438 to leads and residents, 214 notices to client staff.
Phone calls583106689293 inbound, 52 outbound, 344 older calls with no direction stored.

ElevenLabs, counted independently by phone line with the same team and automated numbers removed, gives 688 answered calls, within one of the database's 689. Twilio gives 339 inbound / 1,088 outbound texts on a looser exclusion list; the text gap to the database is not fully reconciled.

Growth, month over month

Charts, growth table and the caveat
Conversations handled 0 75 150 225 300 Apr: JP&Co 114 114 Apr May: JP&Co 133 133 May Jun: JP&Co 125 125 Jun Jul: JP&Co 166 166 Jul Aug: JP&Co 176 176 Aug Sep: JP&Co 145 Sep: Western Slope 110 255 Sep

Stacked by customer:JP&CoWestern Slope

Phone calls 0 50 100 150 200 Apr: JP&Co 80 80 Apr May: JP&Co 106 106 May Jun: JP&Co 99 99 Jun Jul: JP&Co 96 96 Jul Aug: JP&Co 108 108 Aug Sep: JP&Co 86 Sep: Western Slope 81 167 Sep

Stacked by customer:JP&CoWestern Slope

Tours booked 0 10 20 30 40 Apr: JP&Co 7 7 Apr May: JP&Co 13 13 May Jun: JP&Co 4 4 Jun Jul: JP&Co 20 20 Jul Aug: JP&Co 22 22 Aug Sep: JP&Co 15 Sep: Western Slope 18 33 Sep

Stacked by customer:JP&CoWestern Slope

Ways to say it

+70%
tours, last 30 days vs the 30 before
34 vs 20
+61%
conversations, last 30 days vs prior 30
281 vs 174
+79%
phone calls, last 30 days vs prior 30
190 vs 106
4.7×
tours per month, September vs April
33 vs 7
2.2×
conversations per month, September vs April
255 vs 114
1 → 2
customers live since April
Situs Group connected, ConAm onboarding
Per month, both customersAprMayJunJulAugSepSep vs AugSep vs Apr
Tours booked7134202233+50%+371%
Conversations handled114133125166176255+45%+124%
People talked to889498129117182+56%+107%
Phone calls801069996108167+55%+109%
Texts sent by Clara262878230209265+27%+919%
Emails sent by Clara488129104164199+21%+315%

What else to highlight

Eight more things worth saying
  • Camellia occupancy: from 82% to 92%. Camellia (120 units) sat at 82.5% occupied on Aug 1 and bottomed at 83% in mid-September. Today it is 88.3% physically occupied and 91.7% trending (occupied plus one move-in, two approved and one pending application), the figure the PropFlow dashboard shows. Say "from the low 80s to 92% in two months".
  • Clara answers in seconds. Median reply to a text is 10 seconds (90% under 24 s); to an email 16 seconds (90% under 35 s), measured on the inbound messages she answered.
  • Half of the work is after hours. 48% of inbound texts and emails arrive outside 9–5 Central or on weekends (57% of emails). 20% of calls do. All of it is answered outside office hours.
  • She hands off when a person is needed. Calls are transferred to a human when one is needed; at Western Slope the destinations are a maintenance call center and a leasing ring list.
  • Footprint. Two companies live: Camellia (JP&Co) and Western Slope (about 90 buildings). Situs Group (52 buildings) connected its AppFolio in the last week of September. ConAm is in onboarding on Yardi, our second property-management system. Yale 25 Station is set up but has almost no volume yet.
  • Four ways in. Phone, text and email are live. Video Clara (a live face on a call, can join Zoom, Meet or Teams) is built and deployed, dark, pending first customer. Say "experimental".
  • Hands-free inspections. A Ray-Ban Meta glasses prototype captures move-out photos straight into a turnover inspection, proven on real hardware in developer mode. Say "prototype".
  • 18 integrations built, including Twilio, ElevenLabs, Microsoft 365, Gmail, SendGrid, Stripe and Tavus. AppFolio is live; Yardi and RealPage are built or in progress, not live.

Where this goes next

Scattered portfolios, other asset types, affordable housing

One phone line for a whole scattered portfolio

Camellia is one building. Western Slope is about 90 buildings, many of them single-family homes spread across Grand Junction, run from one office. Clara now answers one phone line and one mailbox for all of them and works out which home a caller means. That is the "distributed portfolio" architecture: the management company is the wall around its data, every phone line, mailbox, calendar and setting hangs on either the company or one of its buildings, and one lookup resolves a caller to a company, then to the buildings in reach, then to the person. A whole portfolio onboards from one AppFolio pull.

  • Status. Designed by Gera in early September; 113 of 179 build steps merged by Sep 13; serving Western Slope's portfolio since Sep 18. Still to finish: the company wall enforced at the database layer, and calendars and policies moved from building to company level.
  • Why it matters. Most management companies look like Western Slope, not Camellia. This is what lets one Clara line serve a company with hundreds of doors in dozens of places.

Other asset types

  • Single-family rentals: live. Western Slope's first single-family leads came in on Sep 21. The first week surfaced five defects (replies filed to the wrong thread, duplicate people across email and text, a dropped handoff, mismatched fees, no per-person pause); all were fixed in the same week.
  • Mixed-use: named, not specified. On the list, no design or date. Commercial and mixed-use units in Western Slope's portfolio are deliberately kept out of Clara's scope today.
  • Not planned. HOA, student, senior, self-storage and short-term rentals do not appear in the product vision or any plan. Say "multifamily and single-family" to investors, nothing wider.

Affordable housing

  • Real demand already in the portfolio. Western Slope manages two income-restricted properties in Grand Junction (Courtyard Apartments, project-based vouchers, and Lincoln Apartments). Situs Group's portfolio is being checked for the same.
  • Built so far. The design for an affordable-housing layer (program tags per unit, income-limit rules, what Clara may and may not say about eligibility, a human-compliance flag) was written Sep 11. The first piece, syncing each unit's program tag from AppFolio, merged dark the same day. The portfolio dashboard already benchmarks subsidized portfolios against the 2024 LIHTC occupancy median.
  • Status. Parked on Sep 13: Courtyard is fully occupied with no prospects in a year, so there is nothing to prove it on yet. Six more pull requests are specified and waiting on three product decisions. Say "designed and started, resumes when a customer has affordable vacancies".

Engineering velocity

Pull requests, deploys, lead time

The repo started on Feb 17, 2026, so everything below is about seven and a half months of work by a small team directing AI agents (two people opened 99.6% of the pull requests).

12,800+
pull requests merged to production this year
5,600+
in September alone, about 188 a day
5.1×
merges, last 30 days vs the 30 before
6,563 vs 1,293
0.7 h
median time from PR opened to merged (September); production follows within 30 min
~89 / day
production deploys in September (2,658)
97%
of September commits co-authored by an AI agent
MonthFebMarAprMayJunJulAugSep
Changes merged to main2691,2577081,1679412,2521,3585,648
Pull requests merged024991,1679412,2521,3585,647
Median open → merged––0.2 h0.4 h0.4 h0.4 h0.7 h0.7 h
Production deploys––5151,1609381,7329882,658

Feb and Mar were direct pushes before the pull-request workflow existed. Deploy records come from Vercel and are raw counts, not filtered for failed builds; from Sep 24 deploys are batched to one every 30 minutes, so deploys fall below merges. About 1.5% of merges were later reverted (220 of 14,565), a rough proxy for change-failure rate. GitHub's own search puts merged pull requests slightly higher (13,030 this year, 5,708 in September, UTC months); the table uses the git history count. Lines of code and test counts are deliberately left off: they are inflated by generated files and say nothing an investor should weigh.

Recount vs the first count

What moved and why

Fede asked for the numbers to be redone from scratch. The second count used its own code, re-read the production table, and bucketed months in Central Time. It then removed two phone numbers the first count had left in: a team test phone and an alarm-system auto-dialer that rang Camellia's line about once a weekday and said nothing (173 silent calls). Texts, emails and tours did not move; conversations and calls did.

MetricFirst count (Trinity)RecountWhy it moved
Tours booked103101Two tours belonged to conversations with a team phone number on them.
Conversations handled1,127945173 silent auto-dialer calls and 8 team-phone threads removed, plus the extra phone and email-domain checks.
Texts in / sent299 / 897305 / 902New rows written after the first scan, minus 1 blocked draft that was never sent.
Emails in / sent351 / 660350 / 6526 blocked drafts and 5 rows from team email domains removed.
Emails to leads and residents~580 (estimate)438 (exact)The first count guessed the staff share. Counted by message kind, 214 of Clara's emails are notices to client staff, 98 of them the "tour confirmed" email to the leasing desk.
Phone calls870689173 auto-dialer calls and 8 team-phone calls removed; older calls bucketed by Central day instead of UTC day.

Two provider cross-checks land on the same number: ElevenLabs shows 688 answered calls with the same numbers removed (database 689), and Twilio's inbound call counts for Camellia match ElevenLabs month by month. Separately, Twilio's lifetime text total is inflated by about 5,000 texts sent to one test number in March, mostly failed, before the first real customer message on Apr 6.

Method and caveats

Sources, exclusions, open questions

Working files: the recount report, monthly JSON, provider cross-check and engineering report live in the app repo's ignored tmp-artifacts folder on Fede's Mac (ytd-recount-2026-10-02.md, provider-crosscheck-2026-10-02.md, engineering-velocity-2026-10-02.md).

PropFlow Docs