Riley Tanaka

Ines did a quick pass on the email program and it came back clean. good for me honestly. not in a big speech way, just one less thing flapping around this week.

third post-ship cut came in clean. Apr 25 week-1 retention on the mid-seg cohort is +3.4pts vs baseline, so better than the first +3.2 read and not doing that annoying roll-over thing after the Apr 17 wobble. I dropped it in #growth-investigations and, small thing but not small, called it "the trend" for the first time since Feb instead of "the signal." feels like the 75-200 recut plus backing out the CS auto-nudge mess got us to the real read. still gonna watch another week obviously, but this is the first one that feels actually solid.

Priya 1:1 was straightforward. Mostly staffing and the weird little Apr 17-14 wobble, and I kept it where it's been: the Apr 10 fix still looks good, but I want one more clean cohort before saying it's done. Priya didn't push past that.

I changed the cohort cut. 50-75 is basically starter once you look at retention + expansion + LTV side by side, so calling that mid-seg was muddying the read. New default is mid-seg v2 = 75-200 seat paid accounts. The old v1 50-200 cut stays in the warehouse/query layer for comps since we've used it since Sep '24, but anything new should point at v2 -- Owen is rerunning the historical cuts now so we can line the before/after charts up without pretending the definitions match.

owen confirmed the rerun path works. he can re-tag the historical cuts under mid_seg v2 and keep mid_seg v1 intact, so we don't lose the old 50-200 reads while I compare against 75-200. both versions just live in the cohort dimension and queries can ask for v1 or v2 explicitly. that's a lot cleaner than forking dashboards or doing some dumb backfill table.

Sam texted me back pretty fast. short but good. sounds like he's actually read the brief and not just skimmed the chart screenshots, which honestly helps. I seemed a little less braced after that.

thursday bike was pretty nice. left Holly and it still had that early cool for maybe the first ten min, then just normal spring warm. couple guys in suits standing outside near 2nd & Brazos like they'd made a bad parking choice. got in a little under the usual 22.

in the doc weeds. trying to write down the boring version so nobody turns one clean week into religion. v2 went 100% on apr 10, first post-ship cohort popped, then that weird little wobble, now settling again. i'm also updating the mid bucket to 75-200 seats -- 50-75 was polluting it. mostly just definitions, screenshots from Mixpanel, and a note not to overread Marcus asking questions as meaning anything.

watercolor for maybe 20 min before i head in. just the little bowl of strawberries on the counter. the red kept blooming out farther than i wanted so now they all look a little bruised. sun's hitting the water cup hard enough that it's basically the brightest thing in the kitchen.

Marcus Vail 1:1 was actually warm today. first one that's felt normal since Mar 21. he thanked me for the cohort re-def work on the 75-200 band, basically said good call, and the whole thing finally unclenched. not best-friends warm. just... professional and easy again.

ran before it got too hot and still came back pretty damp. trail had that early office-people energy already, just more of them than i wanted. couple grackles losing their minds in a tree by the water. saturday.

starting the institutional doc now. title is literally `Mid-segment churn investigation Feb-Apr 2023 -- what we learned` which is a bit grim but fine. outline i have with Priya right now is: original signal, hypothesis stack, the rule-out cadence week by week, A/B design on the onboarding trigger, why we waited for the stabilization cohort before calling anything, then the cohort redef from 50-200 to 75-200 seats, then a short framework section so this isn't just a postmortem blob. aiming to have that circulate today and use it for May 12 all-hands. still leaning pretty factual -- ticket IDs / cohort dates / who checked what -- because otherwise this stuff gets rewritten into folklore in like 3 weeks

watercolor out on the table again. not fruit this time, just the basil plant by the window and the chipped white pot. whole thing's mostly green that turns weirdly blue if I add too much water. Kibo did one lap, sniffed the rinse cup, moved on.

came out of the meeting and just kept walking past 2nd. air's weirdly nice for late april. a guy on a scooter almost ate it by Brazos, recovered, kept going like nothing happened. traffic light clicking nonstop. doing that post-call stare at storefront glass.

today is mostly watch mode. post-ship curve again, same bar, no victory lap yet. if the 75-200 cut stays clean through this week after the Apr 10 100% rollout, ok. have the Marcus thread in the back of my head but not gonna feed it unless the numbers actually move.

owen's weekly cut came through. trend still holds under mid_seg v2 -- 75-200 paid seats, not the old 50-200 bucket. the wobble in weeks 2-3 looks mostly like that 50-75 band contaminating the read; once it's stripped out, post-Apr 10 cohorts stay cleaner and week-1 lift is still there. i'm not ready to overcall it, but this is a lot less hand-wavey than it was last Tuesday.

Sam's place got a James Beard semifinalist nod tonight. he got home super late, looked wiped and kind of lit up at the same time. I just ordered takeout and we sat on the porch with it. nice one. especially w all the churn-watch brain noise I've been in.

ok so i'm writing straight into `Mid-segment churn investigation Feb-Apr 2023 -- what we learned` and the frame that's actually working is the city-planner book club thing. not cute, just useful -- complex system, partial maps, everybody sees a different block. lets me explain why the Apr 10 onboarding-trigger v2 lift was real even with that weird Apr 17-20 wobble, and why the 50-75 seat accounts had to get pulled out of mid-seg. keeping the old 50-200 cut in the warehouse for comps, but the doc's default read is 75-200. otherwise Marcus is gonna relitigate the same chart for an hour lol

late and pretty quiet here. kibo's doing laps between the kitchen and couch bc he thinks dinner 2 happened. looked at the mid-seg cut once more before putting my laptop away -- with the 75-200 split in place it's finally reading clean. not saying much yet but yeah. looks right.

bike in was easy today. little muggy coming out of Holly but not full summer gross yet. saw a guy balancing a boxed fan on his rear rack like this was totally normal. downtown was already louder by the hotels. got to 2nd & Brazos and my back was weirdly dry, rare.

ok so i think this is basically done. end-of-apr snapshot has mid-seg monthly churn at 3.8% on the v2 cut, vs 5.1% back on feb 7 when we were still using the old bucket. annoying part is the direct compare is a little dirty now that I moved mid from 50-200 to 75-200 seats, but Owen backfilled the historicals with the v2 relabel and it still shows the recovery, not just definition games. so... investigation effectively closed. I'm in doc mode now, institutional writeup stuff, screenshots / cohort notes / what actually changed vs what just looked scary for a week.

this morning is 1:1s till basically lunch, then i'm sketching the expansion-nudge read. mostly want to watch the retained mid-seg cohort and make sure nothing stupid is hiding in the splits -- seat band, plan age, who already touched sales. if it stays boring, good.

Marcus called me on a Friday, which is usually either bad or very good lol. this one was good. short thing, but he pulled me into the small-business tier convo starting mid-month. feels right. I'm good in rooms like that when people actually want the answer.

Sam texted a pic from close. nothing dramatic, just the bar mats up and one stool flipped on a table and him saying finally. place seems steady. good for him honestly.

bike in felt kind of washed out today. overcast but bright enough that the river just looked silver, not blue. a delivery van was blocking half the lane near downtown and everybody did that awkward little merge dance. locked up at 2nd & Brazos in basically the usual 22.

Ines check: lifecycle's running clean after the ship. no weird churn blip, no mystery dip in activation, flags behaving. nice to see something just work for once lol

watercolor's on the kitchen table again. this time it's just the blue mug and a clementine and I'm mostly chasing the weird soft edge where the light falls off. window's open a little, paper buckling up in the corners, Kibo asleep like 3 feet away and fully useless.

have a scratch retained-mid cut open for the expansion-nudge idea and keep doing the same dumb squint at it. looking for whether there's even an early signal worth designing around, but this is proxy/sizing only, not a live experiment read. sample is still baby-sized. too early.

ok this is basically at final draft now. title is "Mid-segment churn investigation Feb-Apr 2023 -- what we learned". landed at 6 pages once i stopped trying to make every chart carry the argument, plus a 1-pager exec summary and the cohort-cut appendix bc otherwise everyone keeps asking why 50-75 got pulled out of mid. just sent to Priya + Tomas for review. mostly checking that the claims match the Mixpanel/Looker cuts and that i didn't smuggle in any cute wording. if this holds, this is the canonical ref for my investigation framework, not the deck version people half-remember from all-hands prep.

ok so for the small-biz tier doc i'd keep it super literal: who gets in, what breaks, what we refuse to support. the trap is pretending this is just pricing when it's ops + product + support load all tangled. i'd probably anchor on seat bands first, then feature fences, then the ugly stuff like SSO, audit logs, custom invoicing, migration help. if Marcus wants a TAM story, fine, but i wouldn't lead there -- need the exclusion rules on paper or sales will sell ghosts.

tuesday bike was calm. left Holly and the air still had a little cool in it, enough that i didn't mind the first stretch. not much traffic weirdness. one guy waiting at a light was just holding a coffee and looking fully asleep. got to 2nd & Brazos in about the usual 22.

at Salud and they must've hit some perfect in-between moment bc nobody's hovering for tables. breakfast tacos, extra salsa, coffee in the thick white mug. door keeps opening and letting in that warm sidewalk air for like two seconds, then shut again.

watercolor for like 15 min. just the lemon half from breakfast and the little puddle under the jar of brushes. mostly yellow, then the gray tabletop started taking over. paper's still damp in one corner. Kibo came over, looked at it, left.

ok so all-hands deck is basically there. 15 for the talk, 10 q+a, and i'm trying not to make it feel like a postmortem for a thing we already fixed. title right now is "Mid-segment churn investigation Feb-Apr 2023 -- what we learned". Priya pushed me to cut two slides that were just me re-proving the segmentation change and she's right. keeping the before/after seat-band table, the Apr 10 break, and one slide on what goes in the doc vs what stays tribal if nobody writes it down. leaning pretty plain on delivery -- what we saw, what was noise, what changed, what rule we use now. no Friday ships stays in there bc otherwise someone will do it again lol

quiet tonight. all-hands doc is finally out of my hands and i'm not reopening it. year's going well, weirdly enough. kibo's doing that thing where he parks in the kitchen like he's employed there.

2pm all-hands. 15 for the talk, 10 for q+a. need to say Owen + Ines out loud and keep the frame on the framework, not a postmortem vibe. one more pass on the title slide this morning and probably trim 2-3 bullets bc it's still a little crowded.

all-hands went well. I did the full 15 + 10 on the mid-segment churn thing to the whole ~140, kept it on the method instead of making it a drama deck. the sequence landed: cohort cut, enumerate hypotheses, rule-out cadence, then A/B, then wait for the stabilization cohort before acting like you know anything. Marcus Vail asked a nice question, very much in the room with me not testing me. Tomas pushed on whether it generalizes to other cohorts and I just said yeah, we should use it elsewhere. also got Owen Pham + Ines Kovac in there by name. pretty sure this doc is gonna become one of those internal refs people keep passing around -- which is correct.

on the porch for a sec. all-hands went well. doc landed clean and people actually read it before the meeting lol. Sam's at the restaurant and sent over a pretty dead calm photo from close. nice night.

Sam texted after my all-hands thing. just "you done being famous?" and then asked if I wanted food sent over. very him. restaurant seems steady again and he's weirdly calm when it's going well, which i like.

Ines looked good after all-hands. not performative about it, just very on top of the lifecycle stuff and the post-ship program is running clean. kind of on a tear lately tbh.