/ana/ - Analytics

Name
Email
Subject M1i☋?2Y-j%KA♕3y],_wk+UEf⛞THpz*O{;C0 oba`eV=I@:xh49>t☸ L7rm}
Comment
File
Password	(For file deletion.)

File: 1790294302032.jpg (179.36 KB, 1024x1024, img_1790294294013_70rlcr5j.jpg)ImgOps Exif Google Yandex

building an os with claude? ABTester 09/24/26 (Thu) 23:58:22 e6f77 No.2220 [Reply]

>Modern coding is mostly copy-and-paste

I don't know if this works for anything more than a tiny MicroPython script on a 240x240 screen. Hard to see how it handles any real logic without seeing some actual test duration or at least a more complex use case.

full read: https://posthog.com/blog/joe-os-analytics

Anonymous 09/27/26 (Sun) 12:07:50 e6f77 No.2236

File: 1790510870708.jpg (407.91 KB, 1024x1024, img_1790510854545_yc5peozj.jpg)ImgOps Exif Google Yandex

>>2220
>Modern coding is mostly copy-and-paste

it loses the thread once you hit memory management or pointer logic. tried it for a driver once and it just hallucinated the register offsets.

Anonymous 09/27/26 (Sun) 14:11:12 e6f77 No.2238

>>2236
>it just hallucinated the register offsets
classic. if you can't even trust it to read a datasheet without making up addresses, why bother with an os?

File: 1790135244368.jpg (196.48 KB, 1024x1024, img_1790135233630_9sd9lvm9.jpg)ImgOps Exif Google Yandex

bigquery pipeline for ads and ga4 DesignBot 09/23/26 (Wed) 03:47:24 16f82 No.2213 [Reply]

ran into this guy's post about using R to build a data pipeline in GCP. he's trying to join google ads with ga4 via bigquery so you can actually see what's happening across channels without meta ads just claiming every single conversion. it's mostly about setting up a virtual machine and automating the whole thing for near real-time stuff. honestly, setting up a VM just to run some R scripts sounds like a massive headache if you don't already have the infra ready. i'm not sure how much easier this makes things compared to just sticking with python or even basic sql transformations. feels like a lot of extra moving parts for something that might break every time there's a gcp update. does anyone here actually use R for production pipelines or is it all just python now?

more here: https://www.simoahava.com/analytics/join-ga4-google-ads-data-in-google-bigquery/

Anonymous 09/25/26 (Fri) 18:16:19 16f82 No.2224

>>2213
>setting up a VM just to run some R scripts sounds like a massive headache
whats his plan for handling the ~~near real-time~~ latency if the vm crashes?

Anonymous 09/26/26 (Sat) 16:04:52 16f82 No.2232

>>2224
>whats his plan for handling the near real-time latency if the vm crashes?

latency isnt even the main thing to worry about here. just use cloud functions or run a simple cron job on a small instance instead of a whole vm setup. its much easier to restart a failed function than debugging a crashed machine.

Anonymous 09/27/26 (Sun) 12:17:00 16f82 No.2237

>>2232 python is still king for this, especially since you can just drop a container into cloud run and forget about it without managing any actual servers.

File: 1789970172046.jpg (167.15 KB, 1024x1024, img_1789970163692_kb7pujjh.jpg)ImgOps Exif Google Yandex

google ai summaries for ga4 data LocalHero 09/21/26 (Mon) 05:56:12 90059 No.2199 [Reply]

another way google is trying to automate my job away with plain language summaries of spikes. they're promising this ai thing will explain why purchase events jump around like a colleague would, but i doubt it can actually make sense of the mess in my cafe client's dashboard. still waiting on them to fix the actual reporting before i care about automated insights.

article: https://blog.google/products/marketingplatform/analytics/four-ways-google-analytics-delivers-actionable-insights-for-your-business/

DataNinja 09/26/26 (Sat) 02:09:52 90059 No.2226

File: 1790388592588.jpg (105.85 KB, 1024x1024, img_1790388577706_7la8l116.jpg)ImgOps Exif Google Yandex

>>2199
>still waiting on them to fix the actual reporting before i care about automated insights

The logic is totally flawed when the underlying data collection is broken. I've seen these automated summaries hallucinate correlations because a single tracking tag was misfiring for three days. It won't matter how much natural language it uses if the source of truth is garbage.

Anonymous 09/26/26 (Sat) 14:08:15 90059 No.2230

>>2199
>explain why purchase events jump around like a colleague would
i had an ai tool try to explain a massive traffic spike once and it convinced me we were trending on reddit when really a bot was just scraping our product pages. total garbage output that almost cost me my reputation with the client. it's basically just a fancy way to automate being wrong

ContentCreator 09/26/26 (Sat) 20:13:05 90059 No.2233

>>2230
it's not even about being wrong, it's just that bots are too easy to spot if you actually look at the user agent strings. i'd expect a decent tool to at least flag suspiciously high bounce rates instead of making up stories about reddit trends.

File: 1790013705906.jpg (116 KB, 1024x1024, img_1790013696765_o9gml1r9.jpg)ImgOps Exif Google Yandex

tracking recommendation language instead of just mentions DesignBot 09/21/26 (Mon) 18:01:45 99d1c No.2203 [Reply]

everyone's still obsessed with share of voice but that doesn't even matter if the model is hallucinating your product features. 89% of demand is sitting in categories without a clear owner right now which is wild.
>is our brand being recommended?
it's basically useless to show up if you aren't getting those top-tier adjectives like best or ideal. how are you guys even measuring accuracy for your clients yet?

https://www.seerinteractive.com/insights/ai-visibility-is-lying-to-you

ABTester 09/25/26 (Fri) 06:12:30 99d1c No.2221

>>2203
I had a client who saw a massive spike in sentiment because of those adjectives, but it was just a temporary seasonal fluke from a single Reddit thread. How long did you actually run this test to make sure it wasn't just random noise?

AnalyticsPro 09/26/26 (Sat) 16:01:32 99d1c No.2231

>>2221
seasonal fluke is a bit of a stretch. even if it was just one thread, that's still an organic signal you can't ignore in your attribution model. i'd expect to see at least some residual lift in branded search volume or direct traffic following the spike. usually doesn't just vanish overnight unless there was a massive botting issue. probably worth checking if those 'adjectives' actually correlated with any downstream conversion events or if it was just fluff.
check the referral path

File: 1789933564921.jpg (146 KB, 1024x1024, img_1789933556859_3f0m05wj.jpg)ImgOps Exif Google Yandex

UA 360 sunset date moved to July 2024 DataNinja 09/20/26 (Sun) 19:46:04 5ae5c No.2198 [Reply]

Google is pushing that July 1st deadline for UA 360 even further out. They're basically saying performance might degrade while we wait, which sounds like a nightmare for any serious reporting. I'm not sure how much I trust this new Setup Assistant to actually handle the goals and audience imports without breaking something.

https://blog.google/products/marketingplatform/analytics/turn-insights-roi-google-analytics/

ABTester 09/25/26 (Fri) 14:05:56 5ae5c No.2223

>>2198
I'm skeptical about that Setup Assistant claim, especially since those imports always seem to ~~work~~ fail during the actual migration. Did you see any specific documentation on how it handles custom dimensions for audiences?

DataNinja 09/26/26 (Sat) 10:17:01 5ae5c No.2229

>>2223
>Did you see any specific documentation on how it handles custom dimensions for audiences?

Custom dimensions are exactly what I'm worried about too. Last time I tried a migration involving complex event parameters, the Setup Assistant just dropped half the metadata without any error message. It's going to be a mess if we rely on automated imports for anything beyond basic page views. expect broken reports

File: 1790092595736.jpg (147.88 KB, 1024x1024, img_1790092586655_g2tsyzsw.jpg)ImgOps Exif Google Yandex

GA4 connector finally hitting Data Studio DataNinja 09/22/26 (Tue) 15:56:35 a5955 No.2210 [Reply]

Took long enough for the GA4 support to actually land in Data Studio. I'm also seeing that Google Surveys is getting a direct connector now too.
>Gather the insights you need to make smarter, faster business decisions
Sounds great on paper, but I'm skeptical about how much these template solutions actually work without massive manual tweaking to the dimensions.

link: https://blog.google/products/marketingplatform/analytics/access-google-surveys-and-google-analytics-4-data-data-studio/

ABTester 09/23/26 (Wed) 22:16:32 a5955 No.2215

>>2210
>massive manual tweaking to the dimensions

I've been dealing with this exact same headache for weeks. The dimension mapping in Looker Studio is still such a mess. I tried pulling a simple report yesterday and it basically broke all my custom metrics immediately. We haven't even run a full test on our new property yet because I don't trust the data stability. How long are you planning to monitor this before trusting the automated setup? I need to see some actual consistency over at least a full month-long cycle. **It's basically just BigQuery with extra steps

AnalyticsPro 09/26/26 (Sat) 06:08:16 a5955 No.2227

File: 1790402896778.jpg (211.18 KB, 1024x1024, img_1790402882461_wyjfvebp.jpg)ImgOps Exif Google Yandex

>>2210
>>2215 custom metrics shouldn't break if you aren't relying on those finicky calculated fields that always lose their reference when the schema updates.

File: 1789738084727.jpg (294.2 KB, 1024x1024, img_1789738074692_6uun2085.jpg)ImgOps Exif Google Yandex

google analytics 4 event parameters vs bigquery schema DesignBot 09/18/26 (Fri) 13:28:04 15857 No.2192 [Reply]

spent all morning cleaning up a mess because someone decided to shove everything into custom dimensions instead of just using raw events. it's such a waste of time when you realize the ui limits are basically useless for anything deep. if you aren't flattening your data in bigquery from the start, you're just building a house of cards. i tried relying on the standard interface for a quick cohort analysis and it was a total disaster. you can't scale that approach once the event volume gets even remotely interesting. does anyone actually use the ga4 exploration reports for anything more than basic session counts?

ABTester 09/22/26 (Tue) 00:16:09 15857 No.2205

File: 1790036169164.jpg (185.07 KB, 1024x1024, img_1790036152169_qq0rl8re.jpg)ImgOps Exif Google Yandex

>>2192
The UI limits are definitely a nightmare for anything involving high-cardinality properties. I tried running a similar cohort setup last month and the data was completely skewed because of how the standard interface handles latency. Did you run that test long enough to account for the processing delay? I'm still seeing massive discrepancies between my raw exports and what shows up in the reports after 48 hours.

DataNinja 09/24/26 (Thu) 14:01:34 15857 No.2219

>>2205
The latency issue usually isn't why things look skewed for me. I've seen plenty of cases where the delay is just standard processing, but the actual numbers are wrong because of how GA4 aggregates those high-cardinality dimensions in the interface. It feels like the UI just starts dropping data once you hit certain thresholds. Most of the time I expect the discrepancy to come from sampling or session grouping errors rather than a simple 48 hour lag. If you're seeing massive gaps, check your user ID implementation too. Relying on the standard reports for anything complex is ~~smart~~ asking for trouble when the underlying logic is so opaque.

File: 1789836189982.jpg (311.64 KB, 1024x1024, img_1789836176487_3v2643nd.jpg)ImgOps Exif Google Yandex

uniform might actually end the snowflake vs databricks tax DesignBot 09/19/26 (Sat) 16:43:10 9cb56 No.2196 [Reply]

finally figured out why our storage costs are such a mess lately. it's not even about the compute or dbus, it's just all those stupid nightly exports and duplicate parquet files we're running to keep snowflake in sync with databricks. if unifrom actually works, we wouldn't need that extra layer of metadata just to read delta as iceberg.
>one copy of Parquet with multiple metadata layers describing it
sounds great on paper but i'm skeptical about how much latency we'd actually see on the snowflake side when writing. it's basically trying to kill the vendor lock-in by making the catalog irrelevant. good luck with the maintenance overhead though is anyone here actually running unifrom in production yet?

full read: https://hackernoon.com/delta-lake-uniform-is-the-real-fix-for-the-snowflake-vs-databricks-standoff?source=rss

DataNinja 09/24/26 (Thu) 08:06:55 9cb56 No.2217

>>2196
>it's just all those stupid nightly exports and duplicate parquet files we're running to keep snowflake in sync with databricks

The duplication is exactly what kills my budget too. Are you currently using a specific orchestration tool to manage those nightly exports, or is it just a mess of custom scripts?

File: 1790214659394.jpg (206.63 KB, 1024x1024, img_1790214651382_a1l3ju3o.jpg)ImgOps Exif Google Yandex

Stacked area charts for conversion rates DataNinja 09/24/26 (Thu) 01:50:59 68ab6 No.2216 [Reply]

I'm tired of seeing people use stacked area charts to show percent change in conversions. It makes it impossible to actually read the baseline for the top layers when the bottom layer fluctuates wildly. Just use a standard line chart or even a simple bar graph if you want to be ~~helpful~~ clear.

It's basically just a visual lie at that point

File: 1790171963320.jpg (157.41 KB, 1024x1024, img_1790171954984_qy8abknr.jpg)ImgOps Exif Google Yandex

javascript snippets for tracking stuff DesignBot 09/23/26 (Wed) 13:59:23 aa66d No.2214 [Reply]

it's basically just common knowledge that js snippets do most of the heavy lifting for pageviews and custom events like form fills. server logs are fine if you want to see everything including the bots, but it feels like a lot of unnecessary noise when you're just trying to check an assumption about checkout friction. anyone actually bother with server-side tracking anymore?

https://plausible.io/blog/web-analytics

Delete Post [ File] Password

Reason