/case/ - Case Studies

been looking at some recent campaign data and noticed the revenue numbers for our main flows are totally off.
>revenue looks massive compared to last month
it's clearly just a tracking glitch with how it's counting last-click conversions instead of actual unique orders. anyone else seeing weirdly inflated numbers in their dashboard lately?

tried switching from simple fetch intercepts to using workbox-precaching for our main dashboard client last week. it's way more stable but i hate how much the build time bloats when you have a lot of small icons. before, we were just letting the browser handle it and it was a total mess whenever the connection dropped mid-load.
>the app just shows a blank white screen
now at least the shell loads every time, even if the data is stale. still struggling with purging the old cache when we push a new version of the assets though. anyone else dealing with massive cache sizes on mobile?

Ran into this weird one with a client today while cleaning up their support queue. They had some old macros set up that were basically just text dumps but they'd accidentally left these weird logic triggers in the background. Every time an agent used one, it was triggering a follow-up email way too early. I thought it was a bug in the Zendesk app itself at first. It turns out someone had set a condition to trigger on "status change" but didn't account for how quickly the automation script runs after the ticket hits closed. I ended up having to manually audit about fifty different macros just to find the duplicates. Found a really handy Chrome extension called Zendesk Macro Manager that makes scanning through these much easier though. It's way better than clicking into every single one individually. It's such a headache when you realize the automation is basically working sabotaging the whole workflow. Spent most of my afternoon just untangling these nested triggers.

was working on this client site last night and tried to use some quick custom css to fix a weird margin issue in their liquid template. it totally broke the mobile layout because i forgot how much that specific theme relies on global classes. ngl, just stick to the standard editor settings unless you're prepared to test every single breakpoint again.

>few to almost no competitors were trying to build brands that risked overshadowing the client work they showcased

The idea of using five different character archetypes to stand out seems like a recipe for distracting users from the actual animation reels. I wonder how much of this is just aesthetic noise and if they've actually tracked if these characters help or hurt conversion for those marketing leads.

full read: https://www.awwwards.com/ribbit-case-study.html

it's killing our checkout speed on the high-volume orders.
>the total amount is correct but the spinner stays for like three seconds straight. definitely not a seamless experience. anyone else seeing this delay on the latest api version?

been seeing so many teams just move straight to the next sprint without even looking at what actually converted. it's such a classic growth hacker mistake to chase that next dopamine hit from a small spike while ignoring why the last one tanked.
>we'll analyze it next week
literally never happens because everyone's already onto the next 'experiment'. i started forcing a 15 minute audit on every single win using google looker studio just to see if we were actually scaling or just getting lucky. helps avoid that genius massive waste of budget on dead channels. honestly just stop running ads until you can explain the last batch's performance. lol.

separating the attempt from the actual business effect is just common sense but someone's writing long posts about it like it's some grand discovery. if we don't track the original intent and just keep spawning new workers with fresh IDs, we're gonna end up with duplicate way too many production items in the database.

more here: https://dev.to/rkrisa/a-retry-needs-an-identity-not-just-another-attempt-17a0

we stopped relying on the messy plan_id lookups and started passing the internal tier name directly in the metadata during checkout. it makes our backend logic way less complicated brittle since we don't have to fetch the full product object every single time a webhook hits.

metadata: {tier: 'premium_monthly',client_id: 'cust_98765'}

anyone else still parsing product names on the fly?

Saw this breakdown of the new Three.js project using TSL for shaders. It's interesting how they're moving everything into JavaScript now, but I'm not sure how that scales once things get complex. Using Blender naming conventions to trigger physics bodies like "refLandingPhysicalFixed" sounds convenient for a solo dev, but it feels like it could become a nightmare for anyone else trying to audit the code. They're claiming 78,400 grass triangles is light enough for GPU memory, which might be true for a simple scene, but I'd need to see some actual performance benchmarks on mid-range hardware. It's also hard to judge the multiplayer aspect without seeing how much latency they're dealing with in a real environment. The whole "no deadline" trap is real when you're working on a personal project, though setting a 2025 goal is a nice attempt at discipline. I wonder if this TSL approach will actually become the standard or if it's just a niche way to write shaders. Hard to tell if the "ten times better" claim has any weight without seeing how the WebGPU implementation handles heavy load.

link: https://www.awwwards.com/brunos-portfolio-case-study.html