So an interesting observation from vaccinateCA.com: we're effectively discovering availability bottom's up, starting from where the patient would perceive supply, rather than top down, starting from where a hypothetical national logistics coordinator would see supply.
That hypothetical coordinator might scratch their head at some of the results: "That's funny, I dispatched to X, Y, and Z on the same date. Why does X have it but Y and Z do not?"
A colleague, explaining our exploration strategy:
We have no prior knowledge of what the logistics system looks like for Safeway, for example, versus a public hospital versus a competing pharmacy chain.
All we know is what happens when we call, say, 10 Safeways.
We, inadvertently then advertently, put a profiler on Safeway.
When we discover a pattern, like "Holy cow, Safeways seem to have a ridiculously high hit rate", that is often downstream of some decision made at a level which is opaque to us. But we can make informed guesses.
For example, that Safeway's other brands share similar logistics.
And lo, that seems to be accurate when we surge calling attention on them.
Which is the sort of thing that you probably never once think about in your life unless you decide to blackbox reverse engineer national medical delivery infrastructure.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
You know the sales team tradition of banging a gong when you get a commit?
We've got a gong bot that flags every time we find accessible vaccine.
I like this gong so much. (That it is not audible is a plus.)
Also like the emergent culture of people reacting to it with the :vaccine: emoji.
It's fundamentally the same purpose as the sales gong, right?
You're going to get told No, a lot. That's a useful result! Pick up the phone immediately and try to get some more Noes. Occasionally you get a Yes in there. *gong*
Happy MLK Day. Here's what vaccinateCA.com hopes to accomplish today:
* Call ~500 pharmacies in California to learn their current status of vaccine availability.
* Improve website UX on mobile devices after relaunch on new tech stack at ~midnight.
* Continue learning nuances about pharmacy IVR systems to improve call efficiency, both for us and the responding pharmacist.
* Start scoping out multiple ways to expose our data, via e.g. a text-based interface, chatbot, etc, and APIs / ways to package for gov't frontline.
Picked a pin at random to check for some formatting issues. I'm just struck by how we got this info just by asking nicely.
(I'm obscuring identifying information because this information will age rapidly; vaccinateCA.com will update but a Twitter screengrab will not.)
This is *not* the sort of petabyte scale computation that the tech industry often thinks of when we think data problem. It's a very, very human scale dataset.
But as we get good at ingesting updates rapidly, we can ask it useful questions, like "Where's nearest short waitlist?"
"But how will you go from unstructured text notes to something a computer can operate on? Spin up an ML team to do natural language parsing and then..."
No it probably sounds closer to "Read from computer, type into a computer, ideally very quickly."
We called ~80 pharmacies in California today to find one that had the vaccine available.
Seventy.
Luckily, we have a lot of dedicated people calling. If someone in your life can't do that many calls, point them at vaccinateCA.com to piggyback off our work.
My kingdom for an edit button, including after I fumble-fingered the domain name.
Anyhow, it isn't entirely obvious yet what is causing the low hit rate today. We're mixing in new pharmacy chains, and therefore supply lines, after completing the census of Safeways.
We also learned a fun new thing about pharmacy hours, which is that waking up bright and early at 8:30 AM on a Sunday means you get a 90 minute coffee break before you can actually talk to anyone.
Day 4 of VaccinateCA.com (we rebranded; we couldn't remember the old domain, either).
We're shooting for 500+ calls today. That should uncover another 50+ locations with the vaccine in California to go with the 100 we have already found.
We're also relaunching the site after a rewrite in a static site generator (to allow us to greatly improve the experience on mobile, 60~80% of clients).
ETA: a few hours into the day. We'll then launch a simplified experience where ~3 facts about patient => show best result.
We're still working on getting the word out. Please feel free to send the site to older folks in California or to your networks.
If you want to help with calling, and haven't already emailed me, please do. Onboarding volunteers throughout the day, between everything else.
We've now called about 700 California vaccine sites for covid19vaccineca.com and have 98 which are actively administering the vaccine, most of which require appointments.
We will probably ~2X those numbers by tomorrow.
Priorities for rest of the weekend:
1) Continue improving processes for contacting pharmacies.
2) Complete census of Safeway; prioritize CVS/Walgreens/etc based on observe hitrates.
3) Develop repeatable process for daily refresh on pharmacies, via humans or bots.
4) Outreach outreach outreach; we are getting to the point where we are better at data collection than getting data in front of people.
5) Improve usability on mobile; represents majority of clients.