Almost every first call we take includes some version of the same question: the data is public — why can't I just download it instantly?
We used to answer with file sizes. This time we went further: we pulled the wall-clock logs from every extraction job our machines have run — 155 measured runs across 39 payers — and benchmarked how long these files actually take to scan, payer by payer. Here is what the logs say, as of August 2026.
First, the sizes (context, not the story)
Add up the in-network files for the payers we track and you get about 108 terabytes of compressed, nested JSON — wildly lopsided. UnitedHealthcare alone publishes about 63.5 terabytes across roughly 6,800 files — 59% of the entire corpus. With Aetna (~22.6 TB) and Anthem (~13.8 TB), the three largest payers are 92% of the data. Meanwhile 29 of the 34 payers we size fit under one terabyte each, with a median footprint around 98 gigabytes. (These figures move as payers republish monthly; we re-measure them from the live file inventories.)
But size alone turned out to be a bad predictor of work. That's what the job logs revealed.
What the machines measured
Our benchmark: hours of wall-clock per gigabyte scanned, computed per payer as total measured extraction hours divided by total measured gigabytes, from real batch jobs.
- UnitedHealthcare is the slowest payer per gigabyte we process: 0.88 hours per GB. Its most recent full-file run took a single 9.72 GB file and scanned it in 8 hours 10 minutes.
- Anthem is over five times faster per gigabyte (0.16 hr/GB) — but it publishes the largest single files we have ever processed: a run of 74–77 GB files that took 18.0 to 18.6 hours EACH. Those are the longest single jobs in our history. Slowest-per-gigabyte and longest-job are different titles, held by different payers.
- Regional plans are a different world: the biggest Blue Cross Blue Shield of Texas file we've measured (19.3 GB) scanned in 2.25 hours; the biggest BCBS of Tennessee file (35 GB) in about 4 hours. Most regional payers' entire corpus is smaller than one Anthem file.
The five-fold spread in per-gigabyte speed between national payers is the quiet finding here: a gigabyte is not a gigabyte. How a payer packs its provider networks into the JSON changes the work several-fold, which is why honest delivery estimates have to come from measured runs on each payer's files — not from file size and a rule of thumb.
Six machine-years, or a filter
Multiply UnitedHealthcare's measured rate by its 63.5-terabyte corpus and you get roughly 55,600 machine-hours — about six machine-years of scanning on a single computer. That number is why "just download it and look" is not a plan, and it forces an architectural choice.
You can warehouse the corpus — download everything, store it, re-ingest every monthly republication, and carry a permanent storage-and-compute bill that lives inside annual subscription pricing. Or you can filter at the source: the files already sit in payers' public cloud buckets; when someone asks a specific question — these codes, these providers, this payer — machines scan only the relevant files, in parallel, and keep only the rows that answer it. We built DeductibleData on the second model. You pay for machine-hours spent on your question, not for warehousing everyone else's.
What that means when you hit "purchase"
The benchmarks translate directly into what a buyer experiences:
- The largest delivery we have ever shipped — 32 terabytes across two national payers — completed in 40.9 hours end to end, including the aggregation and export stages after the raw scan. That is the measured worst case in our history: under two days, for a third of the entire corpus.
- Targeted requests on sub-terabyte regional payers are the opposite end: single-file scans measured in minutes, full deliveries on a same-day scale.
- And the corollary that matters most: narrowing your request makes it faster and cheaper at the same time, because both are functions of the same measured quantity — gigabytes scanned. Two billing codes at one facility is a small scan even inside a big payer.
A TPA raised the sharp version of this on a recent call: they negotiate single-case agreements, sometimes with a surgery days away — can they get rates for one or two codes at one facility fast enough to matter? That request is a narrow scan of a small footprint: the fast end of everything measured above. The slow, expensive version of this industry exists to answer broad questions; the urgent questions are almost always narrow ones.
What the number is worth
The files are public, and so is the arithmetic — anyone with enough machines and patience could reproduce every benchmark in this post. Raw access to this data is becoming a commodity, and pricing that assumes otherwise is living on borrowed time.
What doesn't commoditize is the answer sized to your question: which rates apply to your codes, your facilities, your counterparty — current vintage, delivered in a format your team can open, at a cost proportional to the question rather than to the universe. These benchmarks are the same numbers we use internally to plan every job; publishing them is the same bet we make everywhere else — that you're better served by the measured truth than by a sales pitch. We don't sit on either side of your negotiation; we make sure you walk into it knowing exactly what the file says, and exactly how long the machines will take to get it to you.