A document-processing company we've been working with runs a wildly seasonal business. During tax season their inbound document volume spikes to something like a manufacturing line running flat out, then drops off for the rest of the year. On a recent working session, the question came up: how do you buy Azure Document Intelligence when your volume looks like that? It's a good question, and the answer is not obvious.
How Document Intelligence is priced
Azure AI Document Intelligence (the service that reads structured data out of forms, invoices, and other documents) charges based on how many pages or documents you process. Microsoft offers commitment tiers: you agree to a monthly volume, and the per-document rate drops as the committed volume goes up. That's the bulk-buying discount. If you go over your committed amount, the overage is billed pay-as-you-go, which is the highest rate you can pay per document.
The key detail people miss: even when you sign an annual commitment, the threshold you're measured against is monthly. So a flat annual number does not automatically smooth out a spiky year.
Committed tiers give you a lower per-document rate. Anything above your tier for that month bills at the pay-as-you-go rate. The tier resets and is evaluated month by month.
Why a cyclical business shouldn't just pick one number
If your volume is steady, you pick a tier that matches your run rate and you're done. If your volume swings hard by season, a single flat commitment costs you either way. Set it high enough for your peak and you're overpaying for the quiet months. Set it low and your peak months rack up expensive pay-as-you-go overage.
The better approach for a seasonal shop is to schedule the tier changes. Ramp the committed threshold up before your busy months so the peak volume lands at the discounted rate, then drop it back down once the season ends so you're not paying for capacity you aren't using. You can still exceed the lower tier in an off month if something unusual comes in; that just bills as overage, which is fine when it's occasional.
Automating the throttle
Doing this by hand means someone has to remember to change the tier at the right time every year, twice a year at minimum. It's the kind of thing that gets forgotten in a busy stretch, which is exactly when it costs you the most. We treat it as a scheduled operation tied to the known season, so the threshold moves on its own and someone reviews it rather than owning it from scratch each cycle.
One more lever that changes the math: what you send in. If a meaningful share of your inbound pages are blank, cover sheets, or organizer junk, every one of those still counts as a page processed. Stripping the pages that carry no data before they hit Document Intelligence is worth real money at high volume. That pre-processing step (dropping blank pages, normalizing oddly scanned images) belongs in the pipeline before the paid service ever sees the document.
What this looks like in practice
- Know your peak-month and off-season document volume before you commit to anything.
- Match a higher tier to your busy months and a lower tier to the rest of the year, rather than one flat number.
- Remember the threshold is measured monthly even under an annual commitment.
- Strip blank and no-data pages before processing so you don't pay to read nothing.
- Schedule the tier changes so nobody has to remember them manually.
Worth thinking through before you commit
If you're standing up an AI document pipeline and your volume isn't flat, the pricing decision deserves as much attention as the technical build. We're happy to walk through your volume pattern and what a tiered commitment would look like against it. Reach out through our contact form or give us a call and we'll talk it through.