What we changed on the pricing page this quarter

We updated the pricing page twice this quarter. The changes are small. The reason we’re writing about them is that pricing-page honesty is the single most reliable tell about a software company, and I want to make our tell public.

If you’d rather read the earlier thinking first: pricing in public, 90 days in and pricing opacity as a market signal frame why we publish prices at all. This post is about what moved between October and December.

The two lines we added

A document-volume ceiling per tier, named. The original pricing page listed three tiers with feature differences but didn’t name a document-volume or usage ceiling. In practice, every enterprise plan eventually asks “how many documents can I ingest” and the sales conversation ends up negotiating the number one-off. That’s bad. It makes the published price look like a starting offer and the real price like a negotiation.

The new page names a document ceiling for each tier — 2,000 / 10,000 / 50,000 KB blocks respectively — and an overage rate for each that is literally the same number we’d quote on a call. A customer who wants to do the math themselves can do the math themselves.

We had a longer internal argument about whether to publish the overage rate at all. The argument against was “this looks punitive to new customers.” The argument for was “pricing a customer finds out about at renewal that wasn’t on the page at signup is a worse experience than any sticker-shock we’re worried about.” We took the second argument.

A named, documented metered line item for AI costs above the baseline. Our tiers include a baseline of AI spend. Customers who run far above the baseline get billed for the difference, metered against provider costs with a published markup. The metered line lived on invoices but not on the pricing page.

It is on the page now. The markup is 15%, which is the number we use internally, which is the number on the page, which is the number on the invoice. If we raise it, we will announce it separately rather than quietly updating the page.

The line we removed

“Starting at” was on the page for the entry tier. It’s not anymore.

“Starting at” is a phrase that lets a pricing page do two incompatible things at once: look concrete (there is a number) and stay vague (the number might be a floor). In our case the “starting at” was technically accurate — the base plan does start at the published price, and some add-ons cost more — but it was doing the wrong work. New customers were asking on every call “what’s the actual number” because the page hadn’t committed to one.

The new page has committed numbers for each tier and an add-on list with prices. The add-ons are named: SSO, evidence vault, private retrieval index, advanced audit logs. Each has a monthly price. No “contact us” on any add-on priced below $2,000 a month. The two enterprise add-ons that genuinely require a conversation — dedicated instance, on-prem deployment — say “contact us” and name specifically what the conversation is about.

Why we care about this

Pricing opacity is the single most reliable tell that a software company is behaving extractively. A company that will not put a number on its product page is a company that prices each customer based on what they will accept, which means the company that asks politely and the company that haggles aggressively pay very different amounts for the same software. That is not a pricing strategy; it is a trust leak.

I wrote in Q3 that the pricing-opacity pattern is a signal category operators should use to read the market. The inverse applies to us: the pricing page is where we are most visibly exposed to the test. If we write category commentary about opacity and then our own page is vague, the commentary is hot air.

What is still imperfect

Two things, named.

The “document” unit is not fully intuitive. We count KB blocks, which are post-chunking artifacts. A customer who uploads 50 PDFs of varying lengths cannot eyeball in advance how many blocks that produces. We publish a rough conversion — the median 30-page PDF produces ~400 blocks — but the actual number depends on the document’s structure. We are working on a pre-ingest estimator so the number is visible before upload, not after. Not shipped.

Enterprise pricing still lives off the page. Customers with compliance, isolation, or deployment requirements that push them past the published top tier end up in a conversation with named pricing ranges but not a fixed number on a page. I would like to publish a range. We haven’t yet because the range is wide enough (10x between floor and ceiling) that publishing it felt misleading. This is an honest limit, not a plan; I don’t know when we’ll solve it.

What pricing transparency is not

Transparency is not the same as cheapness. A cheap product that is vague about its price is still extractive. An expensive product that is precise about its price is not. We are not the cheap option in our category and we do not intend to be. We publish prices because customers deserve to know what they cost, not because we intend to race them to zero.

Transparency is also not the same as “take it or leave it.” Customers with genuine multi-year commitments or genuine volume get negotiated prices. The negotiation starts from the published number and moves against specific commitments. It does not start from a vague “contact sales” that turns into a custom quote based on how well-resourced the buyer looks.

The rhythm

We update the pricing page quarterly at most. Changes are logged on the page itself — a dated changelog footnote — so a customer comparing the page today against the page they signed from six months ago can see what moved. No quiet changes. No revised numbers without a line in the change log. No reserved-right-to-update language.

The next update, if it happens, lands at the end of Q1 2026. If nothing moves, the change log says “no changes this quarter” and the page looks identical. That is also a signal worth sending.