A few weeks ago, an engineering director told me he’d put four AI vendors side by side in a spreadsheet. One column per vendor, a price per row, a tidy comparison he could hand to his CFO. He asked me for our number so he could fill in the last cell. I told him I couldn’t give him one, and I watched his face do the thing faces do when they decide you’re being evasive.
I get it. He wanted to do his job, and his job — as he understood it — was to find the cheapest competent option. The spreadsheet was a reasonable instinct. The problem is that the spreadsheet was already lying to him, and my number would have made the lie more convincing.
Here’s what was actually in those four cells. Vendor A had quoted a chatbot bolted onto a help-center export. Vendor B had quoted a document-extraction pipeline that would touch his ERP. Vendor C had quoted a proof of concept with no path to production. Vendor D had quoted something nobody could quite describe. Four numbers, four completely different pieces of work, lined up in a column as if they were the same thing measured four ways. They weren’t comparable. They just looked comparable.
That’s the whole reason we don’t publish pricing. Not because we’re hiding something, and not because we want to trap you on a sales call. It’s because a price on a page answers a question you didn’t ask, and quietly stops you from asking the ones that matter.
A sticker price answers the wrong question
When you see a number before you’ve scoped the work, your brain does something automatic: it anchors. Every later conversation gets measured against that first figure. If the number was low, anything more feels like a markup. If it was high, you walk in defensive. Either way, you’re now negotiating against an anchor that was set before anyone understood what you actually need.
The deeper issue is that AI work doesn’t have a stable unit to price. When you buy a laptop, the thing in the box is the thing in the box. When you buy “an AI solution,” the words cover a range so wide that a single price is almost meaningless. Consider two requests that sound identical on the surface:
- “We want to summarize incoming documents.” If those documents are clean PDFs in one format and the summary lands in a Slack channel, that’s a contained, well-understood build.
- “We want to summarize incoming documents.” If those documents are scanned drawings, vendor submittals, and email chains, and the output has to write back into your project-management system with an audit trail, that’s a different animal entirely — different by a large multiple of effort and risk.
Same sentence. The cost driver isn’t the AI. It’s the messiness of your inputs, the systems you need to touch, and how wrong the system is allowed to be. None of that fits on a pricing page, which is exactly why a pricing page can’t tell you the truth.
What actually drives the cost
If you strip away the marketing, the real cost of an AI project lives in three places, and almost none of it is the model.
Integration surface
The model is the cheap part now. The expensive part is everything around it: where the data lives, what shape it’s in, which of your systems the solution has to read from and write back to, and who owns the credentials and the failure modes when something upstream changes. A request that stays inside one clean data source is a different project from one that has to reach into three systems your team half-remembers configuring.
Acceptance bar
“Good enough” for an internal drafting assistant and “good enough” for something that touches a client deliverable or a regulated record are worlds apart. The closer the output sits to a decision with real consequences, the more review, evaluation, and guardrail work the project carries. That’s not padding. That’s the difference between a demo and something you can actually depend on.
Unknowns
Every project has a few questions nobody can answer until you’re inside the data. How consistent are the inputs, really? How often does the edge case show up? A good partner prices the known work and is honest that some of it can’t be fully known on day one — and tells you how that gets handled rather than burying it.
You cannot read any of that off a rate card. You can only find it by talking through the specific work in front of you. That’s the conversation a published number prevents, because once there’s a figure on the table, everyone argues about the figure instead of the work.
The questions that actually predict cost and value
You don’t need our pricing to protect yourself in an AI buying process. You need a short list of questions that expose whether a partner actually understands what they’re selling. Ask these of anyone — us included — and the vague vendors separate from the serious ones within about ten minutes.
”What exactly is in scope, and what’s explicitly out?”
A serious partner can draw a hard line around the first piece of work. Vague answers — “it depends on your needs,” “we’re flexible” — usually mean they haven’t thought about your problem yet and are pricing a feeling. You want a crisp boundary you can point at later.
”How will we both know it’s working?”
This is the question that saves projects. Ask for the acceptance criteria before any work starts. What does “done” look like? What accuracy or behavior counts as success, and how is it measured? If a partner can’t tell you how you’ll jointly agree the thing works, you’re buying a maybe.
”Who actually does the work?”
The person in the room is often not the person who builds. Ask who writes the code, who owns the architecture, and whether the senior people you’re talking to stay involved or hand off to whoever’s free. There’s nothing wrong with a team — but you should know whose hands are on your system.
”Who owns what when it’s finished?”
Get clear on ownership early: the code, the models, the prompts, the data, and the infrastructure it runs on. Can you take it in-house later? Are you renting access to something you can’t see inside, or do you own an asset? Both can be fine. Not knowing which one you bought is not fine.
”What happens when the scope changes?”
It will change. Inputs turn out messier than expected, or a stakeholder wants one more system connected. The question isn’t whether change happens — it’s whether your partner has a sane, transparent way to handle it, or whether every adjustment becomes a renegotiation. Ask them to walk you through how the last project changed mid-flight and what they did about it.
If a partner answers these clearly, the price almost takes care of itself, because now you’re both pricing the same well-defined thing. If they can’t, no published rate would have saved you — it just would have hidden the problem until the invoice arrived.
Why a call beats a quote
None of this is an argument for mystery. It’s an argument for sequence. The number has to come after the understanding, not before it, or the number is fiction.
A focused 30-minute call does something a pricing page structurally cannot: it lets us look at your actual inputs, your actual systems, and your actual acceptance bar, and tell you what the work really involves — including, sometimes, that you don’t need us yet. We’ve told a fair number of teams that the right first move is a smaller, cheaper experiment than what they walked in asking for. You can’t say that to someone who already paid against an anchor on a page.
That’s also how we set scope and capacity at Ryshe Forge and Ryshe Labs — not from a menu, but from a real read on the work. The conversation isn’t a sales gauntlet. It’s the only honest way to answer the question you came with. Sometimes the most useful thing we say on that call is “this is simpler than you think,” and sometimes it’s “this is harder than the other quotes admitted.” Both are worth more than a number that lets you skip the thinking.
The takeaway
A price on a page feels like information. Most of the time it’s the absence of information dressed up as a fact. It compares projects that aren’t comparable, anchors you before you understand your own problem, and rewards the vendor who scoped the least.
So use the spreadsheet — just fill the columns with better questions. Scope. Acceptance criteria. Who does the work. Who owns the result. How change gets handled. Those five answers will tell you more about your real cost and your real risk than any sticker ever could. Ask them of every partner you’re considering, and ask them of us.
If you’d like to run your specific situation through those questions with someone who’ll tell you what you need to hear, book a 30-minute call. Bring the messy version of your problem, not the cleaned-up one. That’s the version we can actually price — and the version worth solving.
