Everyone says they want AI integration. Most of them want two tools by Friday.

I spend a lot of time pitching. Some of it is responding to RFPs. Some of it is sitting across from a leadership team that has decided, usually after a board meeting or a competitor's press release, that it's time to "do something with AI."

The conversations start the same way. They want full integration. AI woven into how the business runs. Not a chatbot bolted onto the website. The real thing.

Then I send the proposal, and the tone changes.

The prompt misconception

Most firms I talk to believe, somewhere underneath, that AI is a matter of writing the right prompts. They've used ChatGPT. They've watched it draft an email in four seconds. So the mental model is: a consultant shows up, writes some very good prompts, connects a few things, and the firm is transformed.

If that were true, the price would be absurd. It isn't true.

A prompt is an instruction. An instruction is only as good as what the model knows when it reads it. And what a general model knows about your firm is nothing. It doesn't know how your senior estimator decides a job is worth bidding. It doesn't know which clients your principal will bend the rules for and why. It doesn't know that the template everyone uses has a quirk that only two people understand.

That knowledge is the whole game. And it doesn't live in a prompt.

What integration actually means

In a recent pitch I offered a firm full integration, and I explained where it has to start. Not with tools. With the gap between what the senior people in the company know and what the people under them need to know.

Every firm has that gap. It's the reason the senior people are the bottleneck. Work flows up to them for review, for judgment calls, for "how do we usually handle this," because the answers live in their heads and nowhere else.

Closing that gap is real work, and it runs down a lot of paths at once:

• -Recording the meetings where decisions actually get made, so the reasoning isn't lost the moment the call ends.
• -Understanding how senior people decide. Not the policy manual version. The real version: what they look at first, what makes them say no, what they let slide.
• -Understanding how they communicate. How they explain a decision to a client, push back on a vendor, coach a junior. Voice and judgment travel together.
• -Bringing every document into one place. Proposals, SOPs, past bids, contracts, email threads, the shared drive nobody has cleaned since 2019.
• -Making all of it recallable. Not stored. Recallable. Anyone on the team asks a question and gets the answer the senior person would have given, with the source.

Once that foundation exists, you can build almost anything on top of it. Proposal drafting that sounds like your best writer. Review that catches what your most experienced person would catch. Onboarding that doesn't consume a senior person's month. The tools become easy. The foundation is the hard part, and it's the part that makes every tool after it actually work.

What firms actually want

Here's the trend I didn't expect. When I lay that out, most firms agree with every word. Yes, that's exactly the problem. Yes, our senior people are drowning. Yes, we want that.

And then what they actually buy, or want to buy, is a couple of tools they can use today.

A proposal generator. A meeting summarizer. Something that drafts the weekly report. Something visible by next quarter.

I don't think that's hypocrisy. I think it's two very reasonable instincts colliding. Leadership wants the transformation, and leadership also needs something to show for the spend before the next budget cycle. A foundation is invisible for a while. A tool is a demo.

Both are fine. Confusing them isn't.

Buying a couple of tools is a legitimate choice. If your firm has one painful, well-defined task, a focused tool can pay for itself fast, and I'll happily build it.

The problem is buying tools while believing you bought integration.

A tool built without the foundation only knows what you paste into it. It works on day one and plateaus by month three, because it never learns how your firm thinks. Then the next tool gets built the same way, with its own copy of a little context, and a year later you have five disconnected tools, none of which know what the others know, and the senior people are still the bottleneck.

That's how firms end up concluding "we tried AI and it didn't really change anything." They didn't try integration. They tried tools and called it integration.

How I'd approach it

If you're weighing this for your firm, be honest about which one you're buying.

If you want tools: pick the single most painful task, scope it tightly, and expect it to do that one thing well. Don't expect it to get smarter about your business on its own.

If you want integration: expect the first phase to be about capturing knowledge, not shipping features. Budget for it like the infrastructure it is. Measure it by one question: can a junior person get a senior-quality answer without interrupting a senior person?

If you want both: start with one tool, but build it on the foundation from day one. Every document it ingests and every decision it records becomes part of the shared layer the next tool uses. The first tool is how you pay for the foundation instead of skipping it.

That last path is the one I recommend most. It's also how our own products were built. EchoMe starts by learning how a person writes and speaks before it generates a single post. Closr builds each agent's own knowledge base before its daily brief tells them who to call. Neither one starts with prompts. Both start with context.

Why the price surprises people

So when a firm asks why real AI consultancy costs what it costs, this is the answer. You're not paying for prompts. Prompts are cheap and anyone can write them. You're paying for someone to sit inside your firm long enough to capture what your best people know, organize it, and make it usable by everyone else.

That's not a prompt. That's the most valuable asset your firm has, finally written down.