I’ve had the same conversation probably fifty times this year. A client calls, usually someone technical and says their CEO wants AI.
- Do they know what kind? No.
- Do they know which problems they’re trying to solve? Not really.
- Did someone at a conference show them something impressive and now there’s budget that needs to be spent by Q4? Almost always yes.
Here’s the thing—that awkward gap between “our board says we need AI” and actually shipping something useful is the most profitable consulting opportunity I’ve seen in a decade. And most tech companies are completely ignoring it.
The Gap in the Market#
AI presales isn’t sales. You aren’t pushing licenses. It isn’t delivery either—you aren’t writing code yet. It’s that weird, messy phase where you help companies figure out if they’re about to waste a million dollars.
They need someone who speaks both languages. Someone technical enough to know what AI Engineering can actually do (vs. what looks cool in a demo) and business-savvy enough to know when a spreadsheet is better than a neural network.
But here’s the secret: This isn’t about selling AI. It’s about trust. The trust you build by saying “No, that’s a terrible idea” is what makes them sign the check when you finally say “Yes.”
Structuring Your Services (So You Get Paid)#
You need three specific offerings. If you don’t productize this, you’ll end up doing free consulting over coffee forever.
1. Discovery (Fixed Fee)#
Two to four weeks. You talk to everyone. You look at their data. You map the mess. The deliverable isn’t a wishlist—it’s a ranked list of what is actually possible.
Case Study: The Law Firm We worked with a firm handling mortgage disputes. Their process was a disaster of emails and phone tag. We found five potential AI wins:
- Intake Agent: Chat with clients to get the facts.
- Client Portal: Show them the timeline so they stop calling.
- Report Translator: Turn “Legalese” into plain English.
- Auto-Notifications: Tied to court dates.
- Internal Q&A: A bot that knows their playbooks.
Discovery proved that three of these were impossible with their current data. We saved them six months of failure just by checking the data first. Price this as a fixed fee. It keeps you honest and focused.
2. The Blueprint (PoC Design)#
Once you know what to build, you need to design how. This is the technical spec. RAG vs. Fine-tuning? Vector DB vs. Keyword search?
For that law firm, we designed just two things: the Q&A bot and the Report Translator. We cut everything else. We handed them a plan that any competent team could build.
The Power Move: Make the design agnostic. Tell them, “You can take this blueprint to another dev shop if you want.” They almost never do, but offering it builds massive trust.
3. Education Workshops#
Most stakeholders think AI is either magic or a fraud. Run a workshop before you start. Teach them what a “Hallucination” is. Show them why data quality matters.
For the law firm, we ran “AI for Legal Ops 101” before we wrote a line of code. It reset their expectations from “Magic Button” to “Useful Tool.”
The Hard Part: Business Outcomes#
The hardest part isn’t the algorithms. It’s connecting the tech to the P&L.
Start every meeting with pain, not Python.
- What is too slow?
- Where are you losing money?
- Who is complaining the most?
I watched a manufacturing client burn 110k on a predictive maintenance system. It worked perfectly. Nobody used it. Why? Because the maintenance guys didn’t trust a black box telling them what to do. The problem wasn’t accuracy; it was culture. We should have caught that in week one.
Don’t sell “Machine Learning.” Sell “Stopping customers from leaving.”