The Unicorn Skill Set#
To pull this off, you need a weird combination of skills. This is why you can charge a premium. You aren’t just a dev and you aren’t just a suit.
- The Tech Chops (Non-Negotiable): You have to have built this stuff. Clients can smell a faker. You need to have shipped RAG systems that actually work. You need to know what happens when an LLM hallucinates in production. You need to have debugged a pipeline at 2 AM. If you haven’t built it, don’t sell advice on it.
- The Translator: Can you explain “Vector Embeddings” to a CFO without putting them to sleep? Can you explain why “Accuracy” isn’t the only metric that matters? This is a communication job.
- Context is King: You need to understand their world. In Law, accuracy is everything. In Marketing, creativity matters more. If you try to sell a creative, hallucinating bot to a Lawyer, you’re dead.
- Herding Cats: Discovery workshops are messy. You need to control the room, manage the egos and get to a decision.
How to Ruin Your Reputation#
1. The “AI for AI’s Sake” Trap A CEO read about ChatGPT on a plane. Now they have budget. Say No. Tell them: “I’d love to help, but first we need to find a problem worth solving.” If they insist on building “AI” without a problem, walk away. It will fail and they will blame you.
2. The Demo Trap RAG systems look perfect in demos. LLMs write great poetry. Production is a different beast. Data is missing. Users type weird things. If you promise the Demo version, you will fail in Production. Undersell. Overdeliver.
3. Scope Creep (The Project Killer) Everyone wants “just one more feature.” Remember the law firm? They wanted the Intake Agent added to the Q&A bot immediately. We fought them. We kept it narrow. Scope control is your main job. If they want more, write a new contract.
4. Ignoring Culture The best code in the world fails if the users hate it. Is there executive support? Is there a team to maintain this thing when you leave? If the organization isn’t ready, the tech doesn’t matter.
5. Confusing Experiments with Demos A Demo is for applause. An Experiment (PoC) is for learning. The goal of a PoC isn’t to look pretty; it’s to answer the question: “Should we build this?” Stay focused on the question, not the polish.
The End Game#
AI presales is a massive opportunity because everyone is confused.
If you start with small, honest discovery engagements, you prove your value fast. You stop competing on hourly rates for implementation work.
The companies that figure this out—the ones who say “No” to bad ideas and “Yes” to the right ones—stop being vendors. They become the first phone call the client makes when they have a new idea. That is a very good place to be.