The Problem
List pricing isn't actual pricing. Anthropic, OpenAI, and Google all negotiate with operators who have three things: real volume, technical portability, and a credible threat to walk. Their account executives have discount authority. Their solution engineers can write a custom SLA. Their legal teams will redline a data-retention clause if you push.
Most operators never find out. They pay list rates. They signed an annual commit on a handshake call because their AE was friendly and the discount looked nice. They didn't read the SLA past the headline uptime number, and they have no idea what would actually happen if their primary provider had a 30-hour outage on a critical Tuesday.
The leverage is real. The leverage is in your usage data — the actual minute-by-minute shape of your inference spend, broken down by capability tier. The leverage is in your portable prompts (per Module 2.2) — the kind of codebase where switching the underlying provider is a config change, not a quarter-long migration. And the leverage is in a credible threat to multi-vendor — one you can demonstrate with a printed benchmark, not just assert in a meeting.
Most operators have none of these. They pay list. That's the gap this module closes.
The Core Insight
Lock-in compounds silently. Negotiation requires real portability — if you can't actually switch in two weeks, your “leverage” is theater. The technical prerequisite is portable prompts. The policy work is what this module covers.
You can't negotiate from a single-vendor codebase. The vendor's AE knows it. Their pricing model assumes it. The discount they offer reflects the cost of retention, which is approximately zero when you have nowhere to go.
The asymmetry inverts the moment you can demonstrate — with numbers, not posturing — that your prompt library runs cleanly on the next provider down the list and produces commercially acceptable output. The conversation shifts from “here's our published rate” to “what do we need to do to keep your business.” For real-volume accounts, that shift can move the line item materially and can open a more serious SLA conversation.
Module 2.2 (Multi-Vendor Reality / Prompts as Code) is the technical prerequisite to this module's policy work. Without portability, you negotiate from weakness — and the savings here don't justify the engineering effort to make portability real. With portability, the math flips: the same engineering investment now compounds into recurring annual savings on every renewal.