Three hundred thousand employees does not mean three hundred thousand addressable seats. Separating the two avoided one point two million dollars.
How to Negotiate with OpenAI and Anthropic: The Vendors With Nobody to Call
Fewer than 50 sales reps globally per vendor, focused on $100M+ deals. Below $10M a negotiation rarely starts, discounts run 5 to 25 percent on commitment size, and the only leverage is credible competition between OpenAI, Anthropic, and Gemini with a benchmarked case.
Lowe's is a leading US home improvement retailer running more than two thousand stores with roughly three hundred thousand employees. Its OpenAI estate covers ChatGPT Enterprise seats and OpenAI API consumption.
Those two things get bought together and behave completely differently. Seats are a per person subscription with an annual commitment. API consumption is metered per token and scales with what your systems do, not with how many people you employ.
The engagement avoided 1.2 million dollars across a three year term, almost entirely by getting the seat population right before committing to it.
Three hundred thousand employees, more than two thousand stores, and a technology estate spanning corporate functions, supply chain and store operations.
That shape matters enormously for AI licensing. The overwhelming majority of that workforce is in stores, doing work that does not involve drafting documents, managing an inbox or sitting in meetings that need summarising.
An assistant that works on documents, mail and meetings has a natural audience, and in a retailer that audience is the corporate and support population rather than the headcount.
OpenAI's opening proposal sized the commitment against a broad share of the workforce, with seats and API consumption folded into a single number.
That is a reasonable opening position for a vendor and a poor basis for a three year commitment. It assumes an adoption curve nobody had measured, across a population nobody had segmented.
The blending was the more expensive problem. A combined commitment removes your ability to expand one side while holding the other, which is exactly the flexibility you want when adoption is uncertain.
| Line | How it is metered | What drives it | How it should be sized |
|---|---|---|---|
| ChatGPT Enterprise seats | Per user, annual commitment | The addressable population | Measured weekly active use in a pilot |
| OpenAI API | Per token, input and output priced separately | What your systems do | Modelled from actual workload volumes |
| Combined commitment | One number covering both | Vendor preference | It should not be. Separate them. |
We segmented the workforce before discussing any number. Three questions decided whether a role was addressable: does this person create documents in volume, do they carry an inbox that needs managing, and do they attend meetings that need summarising.
Applied across the estate, that produced an addressable population dramatically smaller than headcount, and one that could be defended line by line to the vendor.
We then separated API consumption entirely and modelled it from the workloads actually planned, rather than from any relationship to employee numbers.
Finally we staged the seat rollout, starting with the population where the case was strongest and instrumenting weekly active use so expansion could be argued from evidence.
These are the moves that produced the avoidance. The first three did most of the work.
The common advice is to roll an AI assistant out broadly and quickly, on the argument that adoption compounds and early movers capture the productivity gain first. We disagree, at least about how it should be bought. Broad rollout and broad commitment are different decisions, and vendors are happy to let them be confused. You can pilot widely, measure honestly and expand fast while still committing narrowly, because the commitment is a contractual term rather than a deployment strategy. Organizations that commit at the pace they hope to adopt end up renewing seats that were assigned and never used, and they have no evidence to argue with because they never instrumented the first year.
If you are sizing an enterprise AI commitment, do these before you commit.
The eleven moves, segmenting the addressable population, separating seats from API consumption, staging the rollout, and the buyer side position at every step of an AI renewal.
Used across more than five hundred enterprise clients. Independent. Buyer side. Built for CIOs running the next OpenAI Enterprise renewal cycle.
Source: Redress Compliance advisory engagement file.
OpenAI framed the OpenAI Enterprise commit as the immediate OpenAI uplift across the broader generative AI. Redress reframed the approach around Lowe's actual OpenAI Enterprise utilization. One point two million dollars in AI cost avoidance against the publisher's opening OpenAI Enterprise quote.
Independent. Buyer side. The advisory firm enterprise software vendors do not want you to hire.
OpenAI Enterprise signals, Anthropic Claude signals, AI commit signals, and the broader generative AI licensing leverage signals across the practice.
