100% independent. Rythium takes no fees, commissions or partnerships from any software or AI vendor. Ever.
Home/Case studies/OpenAI commit

OpenAI contract case study: How a fintech signed 31% below the first proposal

In this OpenAI contract case study, a fintech was about to sign a three-year token commit sized on a hackathon's peak usage. We modeled six months of realistic production traffic and negotiated a ramped commit with rate locks. The signed commit came in 31% below the initial proposal.

Client
A fintech
Vendor
OpenAI
Outcome
31% below the proposal

The situation

The fintech was about to sign a three-year OpenAI commitment sized on a few days of hackathon traffic

  • A fintech was about to sign a fixed annual token commit.
  • It was sized on a hackathon's peak usage.
  • That made it a three-year commitment scaled to a few days of unrepresentative load.

How we did it

We modeled six months of real production traffic and negotiated a commitment that grows with usage

  1. 01

    Model the traffic

    We modeled six months of realistic production traffic.

  2. 02

    Find the steady state

    We established the true steady-state consumption.

  3. 03

    Ramp the commit

    We negotiated a ramped commit with rate locks, so pricing was protected as usage grew, rather than paying up front for capacity that wouldn’t be used for a year.

A commit sized on a pilot's busiest days overpays for years. Model real production traffic before you sign.

The results

The fintech signed 31% below the first proposal, with prices locked as usage grows

Proposed: a fixed commit sized on the hackathon peak

Signed: a ramped commit with rate locks, below usage at every step

Six months of production traffic, modeled

Shape only: the fintech’s volumes are not shown

31%Below the initial proposal

3 yr
Rate protection secured
6 mo
Production traffic modeled

What we delivered

Structure
A ramped commit with rate locks
Pricing
Rate protection across the ramp
Signed commit
31% below the initial proposal

Signing an AI commit soon?

Send us the proposal. A senior advisor will tell you within 24 hours whether there’s money on the table, and how our AI commit sizing works.

Talk to an advisorAI procurement

Who leads our engagements

Independent licensing advisors since 2015: 260+ engagements, led by a former Oracle Vice President

Sheshagiri Anegondi
Expert-ledSheshagiri Anegondi

Managing Partner of Rythium Technologies. A former Oracle Vice-President (1998 to 2015) who now sits on the buyer’s side, and author of “Licensing Oracle”.

Anil Nahata
Expert-ledAnil Nahata

Co-founder of Rythium Technologies and its designated partner since 2015.

Ritesh Verma
Project delivery managerRitesh Verma

Vice President, Delivery and Operations at Rythium Technologies. Former CIO of Indian Oiltanking, with a 25-year career implementing Oracle applications.

Questions about this case study

How do you size an OpenAI commit?

On realistic production traffic rather than a pilot’s peak. Here we modeled six months of production traffic, established the true steady-state consumption, and negotiated a ramped commit with rate locks.

What was wrong with the original commit?

It was a fixed annual token commit sized on a hackathon’s peak usage: a three-year commitment scaled to a few days of unrepresentative load.

What did the fintech sign?

A ramped commit 31% below the initial proposal, with rate protection across the ramp.

Related

More case studies

  1. 01Managed SAMA fast-growing bank in India16,300+ assets under managed SAM, audit-ready all year
  2. 02Managed SAMA private-sector bank in India14,000+ assets, on-premises and on AWS
  3. 03Red Hat auditA savings and credit cooperative in KenyaA Red Hat audit closed with no extra penalties

Every case study, by publisher and issue: all case studies

Your renewal quote deserves a redline of its own.

Send us the quote, the demand letter or the proposal. A senior advisor will tell you within 24 hours whether there's money on the table.

Request a confidential review
WhatsAppChat with an advisor

Cookie settings

Strictly necessaryKeep the site working and remember your cookie choice.
Always on