The evidence.
Everything on this site comes from pages any visitor can open, read on November 22, 2026. I created no account, submitted no form and tested no system.
Where OpenAI already leads
- A rollout guide that starts with owners.Step 1 of the admin rollout guide assigns an owner to each part of the rollout before anything is configured. A separate quickstart gives a nine-item checklist. Rollout guide, Quickstart
- A written method for measuring value, with rules for reading the numbers.Agree the result with the business owner and record the baseline. A zero may be a reporting gap, a sample is not a total, and estimates stay labeled as estimates. Usage Insights
- A published delivery sequence.For managed deployments: outcomes and success criteria, integration, security review, acceptance testing, a staged rollout, then review. Help Center
- Forward-deployed engineering, explained in public.The head of the team has written how it works: start with the outcome, then make success repeatable. The posting’s last duty asks for the same thing. October 22, 2026
- Six months’ notice before a generally available model retires.Longer than the 60 days Anthropic and Azure state, with the shutdown date readable from the models endpoint. Deprecations
- A partner network with a public list.73 partners named on the page, three tiers, and a pilot that aligns partner staff with OpenAI’s own forward-deployed teams. Partner Network
- Uptime in public.The status page shows uptime per component: 99.97% for the APIs from August to November 2026. Status
OpenAI beside six other providers
What a customer’s delivery lead can establish from each provider’s own public pages. Each name links to the page that provider’s row starts from.
| Provider | Rollout guide | Baseline step | Adoption analytics | Delivery people | Partner list | Retirement notice | Reserved capacity | Public uptime figure |
|---|---|---|---|---|---|---|---|---|
| OpenAI | Eight-step guide, and a nine-item quickstart | In the value guide, not in the rollout guides | Usage, task insights, outcomes, surveys, an API | Forward Deployed Engineers; AI advisors | Public, 73 named | 6 months for generally available models | Reserved Tier; Scale Tier | 99.9% on Scale Tier |
| Anthropic | Four-phase admin guide with a checklist | Not found | Dashboard and two analytics APIs | Applied AI engineers | Public directory | At least 60 days | Through sales; the older tier is closed to new buyers | Not found |
| Three-phase guide with named roles, for Workspace | Yes, a baseline survey in phase 1 | Gemini reports; a value tab | Cloud Consulting; engineers placed with integrators | Partner network; list not readable | 12 months of availability, or at least 45 days | Provisioned Throughput | 99.5% for the Gemini API | |
| Microsoft | Five-step admin guide; a success kit | Not found | Copilot Dashboard; usage report | Forward-deployed engineers; FastTrack | Public directories | 60 days; 30 for preview | Provisioned throughput units | 99.9% on the product page |
| Amazon Bedrock | A maturity model; no rollout checklist found | Not found | Usage telemetry; no per-user view found | Forward Deployed Engineering | Partner network; list not readable | 6 months, or 45 days | Provisioned Throughput | 99.9% |
| Cohere | Not found | Not found | Not found | Forward-deployed engineers; architects | Partner page; no list seen | Policy with no period stated | Not found | Not found |
| Mistral | Not found | Not found | Usage dashboard | A named deployment team | Partner page; no list seen | 6 months; 1 month for others | Dedicated capacity mentioned | Not found |
“Not found” means not found on public pages, not that the provider lacks it. An AI research agent gathered the rival rows from each provider’s own pages. I checked the OpenAI row against saved page text. Eight rival facts were spot-checked. The rest were not checked line by line, and Cohere and Mistral got a short look only.
What a delivery lead should know about this market
Four things that shape a plan, whichever provider it is built on.
- Every large provider now names forward-deployed engineers.OpenAI, Google, Microsoft and Amazon all describe them on public pages, and Anthropic funds training for them. The engineers are no longer the difference. How their work is planned and joined to the customer’s is.
- The model can retire during the program.For generally available models, stated notice runs from 45 days to six months across the providers I read. A launch date and a retirement date belong on the same page.
- Measurement has moved from seats to outcomes.OpenAI, Google and Microsoft all publish ways to connect usage to results. Of the rollout guides I read, only Google’s makes a baseline survey a named step.
- Capacity is bought ahead.OpenAI, Google, Microsoft and Amazon all document reserved capacity per model. It has to be sized before launch, from a test and not a guess.
What is not on this page
While reading, I found seven places where OpenAI’s public pages disagree with each other, and a few smaller inconsistencies. Most touch things a customer plans around: capacity, retirement dates and reporting.
Those are in a findings report I will send to OpenAI directly. A public page is the wrong place for them.