Resources
Case studies
Operational stories: the existing AI environment, the control gap, what OpsAI was asked to do about it, and what changed.
No customers to name
And no anonymised ones either. A “global financial services firm” costs nothing to write and cannot be checked, which is precisely why it is worth nothing.
What exists instead is a fully documented fictional estate where every figure is computed and checkable.
Stated plainly
There are no case studies, and there are no anonymised ones either.
An unnamed customer in a named industry with an impressive percentage is the easiest thing on a B2B site to write and the least useful thing to read, because none of it can be verified. Rather than publish that, this page shows the estate the product was actually designed against.
The sample estate is fictional and every page that uses it says so. What makes it more useful than an anonymised story is that it is complete and checkable: 11 AI systems across 8 frameworks and 6 teams, 8 policies with named owners, 11 connected systems, and a stream of 240 decisions. Every number on this site is computed from it rather than written into a page.
It also contains things a case study would quietly leave out. A model is blocked. Two credentials are overdue for rotation. 4 discovered systems can change state and 3 cannot. One system in the register is deliberately not connected. A story built to persuade would have none of those.
AI systems
11
across 8 frameworks
Policies
8
owned by 6 teams
Decisions recorded
240
20 held, 5 refused
Real customers named
0
there are none
IllustrativeA fictional company. Every figure above is computed from the fixture rather than asserted — the register itself.
What a case study would describe
The four parts, and which of them a fixture can honestly show.
A real case study has an existing environment, a control gap, a brief, and an outcome. The first two are demonstrable from a fixture. The second two require a customer, and this page will not pretend otherwise.
- The existing environmentshown
- 11 AI systems built independently by 6 teams in 8 frameworks, reaching 11 systems of record. Nobody coordinated the framework choices, which is the realistic part.
- The control gapshown
- 7 systems found that nobody registered, 4 of which can change state. Two credentials overdue. A model still being called while it retires. Scored across seven categories.
- What OpsAI was asked to doneeds a customer
- A brief is a decision somebody made under real constraints, and inventing one would be inventing the constraints. A fixture can show what the product does, not what an organization asked of it.
- What changedneeds a customer
- The one part that cannot be simulated at all. An outcome requires a before and an after in the same real organization, and any number offered here would be fabricated.
The systems in the sample estate
Two-letter marks rather than logos, and the unconnected one stays in the grid. §54: a logo wall implies a partnership, and naming these as integration targets is the only claim being made.
RazorpayRazorpayPayments
RazorpayXRazorpayXPayouts
Zoho BooksZoho BooksAccounting
NetSuiteNetSuiteERP
SalesforceSalesforceCRM
PostgresPostgresDatabase
ZendeskZendeskSupport
GmailGmailEmail
SlackSlackMessaging
Amazon S3Amazon S3Storage
BoxBoxDocuments- WorkdayWorkdayNot connected
IllustrativeIntegration targets in a fictional estate, not customers or partners.
When there are real ones
What a published case study will and will not contain.
Worth stating now, because the constraints are the same ones that produced this page. A case study written under them will be less impressive and more useful than the genre usually manages.
Case studies
There is no deployment to write about yet. When there is, publishing it needs the customer's written agreement — which is also why no estate on this site is described as a reference.
What it will contain
- A named organization, or nothing. An anonymised industry descriptor is not a middle ground, it is a way of making an unverifiable claim look sourced.
- The environment before, including the parts that were nobody’s fault and everybody’s problem.
- What was refused as well as what was authorized, because the refusals are where a control shows its shape.
- Figures the customer agreed to publish, with the measurement stated. No percentage without the period behind it.
The environment
The register
Every AI system in the sample estate, with its owner, framework, autonomy rung and what it may attempt.
The gap, scored
Posture
Seven categories, each a fraction over the fixture data, with the movement since the last assessment.
The gap, found
Discovery
What the estate did not know it was running, and which of those can change state.
- No logo wall either
- A grid of customer logos reads as endorsement no matter what the caption says, and reproducing a trademark needs permission. There is none on this site.
- Why the sample estate has flaws
- A blocked model, overdue credentials, an unconnected system and unattributed findings. An estate where everything is correct demonstrates nothing about a product whose job is to surface what is not.
Where to go
The sample estate is more checkable than a case study would have been.
Every figure on this site traces back to it, and any of them can be verified against the page that derives it. That is a weaker claim than a customer outcome and a stronger form of evidence.