Anonymization rubric
The trade is strict: drop the identifiers, keep the texture. Everything below serves that one exchange.
An anonymized story that a peer reads and thinks "that is exactly us" has done its job. One that reads "a mid-sized company improved efficiency" has protected the client by making the story worthless, which is the failure this rubric exists to prevent.
The attribute table
Describe the client by durable attributes. Durable means true for years, true for many companies, and not searchable back to one.
| Attribute | Ship it as | Never ship |
|---|---|---|
| Size | A band: 50 to 200 employees, or "roughly 40 people" | Exact headcount, or a band narrow enough to name one company |
| Revenue | An order of magnitude: mid eight figures | The actual figure, or ARR to two decimals |
| Vertical | The specialty: specialty insurance brokerage, industrial parts distribution | A niche so narrow only three firms exist in it |
| Region | Broad: the US Midwest, DACH, the UK | City, or a state with one obvious candidate |
| Structure | Ownership shape: PE-backed, founder-owned, second-generation family | The sponsor's name, the fund, the family |
| Tenure | Rounded: founded in the 1990s | The founding year |
| Situation | The pressure: post-acquisition integration, first outbound hire, a channel that stopped working | The named acquisition, the named channel partner |
| Team shape | Roles and counts: two AEs and a founder selling | Named people, or titles unique to that org |
Combine three or four of these. Two is usually too vague to produce recognition. Six starts narrowing toward one company.
Rounding rules
Rounding is a disguise technique and a fabrication risk at the same time. The line between them:
- Round the identifiers. Headcount, revenue, founding year, contract value. Round outward to a band and say it is a band.
- Never round the results. A 1.7x stays 1.7x. Rounding a result up is fabrication regardless of intent, and it is the specific failure that poisons the whole set.
- Direction of rounding matters. Round result-adjacent figures against your own interest. If the pipeline grew from 11 to 29 deals, "from about 10 to under 30" is honest. "From 10 to 30" quietly inflates both ends.
- Timeframes round to the nearest natural unit. "Inside a quarter" rather than "in 81 days," which is precise enough to identify an engagement.
- Say when a number is a band. "Roughly" and "under" cost nothing and protect against the reader assuming false precision.
What always gets stripped
- Names of people, companies, products, and tools specific to the client.
- Logos, screenshots with visible branding, and document headers.
- Anything from the engagement record covered by an NDA, regardless of how anonymized it looks.
- Dates precise enough to pair with a public event: a funding round, a leadership change, an acquisition.
- Direct quotes with an identifiable speech pattern, unless approved.
The re-identification check
Run on every anonymized story before it ships. Non-negotiable, and it takes ten minutes.
The search test. Take the attributes as written and search them the way a competitor would. If the client appears in the first page of results, the combination is too narrow. Broaden one attribute and repeat.
The insider test. Ask someone who knows the client but did not work on the engagement to read the story cold. If they name the client, so will anyone in that industry.
The stack test. Check whether the combination of attributes describes fewer than roughly twenty plausible companies. Under twenty, the story identifies a short list, and a determined guesser will work through it.
The pairing test. Read the story against your other published stories, your client list, and your public case studies. Anonymity fails through cross-referencing more often than through any single detail. A story that is safe alone can be identifying next to the one published last quarter.
The client read. Send the anonymized draft to the client and ask directly: are you comfortable that nobody else can prove this is you? Their answer is the only one that counts. It also frequently returns permission to include more than you asked for.
Failing any test means broadening an attribute, never deleting the texture. The texture is the asset.
When the client says no to everything
Some clients will not approve any version. Two options remain, in order:
- Ship the mechanism without the client. Describe the problem shape and what was done about it as a method piece, with no engagement attached and no numbers. Weaker than a case study, more useful than nothing.
- Bank it. Approvals loosen when a relationship matures or a champion changes roles. Keep the intake notes and revisit annually.
Never publish against an unclear approval. An anonymized story the client did not sign off on is still a story the client can recognize.