Trust & security

Built so you can trust the screen.

Decifer turns your data into decisions, so the bar for trust is high. Two things have to hold: the numbers must be real, and your data must stay yours. Here is exactly how we make sure of both, and where we draw the line at what we’ll claim.

Every number is computed, never invented.

The separation between maths and language is deliberate, and it is the core of how Decifer stays honest.

Computed, never invented

Code does the maths

The engine computes every figure in code. The AI layer only explains what the numbers mean, it is handed the figures and is not allowed to introduce one of its own. A guard checks the narration against the computed values before you ever see it.

“No clean signal” beats a guess

Silence over fiction

When the data doesn't support a conclusion, Decifer says so plainly instead of manufacturing one. An empty answer you can trust is worth more than a confident answer you can't.

Every figure traces to source

Open the receipts

Each decision carries the metric, the value, the period it covers and the connector it came from. Nothing on the screen is something you have to take on faith, and sample data is always labelled as such.

Your data stays yours.

Connecting your sources should never mean handing over control of them.

Read-only connections

We read, never write

Decifer reads from the tools you already use. It never writes back to them and never changes your data at the source.

Isolated per company

Walled off by design

Each tenant's data is separated by row-level security in the database. One client can never see another's, the isolation is enforced at the data layer, not just the UI.

Credentials in a vault

Not in code or env

The keys to your connected sources live in an encrypted per-tenant vault, not in source code or environment files, and are used only to read your data.

Yours, and only yours

Never shared, never training

Your data is used solely to generate your decisions. It is not shared with other tenants and is not used to train shared models.

Access is controlled and logged.

The same discipline that keeps the numbers honest governs who can reach them.

Tenant resolved server-side

No tampered URLs

Which workspace you can see is determined on the server from your session, never from a value in the browser or the URL. There is no parameter to change to reach someone else's data.

Roles scope what you can do

Least privilege

Agency, owner, admin, member and viewer roles control who can see and change what, so access matches responsibility.

Actions are logged

An audit trail

Sensitive actions are recorded in an audit log, so there is an account of what happened and who did it.

Honest about what we can't prove.

Plenty of tools will tell you what you want to hear. Decifer is built to tell you what the data actually supports, including when that is less than you hoped, or when the honest next step is to run an experiment rather than trust a model.

That candour is the point. A decision is only worth making if the evidence under it is real, so we would rather show you a smaller, true picture than a larger, flattering one.

What we won't do

  • Promise an outcome we can't measure, we recommend a test instead.
  • Hide the assumptions behind a figure like LTV.
  • Show a confident number where the honest answer is “no clean signal.”
  • Use a client's name on this site without their say-so.

Put it to the test on your own data.

Request access, connect your sources read-only, and see decisions you can trace all the way back to where the numbers came from.