# Template: record of processing row and impact assessment sheet

Two working instruments. Copy both.

## Sheet 1: the register

One row per processing activity. Populate from discovery and mapping, keep alive at intake.

| Activity | Owner | Purpose | Legal basis | Data categories | Whose data | Kept until | System of record | Who receives it | Reviewed on |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Example: credit scoring | Head of lending | Assess loan applications | Consent | Identity, income, repayment history | Customers | Relationship + 3 years | Lending app | Credit bureau | |
| Example: payroll | Head of HR | Pay salaries | Employment purposes | Identity, bank account, salary | Employees | 8 years after exit | HRMS | Payroll processor, tax authorities | |
|  |  |  |  |  |  |  |  |  |  |

Rules that keep the register true:

- Every row has a named owner. A register with no owners is a snapshot.
- Every new project, product, vendor or report adds or updates a row at intake.
- No new processing goes live without a row. That is the gate.
- Owners confirm their rows on a cycle; the confirmation is recorded.

## Sheet 2: the impact assessment

One sheet per triggered activity. Triggers: high volume, sensitive data, new technology, profiling, children's data, a material change, or after an incident.

| Step | Your entry |
| --- | --- |
| 1. The processing described | Example: app-based credit scoring using repayment history and device data, all loan applicants, collected directly and from the bureau, to automate approve or refer |
| 2. Necessity and proportionality | Example: scoring needed for turnaround; device data not needed for the decision, dropped |
| 3. Risks to people | Example: wrongful rejection at scale; bias in the model; exposure of financial data. Likelihood and severity rated per risk |
| 4. Mitigations | Example: human review of all rejections (owner: lending ops); model bias testing each release (owner: data science); field-level encryption (owner: security) |
| 5. Residual risk | Example: model drift between tests; accepted and monitored |
| 6. Decision | Example: go, with the dropped data field and human review as conditions. Approved by: name, role, with reasons recorded |

A risk finding with no decision attached fixes nothing. Step 6 is the point of the exercise.
