Skip to content
ASIRI

NDPA modules · Impact assessments

Risk drops on paper the moment somebody promises a fix.

Eight assessments, plotted where they started and where they claim to be now. Asiri only lets a risk move once the fix that moves it is in place — so the credit scoring model still sits at critical, not at high.

Critical · consult the NDPC
  • Credit scoring model v2
High · not launchable yet
  • Facial recognition at sign-up
Medium · fixes owed
  • Call centre recording
  • Loan collections outreach
Low · signed off
  • Agent network expansion
  • Merchant analytics dashboard
  • Customer support chatbot
None · nothing to weigh yet
  • Staff device monitoring
Residual risk, as countedRed · fixes only planned
Assess somethingWhy you hold data

Before and after, per assessment

Two are held in place because the fixes are only promises

Five of the eight show the same rating before and after, and they are not the same story. Two are low or dormant with nothing left to reduce. One was sent back unfinished. The two in red are the ones that matter: the fixes that would move them exist only on paper.

Facial recognition stays at high because biometric data cannot be made less sensitive, only less widely held — and the credit model stays critical because nobody has been assigned to review a decline.

  • Credit scoring model v2

    DPIA-2026-0004 · launch planned 15 Aug · Ngozi E.

    criticalcritical

  • Facial recognition at sign-up

    DPIA-2026-0002 · submitted 3 weeks ago · Tunde A.

    highhigh

  • Call centre recording

    DPIA-2025-0018 · sent back 2 months ago · Ngozi E.

    mediummedium

  • Loan collections outreach

    DPIA-2025-0016 · signed off Feb · Tunde A.

    highmedium

  • Agent network expansion

    DPIA-2026-0003 · updated 5 days ago · Segun O.

    mediumlow

  • Merchant analytics dashboard

    DPIA-2026-0001 · signed off 6 weeks ago · Blessing U.

    mediumlow

  • Customer support chatbot

    DPIA-2025-0014 · signed off Jan · Blessing U.

    lowlow

  • Staff device monitoring

    DPIA-2026-0005 · started yesterday · Segun O.

    nonenone

DPIA-2026-0004 · credit scoring model v2

Five fixes agreed. Two of them exist

Launch is planned for 15 August. Eri's note to the approver is one sentence long: human review of declines and the appeal route are both only planned — ask for those two before anything else, they are what makes the residual risk defensible.

Until they land, this model decides who gets credit with no person in the loop and no way to contest it. That is the version of §37 the NDPC cares about.

Someone it declined

  • Delete scoring inputs 12 months after the decision

    Built into the retention rules and running.

    In place

  • No health or disability data in the model

    Enforced at the feature layer, verified by Asiri against the field list.

    In place

  • A person reviews every decline

    The fix the whole residual rating depends on. Nobody has been assigned to do the reviewing.

    Planned only

  • An appeal route customers can actually use

    Promised in the assessment, absent from the product. Zainab Yusuf has already asked for one.

    Planned only

  • The privacy notice mentions automated decisions

    Notice v4 does not say a model decides on loans. It has to, before launch.

    Planned only

When it stops being optional

An assessment written after launch is a report, not an assessment

The point of the exercise is that it can still change the design. Asiri asks the four questions when you connect a system or add a purpose — while the answer is cheap — rather than producing the document once the thing is already live.

And where the risk stays high after your fixes, the NDPC has to be consulted before you launch. Allow four to six weeks. That is a calendar fact, not a formality.

  1. A machine decides something that matters

    Credit, employment, access to a service. If a person is not in the loop, an assessment is compulsory and so is a way to contest the outcome.

  2. Biometrics, health, or anything special

    A face match, a fingerprint, a sick note. Sensitivity is not reduced by good intentions or by strong encryption.

  3. Systematic monitoring of people

    Staff device logging, agent activity tracking, public-space cameras. The question is proportionality, and it needs answering in writing.

  4. Children, or data at real scale

    Anything aimed at under-18s, and anything touching a large share of your customers. Both push an ordinary purpose into assessment territory.

Twelve templates, and none of them let you skip the hard question

Credit decisions, a new AI feature, face or fingerprint checks, identity at sign-up, marketing to existing customers, behaviour tracking, call recording, branch cameras, a new supplier, moving data abroad, staff device monitoring, health or disability data. Each arrives with its risks and fixes pre-filled, and each refuses to reduce a risk on a promise.

Talk to usHow long you keep it