Data readiness review

A diagnostic inside stage 01 of the KARPINITY Method: in 1 to 3 weeks, every data source an AI use case depends on is scored, and anything below the bar gets a remediation plan.

Duration: 1-3 weeks
Price: On request, scoped to your use case
What you receive: A data readiness scorecard, source by source, scored on completeness, accuracy, accessibility and lineage.
A remediation plan for every source below the bar: the fix required, an owner and a rough effort.
A plain statement of which sources are ready today and which are not.
What this doesn’t include: Fixing the data. The plan is scoped and costed. Executing it is separate work, done by your team, ours or a third party.
Building the prototype. That is the proof of value sprint, stage 02.
Data migration, tooling procurement or master data management as a programme. These can be the right long-term answer but fall outside a time-boxed review.
A verdict fixed in advance. If the data is not ready, you learn it in three weeks instead of three months into a build.
Who you need involved: The owner of each system of record in scope: one to two hours per source, mostly access and a walkthrough of what the data contains against what the schema claims.
An IT or platform contact who can grant or coordinate access: a few hours in total.
Optionally, a security contact if a source holds regulated or sensitive data: one to two hours.

What it is

A fixed-scope diagnostic that scores every data source an AI use case depends on, with a remediation plan for the ones that fall short.

It is a diagnostic inside stage 01 of the KARPINITY Method, not a stage of its own. Commission it when the data is the main risk: the sources are scattered, the labels are unreliable, or nobody is sure who owns what.

We work with the owner of the system of record, not the owner of the reporting layer built on top of it. The reporting layer tends to hide the problems the review looks for.

The success metric, defined up front

In week 1 we agree the bar with you in writing: the minimum score each source needs on four dimensions, set against the use case’s metric.

Completeness: are the fields and records the use case needs present?
Accuracy: do the values match what they claim to describe?
Accessibility: can the system reach the data, legally and technically?
Lineage: do we know where each value came from and what changed it?

Each source passes or fails against that bar, and the bar does not move once the build starts.

Week by week

Week 1

Inventory the sources the use case depends on, working from the business case if one exists.
Confirm access: who owns each source, the access path, and whether getting it is realistic.
Start scoring the first sources on the four dimensions.

Week 2

Score the remaining sources.
Where a source falls below the bar, diagnose why: missing fields, unreliable labels, access friction, unclear provenance.
Draft the remediation plan for each gap: what changes, roughly what it takes, who owns it.

Week 3 (for larger source counts, or where remediation options need more investigation)

Validate the plan with the source owners.
Finalise the scorecard and walk your side through it.

What you decide at the end

You decide one of three things:

Prove the value. The sources score above the bar. The next step is the proof of value sprint.
Remediate first. Some sources fall short. You run the scoped plan on your own timeline and return when the data can support the use case.
Change the scope or stop. The data cannot reach the bar at a cost the business case supports.

Questions

Do we have to buy the assessment first?

No. If you already have a business case and a named metric, you can start here. Without a metric the review has no bar to score against, and we tell you so in the first week.

How many sources can one review cover?

The number of sources and how hard they are to reach set the duration of 1 to 3 weeks. We fix scope and price after a short call, before you commit.

Can you fix the data afterwards?

Yes, as a separate engagement with its own scope. The plan is written so your team or a third party can execute it, and nothing in the review ties you to us.

What if the data is sensitive or regulated?

A security contact on your side joins for an hour or two. Where an AI system would touch personal data, a data protection impact assessment is scoped in stage 01. See how we build and handle data.

What does the scorecard look like?

One row per source, one score per dimension, a pass or fail against the agreed bar, and the remediation item where it fails. You can forward it to the data owners and to your security and legal functions.

What does it cost?

Price is on request, scoped to the number of sources.

What happens next

If the sources score well, the next step is the proof of value sprint. If they do not, you hold a costed plan and set your own timeline. Start a project and tell us which data the idea depends on.