FOXNUT AI

Tool

Data readiness check: is your data ready for AI?

This check asks five questions about your data and returns ready, almost or not yet. The result sets where the work starts, not whether it happens.

By the Foxnut team · Updated

What does this data readiness check tell you?

This data readiness check tells you whether your data is ready for an AI project, using the five questions our founders ask before starting one, in the order they ask them. Five answers produce one of three results: ready, almost, or not yet.

We are Foxnut AI, a senior-only practice in Bengaluru and Paris that builds AI systems and hands them over. There’s a link to how we work at the end of this page. Let’s check your data first.

None of the three results decides whether we would take the work on. That is our founders’ own rule. The answers set where an engagement starts. If the data is not properly structured and maintained, the work begins by fixing that, before any AI implementation, so a below-ready result here is a starting point and not a rejection.

Everything the check needs is on this page in plain text: the questions, what each one tests, a worked example with its result already computed, and what to do if the answer is not yet. If you came for the check itself, jump to the check. This is one of the tools this studio publishes, all built to the same rule: usable without contacting anyone.

What are the five data readiness questions, and why this order?

The five questions below are our founders’ intake questions in their own words:

  1. Is all your data accessible digitally?
  2. Do you have data sanity checks in place to identify and fix corrupted data?
  3. Have you implemented access control for your data?
  4. Are you compliant with the data protection and privacy laws of your country, such as GDPR in Europe or DPDP in India?
  5. Which systems do you currently use for data storage?

Two of the five are gates. Data that is not digitally accessible cannot be reached by any system, whatever else is true. Data whose storage systems nobody can name is undocumented in the way that matters, because there is nothing for a builder to connect to. Answer no to either and the result is not yet, because no amount of engineering works around those two states.

The middle three questions score and do not gate. Corrupted records, uncontrolled access and unresolved privacy obligations are all real gaps, but they are the kind an engagement can close on the way in.

The fifth question is not a yes-or-no question, so the check restates it as one: can you name the systems your data is currently stored in? A list of five tools and a spreadsheet is a pass.

Confusion to avoid: messy and scattered is not the failure state here. Our working boundary for data readiness treats data spread across many tools as workable, and this check applies the same boundary. What fails is not knowing where the data lives at all.

What are the five criteria?

The five criteria are in the table below, and the table is the whole instrument. With JavaScript off you can self-assess from it and reach the same answer the check gives.

The criteria, what each one tests, and what a no usually means
CriterionWhat it testsIf no
All of your data is accessible digitallyWhether an AI system can reach the data at allGate. The work starts with digitisation, before any AI
Data sanity checks are in place to identify and fix corrupted dataWhether the AI would read from records nobody has checkedCorrupted records surface later as confident wrong answers
Access control is implemented for your dataWhether who can see what is a decision, not an accidentAn AI system inherits every access gap it is given
You comply with the data protection and privacy laws of your country, such as GDPR in Europe or DPDP in IndiaWhether the system would process personal data lawfullyA compliance gap widens once software starts reading the data
You can name the systems your data is currently stored inWhether the data is documented well enough to hand overGate. The work starts with mapping where the data lives

Read each row as a statement about your situation today, not about where a clean-up project is supposed to land next quarter. The two gate rows sit first and last because that is where our founders put them. The first question establishes that the data can be reached, and the last that somebody can say where it is.

What does a worked example of this check look like?

We constructed the example below for this page, and it is not a client case. A 38-person distribution company keeps sales orders in an ERP, invoices in billing software, delivery schedules in spreadsheets and customer conversations in a shared inbox. That is four systems, all digital, and the operations lead can name every one of them. Access is role-based, and the customer records it holds are managed under the country’s privacy law.

What it does not have is sanity checks. Duplicate vendor records and stale price lists are known to exist, and nothing catches them.

Run through the criteria table, that case passes both gates and meets two of the three scored criteria. The check returns almost, the result already computed in the check below. The gap is named, and it is exactly the kind an engagement closes first.

The benchmark table that follows states what our founders have said about how these answers are used. Every value in it is theirs to state and is marked accordingly.

Benchmarks to read your own numbers against
BenchmarkValueSource
What a below-ready result decides about working with the studioNothing - the answers set where an engagement starts, not whether the studio takes a client onThe founders' stated intake rule
Where an engagement starts when the data is not properly structured and maintainedWith fixing that, before any AI implementation beginsThe studio's stated engagement practice

How do you get data AI ready if the answer is not yet?

You get data AI ready by doing two plain jobs: get every record an AI system would be expected to use into digital, machine-readable form, and write down each system the data lives in and who owns access to it. A not-yet result means one of two things: some of the data is not digitally accessible, or nobody can name the systems it lives in. In both cases the work belongs to your organisation, not to an outside builder.

Neither step needs a platform or a framework. An organisation that has done them knows its own operation better than any assessment would have told it.

We do not treat this state as a refusal. When data arrives with us in this condition, the engagement starts with fixing it, because that groundwork is the first phase of the work and not a precondition for a conversation. Do it alone or do it with help. Either way the AI implementation comes after, and it goes better for the wait.

How do you run the check?

The check

Worked example, as served

  1. All of your data is accessible digitally.Orders, invoices, schedules and conversations all live in digital systems.Yes
  2. Data sanity checks are in place to identify and fix corrupted data.Duplicate vendor records and stale price lists are known to exist; nothing catches them.No
  3. Access control is implemented for your data.Access is role-based; who can see what is a decision.Yes
  4. You comply with the data protection and privacy laws of your country, such as GDPR in Europe or DPDP in India.Customer records are managed under the country's privacy law.Yes
  5. You can name the systems your data is currently stored in.The operations lead can name all four systems.Yes

Result

Almost. The data is workable, with a gap to close first.

Almost. The worked example passes both gates: the data is digitally accessible, and every system it lives in can be named. Two of the three scored criteria are met; what is missing is sanity checks, so the ERP’s duplicate vendor records and stale price lists would reach an AI system unchecked. In the founders’ stated practice, that result decides nothing about whether the work happens - it sets where it starts. An engagement in this state begins by putting the missing checks in place, then builds on data that has been made trustworthy.

What happens if you do reach out?

The same five questions run in person. The first conversation at Foxnut AI covers your answers to exactly this list: where the data lives, whether it can be reached, what state it is in, and what that means for where an engagement would start. So what the studio’s first conversation actually checks is this instrument, held by a person.

A ready result means the work can start at the AI itself. An almost or not-yet result changes the first weeks of the work, not whether the conversation is worth having. Our founders’ rule, one last time: these answers set the foundation for where the work needs to start, never whether we take a client on.

Most firms would sell you a data project and call it a phase. We work the other way round: the data work and the AI work sit with the same senior people, there is one tier, and the people who sign the proposal do the work. We made and launched our own products first, and what we build is handed over running. You can read about Foxnut Consulting, the practice behind this check, before you talk to us, and you will get a direct response within one business day.

Foxnut AI takes on work like this from Bengaluru and Paris. If you want to see how we do it before you talk to anyone, here is the range for each engagement shape.