Do I need a data warehouse before using AI?

Not necessarily. You need the data for your specific use case to be clean and accessible. Many successful AI projects start with a simple, well-structured database. The full data warehouse can be built in parallel as use cases accumulate.

How long does it take to prepare data for an AI project?

For a contained use case, two to four weeks: identifying sources, removing duplicates, unifying catalogs and setting up programmatic access. A prior data readiness assessment lets you estimate the real effort before committing budget.

What if all my data is in Excel?

It's the most common starting point and it has a direct solution: spreadsheets are migrated to a structured database, single catalogs are defined, and loading is automated so Excel stops being the source of record. What matters is data consistency, not the original tool.

Data

Data Readiness: How to Prepare Your Company's Data for Serious AI

Okun Data Team · April 8, 2026 · 5 min read

The phrase comes up in every failed project: "the AI didn't work." Dig deeper and the problem is almost never the model: it's the data. Incomplete, duplicated, locked inside PDFs, scattered across fifteen spreadsheets with conflicting criteria. Artificial intelligence doesn't fix bad data; it amplifies it.

Before investing in any AI initiative, it pays to answer a more boring but more important question: is my data in shape? That's called data readiness, and this article gives you a practical framework to evaluate it.

The four dimensions of data readiness

Access: can the data be queried programmatically (database, API), or does it live in emails, PDFs and personal spreadsheets? AI needs connectable sources, not screenshots.
Quality: are there duplicate records, empty fields, inconsistent entry criteria? A customer loaded three times under three different names breaks any analysis.
Structure: is there a data model, even a simple one? Tables with consistent keys, single catalogs for customers and products, dates in a uniform format.
Governance: who owns each piece of data, who can see it, and who fixes it when it's wrong? Without owners, quality degrades on its own.

The 15-minute test

A quick way to know where you stand: ask your team for sales by customer for the last 12 months, with margin by product. If the answer arrives in 15 minutes from a system, your data is reasonably ready. If it takes two days and three people cross-referencing spreadsheets, there's your real first project — and it's not AI, it's data architecture.

Preparing data is not a year-long project

The good news: data readiness doesn't mean perfection. You don't need a textbook corporate data warehouse to start with AI. You need the data for the chosen use case to be clean and accessible. If the goal is an agent that reconciles payments, cleaning up bank statements and invoicing is enough; the rest of the company can wait.

This surgical approach — clean the minimum needed for the first use case and expand from there — cuts preparation time from months to weeks. It's the same principle we apply in business intelligence projects: value appears when the first dashboard gets used, not when the architecture is complete.

Signs you're ready

You can move forward with confidence when: the sources for the use case are identified and accessible; someone who knows that data can validate results; the key catalogs (customers, products) have no serious duplicates; and some historical record exists — AI learns from the past, and without a past there is no learning.

Conclusion

The right question isn't "what can AI do for my company?" but "what data do I have in shape to feed it?". Companies that spend two or three weeks on an honest data assessment before starting avoid 80% of AI project failures. Bonus: clean data improves everything else too — reporting, decisions, and daily operations.

Not sure whether your data is ready for AI?

We run a data readiness assessment of your company in one week, with a concrete action plan.

Request a demo

Frequently asked questions

Do I need a data warehouse before using AI?: Not necessarily. You need the data for your specific use case to be clean and accessible. Many successful AI projects start with a simple, well-structured database. The full data warehouse can be built in parallel as use cases accumulate.
How long does it take to prepare data for an AI project?: For a contained use case, two to four weeks: identifying sources, removing duplicates, unifying catalogs and setting up programmatic access. A prior data readiness assessment lets you estimate the real effort before committing budget.
What if all my data is in Excel?: It's the most common starting point and it has a direct solution: spreadsheets are migrated to a structured database, single catalogs are defined, and loading is automated so Excel stops being the source of record. What matters is data consistency, not the original tool.