Notes · Nov 13, 2026 · 5 min read · AI

Is your marketing data ready for AI? A practical checklist

Most AI projects in marketing stall on the data, not the model. Here is what "AI-ready" actually means.

Romel PrashadPrincipal consultant, marketing technology & data

Every marketing team I talk to wants an assistant that answers questions about performance, writes better emails, or picks the next best offer. The demos look easy. Then the assistant says renewals were up 12% when finance says 4%, and trust is gone in a week.

The model was rarely the problem. The data underneath it was. Here is the checklist I use before connecting AI to marketing data.

1. One definition per metric

If "active subscriber" means three different things in three reports, an AI assistant will confidently pick one of them. Write the definitions down, build them once in the warehouse, and point every tool, human or AI, at those tables.

2. Data that arrives on time and complete

Pipelines need checks for row counts, missing values and freshness, with alerts that go to a person. An assistant answering from yesterday's half-loaded table is worse than no assistant.

3. One customer, one profile

If identity is not resolved, "how many customers bought twice" has no reliable answer. Match emails, logins and device IDs first, and measure how many records merge.

4. Documentation a machine can read

Table and column descriptions, metric definitions and example questions are what let a language model write the right query. The same documentation helps new analysts, so it pays off twice.

5. Guardrails on what AI can see and do

• Give assistants read-only access to governed, aggregated tables, not raw personal data.
• Keep consent and opt-outs attached to the data, so AI-driven audiences respect them.
• Have a person approve anything that goes to customers, at least at first.
• Log the questions asked and the queries run, so answers can be checked.

6. A way to prove it worked

If AI writes the email or picks the audience, test it against what you do today, with a holdout. Otherwise you will not know whether it helped.

AI does not fix messy marketing data. It repeats it, faster and more confidently.

None of this is glamorous, and all of it is what I spent years doing for large companies before AI made it urgent. If you want a second opinion on where your data stands, I am happy to take a look.