Is your company ready for AI? 7 signs your data is the real problem
Before you buy any AI tool, run this quick test: 7 signs your data will sabotage the project — and what to do about each one.
Most AI projects that fail don't fail at the model — they fail at the data. The tool works in the demo, the contract gets signed, and months later the answers come back wrong, incomplete, or not at all. Before investing, it's worth running an honest test: is your company actually ready for AI?
Building data platforms and AI environments for companies, we've learned to spot the warning signs from a distance. Here are the seven most common — and what to do about each.
1. Every department reports a different number for the same question
Monthly revenue according to finance: one figure. According to sales: another. If people already don't trust the numbers, no AI built on top of them will be trustworthy either.
What to do: elect one official source per indicator — before any tool, this is a management decision.
2. Closing the monthly report takes days (and involves copy-paste)
Manual spreadsheet consolidation is the classic symptom of scattered, unintegrated data. AI on top of that process just automates the mess.
What to do: automate the data collection behind those reports first. It's also a great first data project: visible value in weeks.
3. Critical data lives in spreadsheets and inboxes
Contracts in attachments, the customer base in a seventeen-tab spreadsheet, price history in someone's head. Data that isn't accessible in a structured way feeds no AI at all.
What to do: map where the critical data of each process lives — the map alone reveals where to start.
4. Nobody can say who has access to what
If data access is "everyone sees everything" (or the opposite: every dataset locked in its own system), any AI project becomes a privacy risk. Under laws like LGPD and GDPR, that's not a detail.
What to do: define owners and access rules for sensitive data before connecting it to any tool.
5. Your history is short, full of gaps, or unreliable
Demand forecasting, churn analysis, prioritization — they all learn from the past. If the recorded past doesn't reflect what happened, the AI confidently learns the wrong thing.
What to do: start recording properly now. Historical data can't be bought later.
6. Every system calls the same customer something different
In the ERP it's "ACME LLC", in the CRM it's "Acme", in support it's the buyer's email address. Without unifying who is who, any "360º customer view" — with or without AI — is fiction.
What to do: standardize identifiers in your core records (customer, product, supplier). Unglamorous work, enormous payoff.
7. Only one person "understands the data"
If the whole company depends on one person to extract any number, you don't have a data capability — you have a single point of failure.
What to do: document the extractions that person runs and turn them into automated, accessible routines.
Failed the test? Good — now you know where to start
No company passes all seven. The good news: you don't need to fix everything before starting with AI. The path we apply in our own projects is to pick one use case with clear value and fix the data that use case requires first — the foundation grows guided by results, not perfectionism. We explain the method in where your company should start and the typical investment in how much it costs to implement AI.
If you'd like an honest diagnosis of your scenario — no hype, no grand project — that's what our Data & AI consulting does. Talk to us.