Veltos.Tech

AI and ML

What to Check in Your CRM and 1C Before an AI Project So the Budget Isn’t Wasted

Most failed AI projects run into the data, not the model — and the client usually only learns the data’s real quality after the money is already spent.

In short

Before paying a vendor for an AI project, it is worth checking four things in your own CRM or 1C first: whether required fields are actually filled on real, not test, records; whether data formats (dates, statuses, categories) stay consistent across employees and departments; whether client or deal records are free of duplicates; and whether there is even enough historical data volume for the task. A project built on top of data with these problems will almost certainly produce a disappointing first result that gets blamed on "AI doesn’t work," when the actual cause is the data, not the model.

Why AI projects trip over data more often than the model

A model, even a less-than-perfect one, usually handles a task well enough if the data it runs on is clean, complete and consistent. The reverse also holds: even the best model on dirty, incomplete or contradictory data produces a poor, unpredictable result. That is exactly why diagnosing data readiness is a self-contained task the client can and should do themselves, before paying a vendor to discover the same problems in the project’s first week.

The four-item checklist

ItemHow to check it yourself
Required field completenessExport 100 random real records and count the share of empty fields
Format consistencyCompare how different employees fill in the same field type (dates, statuses)
Duplicate recordsCheck for clients or deals manually duplicated
Sufficient historical volumeCount relevant records from the last 6-12 months
What to check, and how

Why test records are misleading

Checking a few neatly filled cards an employee shows off during a system demo almost always paints a falsely positive picture. A real database, accumulated over years by different people with varying data-entry discipline, looks different, some fields are missing, some are filled in free-form where a standard format was expected, some contain the same client entered twice under slightly different names. That is exactly why the sample for checking needs to be pulled randomly from real, live data, not from showcase examples.

What to do if the data turns out not to be ready

Discovering a data problem before signing with a vendor is not a reason to abandon the AI project, it is a reason to change its first stage: instead of jumping straight into model development, the first paid stage becomes cleaning up the data, either with the internal team or as a separate, explicitly named part of the vendor’s work with its own timeline and cost, not buried inside a general "AI implementation" quote.

Frequently asked questions

Why can an AI project fail even with a good model?

Because even the best model gives a poor, unpredictable result on dirty, incomplete or contradictory data. Diagnosing data readiness is something the client can do themselves before the project starts, not something discovered after the fact at the vendor’s billing rate.

How do you check data readiness without bringing in a vendor?

Export a random sample of 100 real records and check required-field completeness, format consistency across employees, duplicate records, and sufficient history volume from the last 6-12 months. This can be done by the internal team in a day or two.

Why is checking demo records misleading?

Because a few neatly filled cards shown in a demo almost always look better than a real database accumulated over years by different employees with varying entry discipline. The check sample needs to be pulled randomly from live data, not showcase examples.

What if the data turns out not to be ready for an AI project?

Not abandoning the project, but making the data cleanup a separate, explicitly named first stage with its own timeline and cost, either in-house or as a dedicated part of the vendor’s work, not buried inside a general AI-implementation quote.

Need a hand with this?

We do this work, not just write about it. Describe the task and we will scope it and send a staged estimate.

Related services

Read next