Before AI, every new customer format was a manual engineering problem. A developer wrote a mapping script, another reviewed it, a third maintained it when the customer changed their schema. WeTransform replaces that cycle with five AI capabilities that operate on every file, automatically. This page explains what each one does.
Before AI: every new format was a custom integration
When a B2B SaaS product receives data from customers, two things are true simultaneously. Your system expects a fixed, validated structure. Your customers have their own naming conventions, their own data models, and their own interpretation of what a "clean" file looks like.
The traditional solution was code: a developer built import logic for each customer format, maintained it through every schema change, and debugged it when files arrived in unexpected states. The AI import management category exists because this pattern does not scale. At 10 customers it is manageable. At 100 it consumes a significant fraction of your engineering capacity.
The five capabilities below are what changes when AI handles the data side of customer file onboarding. Each one replaces a task that previously required a developer, a ticket, and a deployment.
Column mapping: AI reads any field name
When a customer exports their data, the column names reflect their internal systems, not yours. "client_ref" from one customer, "customer_id" from another, "ID client" from a third. Before AI, each variation was a new configuration entry: open the mapping table, add the alias, test it, deploy it.
AI column mapping removes this for known patterns. When a file arrives, the AI scans the column headers, samples a subset of rows, and matches each source column to a target field in your schema. It compares column names against patterns it has learned across thousands of previous imports and assigns a confidence score to each candidate match.
Matches above the threshold apply automatically. Matches below it appear in the review interface for a human to confirm. Once a reviewer validates a mapping for a customer, that customer's column profile is saved. Every subsequent file from the same source passes through automatically.
The practical outcome: the first file from a new customer requires one review session. From the second file onward, their format is a solved problem.
Value mapping: AI normalizes what column names cannot
Column names are the first layer of variation. Values are the second.
A customer sends "oui" and "non" for a boolean field your schema expects as 1 and 0. Another sends "FR" where your system expects "France". A third sends product categories that do not match your taxonomy: "Accessoires" where your schema expects "Accessories". These are not wrong values. They are the values that made sense in the customer's own system.
Value mapping recognizes that two different strings represent the same concept. WeTransform's AI identifies equivalences between source values and your target taxonomy, assigns confidence scores, and applies matches automatically where confidence is high. For mappings below the threshold, a reviewer confirms in a single session. Once confirmed, those equivalences become permanent rules for that customer.
Fill the gaps: AI infers what the customer did not send
Sometimes a customer's file is missing a field your schema requires, not because of an error, but because their internal data model does not have that field. Before AI, this produced a validation failure: the file was rejected, a support ticket was opened, the customer had to add the missing column, re-export, and re-upload. Each cycle added days.
Fill the gaps changes this. For missing values that can be inferred from context, the AI proposes what the field should contain. If a product export lacks dimensions but includes a product reference and a category, the AI can infer typical values from similar products in the same category. The reviewer validates the inferred values before delivery. Customers receive fewer errors; your team handles fewer tickets.
The capability is most impactful in catalog onboarding use cases, where supplier files routinely omit standard attributes that a customer success team would otherwise have to request manually.
Autoclean: AI removes noise before it reaches your database
Customer files contain noise. Trailing spaces in string fields. Currency symbols inside numeric columns. Duplicate rows from export artifacts. Encoding issues from Windows Excel exports. Line breaks inside text cells that break JSON serialization downstream.
Before AI, this noise either triggered validation errors or landed in your database as dirty data. The fix was manual: identify the pattern, write a cleaning rule, deploy it.
Autoclean applies a battery of standard operations automatically on every file before validation runs. Trailing whitespace is stripped. Numeric fields with currency symbols are converted to clean numbers. Duplicate rows are detected and flagged. Encoding issues are corrected. The AI identifies what cleaning each column needs based on the data it contains and the target field type it maps to.
Because Autoclean runs before validation, rows that would have failed validation due to avoidable formatting issues arrive at validation clean. Your validation error rate drops. Your support queue shrinks.
Natural language rules: validation without code
Validation rules define what your schema enforces before delivery. In traditional systems, these rules are written in code: regex patterns, range checks, conditional logic. Each rule requires a developer to write, review, and test it. Each business change that requires a new rule opens a ticket.
WeTransform allows validation rules in plain language. "All prices must be positive numbers." "Country must be a valid EU member state." "Quantity cannot exceed 10,000." These are entered as plain text. The AI interprets the instruction, translates it into enforcement logic, and applies it on every subsequent import.
This matters operationally because rules evolve. New verticals, new compliance requirements, new data partners bring new constraints. With natural language rule configuration, a product manager or operations lead can update validation logic without opening an engineering ticket. The people who understand the business requirements can express them directly, without a translation layer.
These five capabilities are what WeTransform's AI import management does at the file level. Together, they make customer data onboarding predictable at any scale: the first file from a new customer takes minutes to configure, every file after that runs automatically.
To see all five in action with your own use case, book a 20-minute demo or see WeTransform pricing.