Data and imports

Most businesses do not begin with clean, unified data.

Customer and operational information may be spread across spreadsheets, a CRM, Notion, shared folders, PDFs, line-of-business software, and handwritten or informal notes. Bringing that information into a new operation is not only a file-upload task. It requires decisions about meaning, identity, quality, and ownership.

Why copying the source is not enough

Imagine a spreadsheet with columns named Name, Status, Owner, and Notes.

Does Name mean a person or a company? Does Status describe a customer, an opportunity, or a follow-up task? Is Owner the account manager or the person who entered the row?

Copying those columns into a new database preserves the uncertainty. A useful import first determines which business concepts exist and how they should behave.

Rows can then become records with stable identity. Columns become typed fields. References become relationships. Status values become controlled state. Actions, workflows, and permissions can operate on that structure.

A dependable import journey

Inspect

Profile the source without changing it. Identify IDs, sample values, missing information, duplicates, formulas, inconsistent formats, and relationships.

Model

Translate the source into business concepts. One flat table might actually contain companies, people, opportunities, tasks, and activities.

Review

Confirm field types, identity rules, transformations, duplicate handling, and which system will own each value after the import.

Validate

Normalize values, detect collisions, and surface unsupported rows. Ambiguous mappings should remain unresolved rather than being silently guessed.

Commit

Create or update records through controlled actions while preserving source references, warnings, and the mapping version.

Reconcile

Compare totals, inspect failures, resolve duplicate candidates, and test the views and workflows that depend on the imported data.

Each stage should produce something a person can inspect before the next stage changes durable state.

Decide which system is authoritative

The source of truth is the system whose value wins when connected systems disagree.

That decision can differ across the business.

Migration

Data moves into Paraline, and Paraline becomes authoritative for the selected operation. This may fit a temporary spreadsheet or retiring tool.

Synchronization

Paraline and another system continue exchanging changes. This may fit a CRM or accounting platform that must remain active.

Reference

The external source remains authoritative. Paraline uses the information as read-only context or links to it.

Two-way synchronization is often the most complex choice. It requires rules for identity, ordering, conflicts, deletion, retries, and partial failure. It should be used because the operation needs continuing exchange—not because it sounds more complete.

A hybrid model is common. A CRM may own customer identity, an accounting platform may own invoices, and Paraline may own service cases and review state.

Provenance explains where information came from

Source of truth describes authority. Provenance describes history.

A useful imported value can identify:

• the source system and external record;
• the import or synchronization run;
• the mapping version and transformation;
• warnings or uncertainty;
• the person who approved the result;
• and changes made after import.

This matters for people and AI. A reviewer can check the source behind a questionable field. An agent can distinguish a verified record from a derived summary or suggestion.

Identity and duplicates deserve special care

The import needs a rule for deciding whether a row represents a new record or an existing one.

An approved external ID is usually stronger than a name. Email, phone, and company domain may help, but they can change or be shared.

The business also needs merge rules. Which value wins? Can an empty source erase existing data? How are relationships combined? Can the merge be reversed?

Low-confidence or high-impact merges should remain reviewable.

Documents need evidence, not only extraction

PDFs, scans, emails, and images do not arrive as clean rows and columns. AI can identify document type, extract proposed values, and connect answers to source locations.

The extracted values should still pass through mapping, validation, ownership, and provenance rules.

For consequential work, each important field should link to the page or source region that supports it. Unsupported fields should remain blank or unresolved instead of being filled from general AI knowledge.

Confidence can help prioritize review. It does not replace source evidence.

Imports are powerful, permissioned actions

Connecting a source can expose sensitive data. Committing an import can change thousands of records.

The business may separate permission to:

• connect or upload a source;
• inspect and map it;
• preview proposed changes;
• approve the import;
• execute it;
• and reconcile failures.

Every run should report records created, updated, skipped, warned, failed, and left unresolved. A successful job summary should not hide bad rows.

Validate the resulting operation

Row counts are only the beginning. A good validation also checks:

• required-field coverage;
• relationship accuracy;
• duplicate rate;
• transformation quality;
• failed and unresolved records;
• source ownership;
• permissions;
• and the workflows that use the new state.

Customer pages should show the right relationships. Queues should contain the expected work. Restricted data should remain restricted. Automated rules should respond correctly.

Paraline’s intended role is to make this path explicit and reviewable. Production connectors, mappings, and synchronization remain planned until listed as available in Capability status.