Getting started

Your first reconciliation, step by step

Follow the versioned workflow from raw source files to a reviewed, validated output. Nothing here requires cloud AI.

Get a sample datasetTry the demo
Input

Start small

Use one reference dictionary and two or three source files for the first run.

Authority

Lock deliberate checkpoints

Dictionary, mapping, and translation locks make every downstream result reproducible.

Output

Verify before use

Treat the final metrics, conflicts, and validation issues as release evidence for the dataset.

Core workflow

Six controlled checkpoints

Work left to right. If an upstream file or rule changes, Datally identifies the downstream work that must be refreshed.

  1. Step 01

    Create a project and add files

    Create a project, add one reference file when you have a target schema, then add every source file you want to reconcile. CSV and Excel sources can use different column names and layouts.

    Confirm each intended file appears in the project before continuing.

  2. Step 02

    Review the dictionary

    Datally profiles the files and builds the working dictionary. Review names, types, formats, allowed values, and constraints. Make any corrections before locking a dictionary version.

    Lock a dictionary version when its target schema is ready for mapping.

  3. Step 03

    Map source columns

    Review suggested mappings file by file, resolve unmapped columns, and add computed custom columns where a target value must be derived. Keep ambiguous mappings manual.

    Lock the mapping only after every required target has an intentional source.

  4. Step 04

    Translate source values

    Standardize codes, labels, and categories into the dictionary vocabulary. Accept safe suggestions, edit uncertain values, and preserve explicit manual mappings.

    Lock the translation version used for the output you intend to create.

  5. Step 05

    Run validation

    Author or review validation rules, run them against the selected sources, and inspect exceptions. Correct the source, rule, or accepted exception instead of hiding failed records.

    Re-run affected rules after changes and review the current issue totals.

  6. Step 06

    Consolidate and download

    Select the intended source files and locked versions, choose exact-row duplicate handling, then consolidate. Resolve any surfaced conflicts and download the accepted CSV or Excel output.

    Verify row counts, source coverage, mapping coverage, and validation evidence before use.

Optional acceleration

Add AI without changing the workflow authority

AI can suggest mappings, translations, types, and rules. You still review the result and lock the accepted version.

Local-first by default

Install Ollama and choose local models in Settings to keep inference on your machine. If no provider is available, deterministic matching and every manual control remain usable.

Local AI setup

Troubleshooting

Common first-run questions

A stage is read-only or disabled

Complete the preceding stage and select or lock the required version. Datally keeps downstream results tied to the exact upstream versions that produced them.

A replacement file changed

Review the replacement analysis and any invalidated downstream work. Re-run only the stages affected by schema or value changes.

Local AI is unavailable

Continue with deterministic and manual workflows, or install Ollama and select models in Settings. Cloud AI is optional and must be enabled deliberately.

Contact support