Draft Documentation

This guide is currently in development. Content may be incomplete or subject to change.

~10 minutes

Datasets & Enrichment

Upload your own operational data — call lists, contact-center exports, campaign rosters — and Auralytik pairs each row with the matching AI evaluation. You get the data you already have, augmented with audio quality scores, typification, sentiment, and a deep link to the recording, all in one place.

What is a Dataset?

A Dataset is an Excel or CSV file you upload to Auralytik — typically a call-list export from your dialer or CRM. Once uploaded, Auralytik analyzes the columns, matches each row to its corresponding AI evaluation, and gives you back an enriched view that joins your operational data with Auralytik's quality & typification analysis.

Bring your file

Excel or CSV up to 100 MB / 250k rows

AI-tagged columns

Auralytik suggests what each column represents

Multi-strategy match

Customer ID → agent → time. Falls back gracefully.

Auditable

Every match shows its strategy and confidence

Excel export

One click to download original + enriched columns

Private or shared

You choose who else in your workspace can see it

When to use this: you already track calls or interactions in a spreadsheet or BI export and want each row paired with the AI's findings — call score, critical-error flag, typification, recording link — without exporting from Auralytik and reconciling by hand.

Uploading a Dataset

From the Datasets module, click New Dataset and walk through three steps:

  1. 1

    Name & visibility

    Give the dataset a name and choose Private (only you can see it) or Workspace (everyone in your workspace with the right permissions).

  2. 2

    Pick your file

    Drop in your .xlsx or .csv. Auralytik reads the first sheet, samples the rows, and asks an AI to propose what each column is — phone, timestamp, agent name, customer RUT, and so on.

  3. 3

    Confirm the schema

    Review and adjust the column types in the wizard. The dropdown values that drive the matcher are Customer ID (RUT), Agent Name, Timestamp, Phone, Campaign Name, and a few others. Pick which enrichment sources to enable, then click Confirm Import.

Dataset upload wizard with AI-discovered columns

Screenshot coming soon

Heads-up: the higher the AI confidence, the more you can trust the suggestion. Anything tagged below ~80% is worth a second look — especially the columns that the matcher depends on (RUT, Agent Name, Timestamp).

Schema & Column Types

Every column gets a semantic type. Most are display-only, but a few drive the matcher:

TypeWhat it does
Customer ID (RUT)Strongest match key. Joins directly to the customer ID on the evaluation. Strip non-digits — "12.345.678-9" and "123456789" map to the same key.
Agent NameDisambiguator. Normalized (lowercased, prefixes like CBZ_C1_ stripped, accents removed).
TimestampUsed to pick the nearest call when several match. Naive times are interpreted in your workspace's timezone.
PhoneReserved for phone-based matching (planned). Stripped to the last 9 digits for Chilean numbering.
Campaign NameDisambiguator for time-only matches.
Agent TypificationMarks a column that holds the agent's manual call result — used later to compare against the AI's typification.
Operation ID, Agent ID, etc.Stored verbatim. Available to future match strategies and to the Excel export.
Pass-through / UnknownCarried through but not used for matching. Pick this for columns that are display-only.

How Matching Works

Auralytik tries to pair each row in your file with one evaluation in the platform. It works through a series of strategies, strongest first, and stops at the first hit. The strategy used is recorded on every match so you can audit it.

Confidence 0.95rut+agent+same-day

Customer ID + agent name + same calendar day. The precise, high-confidence case.

Confidence 0.75rut+agent+nearest-day

Same customer + same agent, but the eval is on a different day within the search window. Common for follow-up calls.

Confidence 0.65rut-only-cross-agent

Same customer, agent disagrees. Useful when the file's agent column is stale or the call was transferred. Worth a quick audit before trusting the typification.

Confidence ≥ 0.65agent-name+timestamp-Nmin

Fallback when no RUT match exists. Pairs by agent name + the call's timestamp within a window you control (defaults to 15 minutes).

Why some rows don't match: the customer ID from your file may simply not have an evaluation in the time window — the call may not have been recorded, the customer may not have been contacted, or the call may have been classified as invalid. Auralytik will tell you exactly how many rows fall in each bucket and which strategy was used for the ones that did match.

Reviewing Results

Once the import finishes, the dataset detail page shows a summary and the row-by-row grid:

Total rows

Every row your file contained, including ones that didn't match.

Matches

Rows where the matcher found at least one evaluation. % shown is of total rows.

Ambiguous

Rows where two or more evaluations were equally good fits — the matcher picks the nearest by time and flags the row.

Typification disagreements

Where the agent's manual call result differs from the AI's — your strongest QA signal.

Dataset detail page with summary tiles and grid

Screenshot coming soon

Click any row to open the drill-down panel. From there, you can:

  • See the AI's scoring, critical-error flag, and typification.
  • Open the evaluation deep-link to listen to the recording.
  • Check which match strategy was used and its confidence.

The toolbar's Export to Excel button downloads the original file plus enrichment columns — typification category, average note, sentiment, evaluation deep-link, and the matcher's confidence — so you can hand the file back to whoever provided it with the AI's analysis baked in.

Editing Schema & Mapping

After confirming the import, you can still change how the dataset is matched without re-uploading:

Edit Schema

Open the column-type dialog and re-tag any column. Saving triggers a re-parse: Auralytik rebuilds every row from your original file using the updated tagging, then re-runs enrichment. Use this if the AI mis-tagged a column or if you forgot to mark the RUT column as a Customer ID.

Edit Mapping

Tune the match settings — enabled sources, time-window minutes — without re-parsing the file. Saving triggers a re-enrich only. Use this to widen or narrow the matcher's tolerance.

Re-parse vs re-enrich: a re-parse re-reads the blob and rebuilds every record (1–3 minutes). A re-enrich keeps the records as-is and only re-runs the matcher (faster). The dialog warns you which one is about to happen.

Permissions

Datasets is governed by a dedicated Datasets permission area in your workspace settings:

  • View — see workspace-visible datasets and their results.
  • Create — upload new datasets.
  • Edit — change schema/mapping and trigger re-enrich on existing datasets.
  • Full — also delete datasets.

Visibility, separately: even with View permission, a user can only see datasets marked Workspace. Private datasets are visible only to whoever uploaded them. Switch the visibility at upload time, or never — there's no exposing a private dataset by accident.