# Enrichments

In addition to the raw job data, the Fantastic.jobs API includes several enrichments. Our goal is to provide you with as many useful job and company enrichments as possible.

## locations_derived

Some ATS platforms are more diligent than others in keeping their location data accurate and structured. To create consistency between sources we run every job's location through a normalization step that returns values in a consistent `city/county, region, country` format.

The normalization is powered by [Geoapify](https://apidocs.geoapify.com/playground/geocoding/#autocomplete), which is built on OpenStreetMap data.

If you already maintain your own location-normalization pipeline, you can use the raw `locations` and `locations_alt` fields over `locations_derived` if you so wish.

## AI_ fields

We run an LLM over each job's description, title, location, and several other fields (including some job page elements that we don't return in the API) to extract structured insights: salary, benefits, experience level, detailed remote/work-arrangement signals, and more. We cover roughly 99.9% of all jobs and have a failover system between several LLMS to ensure that these fields are (almost) always set.

Good to know:

- LLM Extractions are performed by a probabilistic llm model. While it can make mistakes, we've finetuned our process to make this very minimal
- No outside information is included in this process, the LLM will exclusively use the context of the job's data

See the [API Reference](/api/new-jobs) reference for the full list of `ai_*` fields.

## Organization enrichment

We enrich ATS jobs with the matching LinkedIn company profile, plus additional company context from Crunchbase, Glassdoor, and our own derivations. The LinkedIn mapping process covers over 99% of jobs at roughly 99.4% accuracy - if you spot an incorrect mapping, please report it and we'll correct it in our database.

The enrichment is available in two tiers:

### Basic

Opt in inline on `active-ats` / `modified-ats` via `include_basic_organization_details=true`. Covers the most common LinkedIn fields - name, logo permalink, industry, headcount, size, followers, HQ, locations, type, founded date, slogan, description, specialties, and the recruitment-agency flag - plus two Crunchbase fields: `org_crunchbase_categories` and `org_crunchbase_total_investment`.

`active-jb` already returns the basic LinkedIn fields inline on every job - no flag needed, since the org is read directly from the job board.

### Advanced (beta)

Available only via the [`ats-organizations-advanced`](/api/ats-organizations#advanced-organization-details) endpoint. Join back to ATS jobs by `org_linkedin_slug`. On top of the basic set, advanced adds:

- **Extra LinkedIn fields:** LinkedIn URL / ID, short description, full industry list, estimated revenue bounds, largest headcount country
- **Industry classification:** NAICS, SIC, market segments
- **Crunchbase fiscal data:** fiscal year end, acquisition status, IPO date
- **Glassdoor:** scalar ratings (work-life balance, compensation, culture, etc.) and review / salary / interview / benefit counts
- **Other profile links:** Twitter, Crunchbase profile, Glassdoor profile

Heavy nested fields (funding rounds, headcount / follower / revenue timeseries, news articles, Glassdoor reviews) are excluded by default and must be opted into individually via dedicated `include_*` flags. See [Advanced Company Details](/documentation/endpoints/advanced-company-details) for the full field list and flag names.

## `domain_derived`

ATS platforms don't always expose the employer's domain or homepage, so we derive it ourselves. This is useful for joining against external datasets or running further enrichments.

Coverage and accuracy:

- ~96% of jobs have a populated `domain_derived` value.
- Accuracy sits around 98% on the populated subset.
- Accuracy is highest for medium-to-large US companies and lowest for non-US companies and companies with generic names.

## `exclude_ats_duplicate`

We've released a beta cross-feed deduplication system for users who consume both the ATS and job board datasets. Every LinkedIn job is checked against the ATS dataset using two signals:

- Job title + organization name + country
- Job title + LinkedIn company profile + country

The result is exposed as `ats_duplicate` on every `active-jb` row:

| Value | Meaning |
| --- | --- |
| `true` | At least one of the two checks matched an ATS job - this LinkedIn listing is a likely duplicate. |
| `false` | Neither check matched - this listing is treated as unique to the JB feed. |
| `null` | Job was not eligible for the dedup check (see below). |

Setting `exclude_ats_duplicate=true` on `active-jb` drops the rows flagged `true`.

### Known false-positive sources

The dedup logic looks for exact matches only, so the priority is precision over recall - some duplicates will still slip through. During internal testing, the residual false positives clustered around the following patterns:

- The LinkedIn listing is indexed before the ATS listing.
- The job is older than 6 months on the ATS platform but newly indexed on LinkedIn. Jobs older than 6 months are considered expired and excluded from the dedup checks. This can also be a positive signal - the employer is re-promoting an older listing.
- The LinkedIn listing was posted via a programmatic platform (e.g. Appcast, Adzuna) with minor edits to the job title or organization name.
- The LinkedIn listing uses a different title or organization name than the ATS listing.

To fully eliminate cross-feed duplicates, we recommend layering a fuzzy deduplication pass on top of this flag.