Extract data from performance reviews

A performance review is the record of how a manager rated an employee over a period, and HR keeps it as the file that later decisions point back to. Whether it is an annual appraisal, a mid-year check, or a probationary review, it carries the same structure: the employee and their role, the reviewer, the period covered, an overall rating, ratings against individual competencies, the goals set and met, the strengths and development areas, and the employee's own acknowledgment. When Dana Osei, VP of Engineering at Northgate Logistics Inc., reviews James Whitaker, a Senior Backend Engineer, for the 2025-01-01 to 2025-12-31 period and records an overall 4.2 out of 5 on 2026-01-20, that appraisal becomes a data point HR tracks across the review cycle. The part a reader has to keep consistent is the rating structure. An overall score is usually the roll-up of individual competency ratings, so a review that scores technical execution 4.5, ownership 4.3, collaboration 4.0, and communication 4.0 has an overall of 4.2 that should reconcile to the average of its parts. Goals carry a target date and an achievement status, so an objective due 2025-06-30 and another due 2025-12-15 each read as met or missed, and the review flags whether a promotion or a salary adjustment, such as a move to $185,000, is recommended, each of which routes to a different downstream process. Free-text blocks for the summary, the development plan, and the employee's feedback hold the narrative that a rating alone does not. Talonic reads the performance review and returns the employee and reviewer, the review type and period, the overall and per-competency ratings, and the goals and development plan as typed fields, with the competency scores kept alongside the overall rating. An annual review dated 2026-01-20, acknowledged by the employee on 2026-01-24, with an overall 4.2 built from four competency scores and a 2027-01-19 next-review date, loads into the HRIS instead of a retyped form, so a people team analyzes ratings from structured data. The extraction captures what the review records and makes no employment decision or recommendation of its own.

What gets extracted from performance reviews

How extraction works for performance reviews

Performance reviews are produced in platforms such as Lattice, Culture Amp, and Workday, and in document templates that vary by team, so the rating scale and the placement of the competency and goal blocks differ from one form to the next. Talonic classifies the review and maps it to the appraisal schema in the Field Registry, which separates the employee and reviewer header, the overall and per-competency ratings, the goals and objectives, and the development plan. Competency ratings are typed as numbers on the form's own scale and kept alongside the overall rating so the roll-up can be reconciled to the average of its parts, goals keep their target date and achievement status, and the promotion and salary-adjustment recommendations are captured as their own flags. Review type, annual, mid-year, probationary, or 360-degree, and the review period and next-review date are parsed to ISO 8601, and the free-text summary, development plan, and employee feedback are kept as labeled text. Every value returns with a confidence score and a pixel-region pointer under DIN SPEC 91491, so a people-operations analyst can verify a rating or a recommendation against the source review before it feeds a talent process. The extraction structures what the review records and does not make or advise on any employment decision.

Sample extraction

An annual performance review with four competency scores

{
  "document_number": "PR-2025-0771",
  "document_date": "2026-01-20",
  "review_period_start": "2025-01-01",
  "review_period_end": "2025-12-31",
  "employee.name": "James Whitaker",
  "employee.id": "EMP-4471",
  "employee.position": "Senior Backend Engineer",
  "employee.department": "Engineering",
  "employer": "Northgate Logistics Inc.",
  "reviewer.name": "Dana Osei",
  "reviewer.position": "VP of Engineering",
  "review_type": "annual",
  "overall_rating": 4.2,
  "performance_rating": 4.2,
  "promotion_recommendation": false,
  "salary_adjustment_recommended": true,
  "next_review_date": "2027-01-19",
  "acknowledgement_signature_date": "2026-01-24",
  "competency_ratings": [
    {
      "competency_name": "Technical execution",
      "rating": 4.5,
      "comments": "Led the storage migration with no production incidents"
    },
    {
      "competency_name": "Ownership",
      "rating": 4.3,
      "comments": "Drove the on-call reliability work end to end"
    },
    {
      "competency_name": "Collaboration",
      "rating": 4,
      "comments": "Effective across teams; mentors two juniors"
    },
    {
      "competency_name": "Communication",
      "rating": 4,
      "comments": "Clear written updates; can lead more in reviews"
    }
  ],
  "goals_and_objectives": [
    {
      "goal_description": "Ship storage migration",
      "target_date": "2025-06-30",
      "achievement_status": "met",
      "achievement_notes": "Completed ahead of schedule"
    },
    {
      "goal_description": "Reduce on-call incidents by 30 percent",
      "target_date": "2025-12-15",
      "achievement_status": "met",
      "achievement_notes": "Down 34 percent year over year"
    }
  ]
}

Frequently asked