Software, security and customer operations / Managed BPO

Dedicated data annotation teams for your labeling platform

Actigy BPO provides dedicated data annotators for recurring work in your labeling platform. You define the labels and acceptance rules. Our team labels the batch, tracks corrections and sends ambiguous items to your dataset owner with the guideline version used.

Agree the scope, access and reviewer before a paid pilot.

Illustrative handoffClient approval retained
A label decision everyone can reproduce
Object
Cyclist partly hidden by a parked car
Option A
Box around the visible area
Option B
Box around the estimated full extent
Decision
Dataset owner confirms the occlusion convention

A worked example of the record and decision flow. Not a client case or promised result.

The work. The handoff. Your decision.

A defined workflow in your systems

For computer vision teams, dataset providers and annotation agencies that need steady labeling capacity in a client-controlled workspace, including projects that restrict where reviewers may work.

  1. You provide

    Approved batch, labeling taxonomy and guideline version.

  2. Actigy BPO prepares

    Labeled units, reviewer checks and an ambiguity log.

  3. You retain

    Label definitions, reference answers and acceptance rules.

Decision owner: Dataset or ML owner. Confirm this role and the escalation contact in the written scope.

Service detail / 01

How does a dedicated annotation team work in a client platform?

Agree the task unit, platform roles and volumes before assigning people. For a visual dataset, define whether reviewers classify an image, place a bounding box, trace a polygon or flag an unusable frame. Assign each operator an individual account and document where project files may be accessed.

  • Train annotators against the current guideline version and a client-approved reference set.
  • Reserve review capacity for ambiguous items and corrections, including cover when a named operator is away.
  • Agree whether Actigy manages daily assignment and review or supplies staff under the buyer’s own supervisor.

Service detail / 02

What happens when two annotators disagree?

A cyclist is partly hidden behind a parked car. One annotator draws a box around the visible part of the cyclist. Another draws the estimated full extent. The reviewer checks the rule for occlusion. If it is silent, the dataset owner chooses the convention. We record that choice, update the examples and identify the batches to recheck.

Service detail / 03

Can we specify EU-only reviewer access?

State the countries allowed for annotators, supervisors and backup reviewers. We confirm the proposed roster and access locations in the written scope. Agree storage location, platform access, support escalation, download permissions and whether any images may be exported. EU hosting on its own does not establish where a person will view the data.

Service detail / 04

What belongs in a paid calibration batch?

Choose a representative batch and reserve a separate reference set for scoring. Include rare labels, partial objects and unusable images. Agree what counts as an accepted unit and measure first-pass acceptance, reviewer disagreement and correction time. Use the results to estimate throughput for the next batch.

Service detail / 05

How do we compare annotation quotes?

For a batch quote, divide the fee by the number of accepted units. For a dedicated team, check reserved hours, reviewer coverage, training and rework. Compare the same task and mix of difficulty. A classification label and a detailed polygon can take very different amounts of time.

Before the first live queue

Bring the inputs that shape your scope

The proposal should identify what can be assigned, who can decide, and how you will accept the work. Start with one representative queue; agree any expansion after reviewing the pilot.

  1. Define the workload

    Modalities, class balance, unit definition and a representative batch.

  2. Agree the controls

    Tool access, work-location rules and adjudication ownership.

  3. Review the right measures

    Accepted units, correction effort and disagreement by error category.

Before you decide

Buyer questions

Clarify the work, access and approval boundaries before assigning a team.

Can we use our own image labeling platform?

Yes. Provide the platform, available task and reviewer roles, output format and access restrictions. Scope the first approved batch in that environment and confirm export responsibilities.

Can you work behind our brand for another dataset owner?

An agency or dataset provider can scope a white-label arrangement. Agree who may access the end client’s platform, which communications use your identity, and who approves feedback and releases data. Verify permissions before live work.

How does this differ from RLHF or general AI outsourcing?

This offer centers on assigned people running repeated annotation batches under your taxonomy. Language model evaluation and preference ranking have different task definitions, reviewer criteria and deliverables.

What should we send for planning?

Describe the modality, annotation task, expected units per month, country limits, platform and guideline maturity. Use counts and a high-level description in the form. Arrange a permitted sample after access and confidentiality terms are agreed.

What happens if the annotation guidelines change during a batch?

Identify the guideline version used for each batch and isolate items affected by the change. Your dataset owner decides the new rule and the examples used to recalibrate reviewers. Agree which completed units need rechecking before applying the revision, so accepted work and additional correction effort remain distinguishable.

Request an annotation team estimate

Tell us the labeling task, monthly units, platform and permitted countries for annotators and reviewers. Arrange sample access after initial scoping.

Your enquiry has been received. We will review the workflow and contact you using the details you provided.
Share process details and estimated volumes. Do not include customer records, patient information or confidential datasets.
Add volume, coverage and timing (optional)

Your details are used to respond to this inquiry. Read the privacy policy.

This form is for companies seeking process support. Job applications sent through it are not reviewed.

For attribution, this browser tab stores your first landing path, the referring site's origin, standard campaign tags and whether you arrived from an AI assistant in session storage, then sends those details with your inquiry.

Your inquiry

Dedicated data annotation teams

Describe the workload and the decision owner. Keep confidential records out of the form; agree a suitable channel for any later sample.

  1. Review the fitWe review the workflow and the inputs you provide.
  2. Define the scopeDiscuss access, coverage, review and the proposed fee.
  3. Agree a paid pilotApprove the written scope before work begins.
Published · Updated