Get Your First Client

Your Commercial Offer

Title: Consistent, Documented Data Annotation

Target Buyer: ML engineers, data teams, and annotation project managers who need reliable human labellers with low disagreement rates and clear escalation discipline.

Problem: Inconsistent or rushed annotation introduces label noise that degrades model performance and requires expensive re-labelling.

Deliverable: A verified, formatted annotation batch with a quality summary, disagreement log, and escalation notes delivered in the client's required format.

Scope Boundaries: We apply the client's label taxonomy as specified; we do NOT modify or reinterpret the taxonomy without written instruction., We do NOT guarantee specific throughput numbers before measuring actual pace on the client's task type., We do NOT guarantee task availability on third-party platforms—that depends on their project pipelines., Privacy-sensitive data is handled per the client's data-handling instructions; we do NOT share, copy, or retain data beyond the engagement.

Illustrative Pricing (India)

  • Currency: INR
  • Range: 0.5 - 5
  • Per Unit: per labelled item (illustrative range; highly variable by task complexity and platform or contract type)
  • Illustrative

Illustrative Pricing (US)

  • Currency: USD
  • Range: 0.02 - 0.3
  • Per Unit: per labelled item (illustrative range; highly variable by task complexity and platform or contract type)
  • Illustrative

Outreach Strategy

Target: ML engineers, data team leads, or startup founders building AI products who post about training data challenges on LinkedIn or in ML-focused communities.

Channel: LinkedIn message or direct application through annotation platform qualification portals.

Follow-Up Limit: 1

Sequence:
  1. Identify a team that mentions training data needs publicly.
  2. Apply directly with a clear statement of task types you have experience with and your quality discipline.
  3. Request a qualification batch to demonstrate labelling consistency before discussing volume.
  4. Follow up once if no response.

Outreach Template

Hi [Name], I work on data annotation and labelling for machine learning teams. I am familiar with [TASK TYPE—e.g., text classification, image bounding boxes, NER] and prioritise low disagreement rates and thorough edge-case logging. If your team runs annotation projects and needs reliable reviewers, I would welcome a qualification test to demonstrate quality on your specific taxonomy. What does your current project pipeline look like?

Objections & Replies

  • Objection: We use automated labelling tools for most of this.

    Response: Automated labelling is efficient for clear-cut cases. The value of human annotation is concentrated in the borderline cases that automated tools resolve incorrectly at high rates. If your team reviews auto-labelled data, I can contribute to that QA layer specifically.

  • Objection: How do we know your labels will be consistent?

    Response: Consistency is measurable—which is why I ask for a qualification batch before discussing volume. You can calculate my inter-annotator agreement against your team's benchmark labels on a small test set. I also maintain a disagreement log so borderline decisions are documented, not guessed.

  • Objection: Can you handle [specific domain—medical, legal] annotation?

    Response: Domain-specific annotation requires genuine expertise in that field—it is not something I would take on without the relevant background. If [domain] is outside my expertise, I would rather be upfront about that than deliver unreliable labels for high-stakes data.

Turnaround Time

Agree on a specific batch size and delivery date before starting. Throughput depends on task complexity—text classification batches of 100 items and image segmentation batches of 100 items are not comparable in time. Measure your actual pace on a small pilot batch first.

Tracker Guidance: Track each client or platform by task type, label taxonomy, batch size, and your measured disagreement rate. Record escalation frequency—how often you log borderline cases versus guessing. Use these numbers to calibrate what tasks you can handle reliably.

Delivery & Fulfillment Mechanics

  1. Guideline Review: Receive and read the complete annotation guideline document before starting. Use Prompt 1 to confirm understanding.
  2. Qualification: Complete any required qualification or test batch. Do not begin the main dataset until qualification is passed.
  3. Data-Handling Agreement: Confirm the client's data privacy requirements. Agree on how data is accessed, stored (if at all), and deleted after the engagement.
  4. Pilot Batch: Label a small pilot set (10–20 items) and submit for inter-annotator agreement check before proceeding to the full batch.
  5. Full Batch Labelling: Work through items at a pace that maintains accuracy. Maintain the disagreement log for borderline cases.
  6. Self-Consistency Check: Spot-check 10% of completed labels before delivery. Note any cases where your label would differ from your first pass.
  7. Quality Summary: Prepare a delivery note with: total items labelled, disagreement log, escalated items, and any guideline ambiguities observed.
  8. Delivery: Submit in the client's required format (CSV, platform upload, JSON). Confirm receipt.
Supplementary Methodology & Evidence
Commercial & Acquisition: Platform-based annotation provides immediate access to tasks without client outreach—suitable for building a quality track record before approaching direct clients. Direct contracting typically requires demonstrating quality metrics first, so platform experience provides the evidence base.