Most annotation vendors make you sit through a discovery call to learn how they're even structured. Here's the whole thing up front — the three ways we engage, what each one includes, how fast each moves, and how we scope the work.
Three ways in
Most teams start with a pilot, run a sprint against their next training cycle, and move to an embedded team once the pipeline is continuous. You're not locked into a path.
You're evaluating vendors
You have a training cycle to hit
Annotation is continuous now
Scoping
We use one of three structures depending on how stable your guidelines are. We'll tell you which one fits after seeing a sample — and we'll tell you when a cheaper structure is the right one.
For well-defined, repeatable tasks where the spec is settled and the edge cases are known. Predictable for you, but only honest once guidelines have stopped moving.
For work where guidelines are still evolving or throughput needs to flex with your training schedule. You're buying a team's capacity rather than a unit count.
For fixed-scope work with a defined deliverable — a gold-set build, a quality audit of an existing vendor's output, or a one-time backlog clear.
Start to production
You send data and whatever spec exists. No procurement process required to get this far, and no NDA needed unless you want one — we'll sign yours.
Two annotators label independently, QA scores agreement, and we send back the output plus a written note on where your guidelines are ambiguous. This is the part most vendors skip.
With the pilot in hand we can size the real thing accurately — team size, throughput, accuracy targets, and cadence. One agreement, no procurement theatre.
Annotators are recruited and domain-trained against your guidelines rather than pulled from a general labeling pool. That's why we can hold quality at week one.
Delivery on your cadence, IAA reported every batch, and guideline changes absorbed as your model and edge cases evolve.
Common questions
Because a per-item number without knowing your annotation density, class count, and QA requirement would be a made-up figure, and you'd rightly discount it. What we can commit to publicly: the pilot is free, sprint work is monthly with no long-term contract, and you'll have a real number within a day of us seeing a sample.
Sprint engagements run in one-month increments and can be paused between training cycles. Pilots require nothing at all.
Yours, in most cases — CVAT, Labelbox, Scale, SuperAnnotate, or an internal tool. If you don't have one, we'll bring ours and hand over the data in whatever format your pipeline expects.
Accuracy SLAs are written into sprint and embedded agreements with defined remediation — re-annotation at our cost, not yours. IAA is reported every batch so you see it before we do.
NDAs, confidentiality terms per engagement, and deletion on request. Details are on our security page.
Different structure entirely — see technical staffing, healthcare, or eCommerce operations.
500 frames, five days, an IAA report, and no contract. You'll know more about how we work than any call could tell you.