Artifact · v0 (format-only stub)

Workflow Map of Labeling Pipeline

Step-by-step documentation of how SmartOne's annotation pipeline actually runs. What is rule-bound, what requires human judgment, where quality control sits, and which steps are candidates for AI augmentation.

2.2 · Workforce and Augmentation · artifact id: workflow-map-v0.html · 2026-05-28 · v0 format stub

This artifact is a format placeholder. The Phase 2 engagement builds this map through direct workflow observation and interviews with Jeritiana Ravelojaona and the Madagascar ops team. No outsider can build this accurately without your inputs. The structure below shows what the live workflow map will look like.

Why this map matters

On the May 15 call, Amyn asked: "could you improve them if you increase the efficiency of your people? And if you do win that big contract, do you need to go out and hire three hundred more, or do you just need to hire a hundred more and you pocket a bunch?"

That question cannot be answered without knowing which steps in the pipeline are capacity-constrained and which are already efficient. This map is the diagnostic that answers it.

Map structure (when built)

Pipeline step table

Step	Description	Rule-bound or judgment?	Avg time per unit	Current bottleneck?	AI augmentation candidate?
1. Intake and job setup	[Data received, format validated, task spec loaded]	Rule-bound	[X min]	[Yes / No]	[Yes / No / Partial]
2. Initial labeling	[First-pass annotation by annotator]	Mixed	[X min]	[Yes / No]	[Partial: pre-label assist]
3. Quality review	[QA reviewer checks against spec]	Judgment-heavy	[X min]	[Yes / No]	[Limited: edge case judgment required]
4. Reprocessing	[Labels below threshold recycled to annotator]	Rule-bound	[X min]	[No / Sometimes]	[Yes: routing automation]
5. Delivery	[Final package formatted per client spec and delivered]	Rule-bound	[X min]	[No]	[Yes: formatting automation]

Augmentation opportunity summary (output of map analysis)

Step	Current headcount at that step	With AI assist (projected)	Headcount delta	Quality risk
[Step N]	[N people]	[N people]	[+N or -N]	[Low / Medium / High]

What you need to send us to build this

Inputs needed. To build the live workflow map: (1) a 60-minute call with Jeritiana to walk through the actual pipeline step by step, (2) the current time-per-unit data for the highest-volume work type, (3) a description of how quality thresholds are set (what triggers reprocessing), and (4) the tool stack currently in use at each step (the 10-20 open-source tools referenced on the call).