WhichAI research library
Inspect the evidence before you trust the recommendation.
This library publishes the artifact, method, limitations, and primary sources behind practical AI workflow decisions. The goal is a conclusion another operator can inspect, challenge, and adapt to local conditions.
Evidence boundary
A template is a template. A synthetic fixture is synthetic. An observed output proves only what was observed.
No artifact on this page becomes a customer result, testimonial, or guaranteed outcome by implication.
One inspectable artifact
A template, dataset, frozen output, protocol, or other evidence asset is available from every published research page.
Method before conclusion
The page states how the evidence was produced, what was checked, and which conclusion the evidence can support.
Limitations stay visible
Synthetic fixtures, observed product output, and unrun protocols are labeled. They are not presented as customer results.
Standalone research
Research built around a decision and a public artifact.
Each publication answers one operating question, documents how the work was evaluated, and exposes what the evidence cannot establish.
AI Workflow Process Map Template
Map a current workflow before proposing AI. Document the trigger, handoffs, evidence, exceptions, human decisions, and excluded actions with a blank template and synthetic example.
Artifact: Current-state AI workflow process-map pack
Inspect the researchAI Workflow Readiness Assessment Template
Use a free evidence-led template to triage one recurring workflow, expose hard stops, and decide whether contained AI test planning is justified.
Artifact: AI workflow readiness operational triage pack
Inspect the researchEvidence-backed field guides
Start with the artifacts already available.
These implementation guides include downloadable evidence packs with a documented method and explicit limitations.
Evidence v1 · 2026-07-14
Should I Hire or Automate This Operations Role?
An editable matrix separates recurring preparation, consequential judgment, exceptions, service demand, and management work before a staffing decision is made.
2 inspectable assets
Open the evidenceEvidence v1 · 2026-07-14
AI Does Not Save Time Until Work Disappears
The ledger records whether a task disappeared, moved, was duplicated, or returned as review, correction, exception, maintenance, or downstream cleanup.
2 inspectable assets
Open the evidenceEvidence v1 · 2026-07-14
What a Former WhichAI Intake Prototype Returned for One Frozen Task
This pack preserves the deterministic intake prototype that WhichAI previously showed for one frozen reporting task. It is historical evidence, not a current visitor entitlement or the paid researched workflow report.
5 inspectable assets
Open the evidenceEvidence v1 · 2026-07-14
How to Measure Time Saved by an AI Workflow
An editable case-level worksheet, a deliberately small synthetic example, and calculation notes that count review, correction, failure recovery, maintenance, and downstream cleanup instead of reporting model response time as time saved.
3 inspectable assets
Open the evidenceEvidence v1 · 2026-07-15
Proof-of-Delivery Automation: Turn Delivered Loads Into Billing-Ready Files
A public protocol, seven structured synthetic document cases, a machine-readable case matrix, and expected candidate records for testing normal, missing, mismatched, unreadable, duplicate, and field-exception paths without authorizing an invoice.
4 inspectable assets
Open the evidenceHow the library is governed
Read the evidence classes and correction policy.
The public methodology explains source selection, evidence labels, claim boundaries, review status, corrections, and what is excluded from publication.
Read the methodologyUse the evidence
Move from research to your own workflow.
Use self-serve when the task is already concrete. Use WhichAI Solutions when the problem spans recurring work, multiple systems, sensitive controls, or implementation complexity.