Graph consistency tests
Check whether public outputs and sources agree with approved client facts, policies, offers, and actions.
AI answer and path checks
Catch wrong facts and broken paths before AI repeats them.
Answer QA tests whether AI systems, public sources, endpoints, and customer paths match the approved client Commerce Graph.
5
Test surfaces
AI answers, sources, endpoints, MCP, actions
Tracked
Defect states
Open, in review, fixed, retested
Attached
Evidence
Each defect can carry proof and a recommended fix
Endpoint validation
MCP validation
Before and after evidence
Platform layer
What Akii powers
Each part of Akii helps the business become easier for AI chatbots and agents to understand, compare, and route to the right customer action.
Defect workflow
Classify wrong facts, broken action paths, source conflicts, endpoint issues, and missing fields by severity.
Retesting
Attach evidence, create owned fixes, publish graph updates, and retest after changes.
Multi-surface evidence
Store proof from AI answers, source pages, endpoint output, MCP responses, screenshots, and action-path checks.
Workflow
How QA closes the loop
The workflow turns scattered business information into approved data, customer paths, checks, and fixes that can be repeated as the business changes.
1
Test runRun tests
Compare approved graph data against public sources, endpoint responses, MCP output, and AI-surface answers.
2
Evidence setCapture evidence
Attach source snippets, endpoint payloads, screenshots, response text, and path-check output.
3
Defect tableCreate defects
Classify severity, confidence, owner, affected entity, recommended fix, and retest criteria.
4
Retest resultRetest after fix
Publish graph or action updates and rerun the failing check to prove the issue changed.
How it works
Defect taxonomy
The important details are structured so businesses, agencies, and AI systems can rely on the same approved version of what is true.
| Area | Input | Output |
|---|---|---|
| Wrong fact | AI or source says something that conflicts with the approved graph | Critical or warning defect with evidence |
| Missing data | Required graph, readiness, policy, or action field is absent | Fix task tied to entity and owner |
| Broken action path | Booking, form, phone, WhatsApp, or provider route fails a safe test | Action Registry defect and retest |
| Endpoint issue | Schema mismatch, stale graph version, unavailable endpoint, MCP mismatch | Access fix and monitoring event |
Use cases
Where it is used
Pre-publish validation
Check a graph and endpoint set before partners, agents, or customers rely on it.
Ongoing monitoring
Catch drift when websites, menus, listings, prices, policies, or action destinations change.
Operational accountability
Turn vague AI-answer issues into owned fixes with evidence and retest results.
Common questions
Short answers for agencies and partners evaluating how Akii helps service-business clients become easier for AI systems to understand.
Does Answer QA test AI responses?
The dashboard is structured for multi-surface QA runs and evidence capture across agent answers, sources, endpoints, MCP, and action paths.
What makes a defect critical?
Critical defects usually affect customer action, payment or deposit clarity, safety-sensitive policy, endpoint availability, or materially wrong client facts.
Can a fix be retested?
Yes. Fixes are designed to carry evidence, owner, recommended action, and retest status after graph or action updates.
Create your agency account, then publish the client AI commerce layer.
Start with one client. Structure what the business sells, publish approved AI-readable outputs, and test whether agents can understand and route the right next step.