Pod AI Workbench

Context

AfterShip's multi-product merchants — running Tracking, Returns, Email, Affiliates, and Personalization simultaneously — face a compounding attention problem. Every product surface shouts for attention. No surface tells them what to prioritise. No surface connects the signals across products into a decision.

Pod is the answer to that gap: an AI workbench that watches your business across all AfterShip products and brings you the decisions worth making today — with evidence, context, and a recommended action already prepared.

Positioning

Organised around decisions, not conversations.

Every major productivity tool organises work around a different object. Slack organises around conversations. Gmail around messages. Linear around issues. ChatGPT around prompts. Pod organises work around AI-detected business decisions — the ones that already have evidence, a recommendation, and a clear action attached.

This single positioning choice determines all subsequent design decisions: Pod doesn't do chat, doesn't do inbox, doesn't do issue tracking. It does "what the AI has already assessed and determined is worth your decision today."

Product form

Three core pages and a settings modal.

Pod has four product surfaces, deliberately separated by usage frequency. The hierarchy of the sidebar reflects that hierarchy directly — Today's Actions always sits first.

Settings becoming a modal — not a fourth sidebar tab — was a deliberate decision. When it was a tab, its visual weight implied parity with the three core surfaces. Moving it to a modal makes clear that it's configuration, not navigation. Closing it returns the merchant to exactly where they were.

Usage rhythm

Designed around how merchants actually use it.

The entry hierarchy in the sidebar mirrors the intended usage cadence. This isn't decorative — it's the core operating model of the product. Each surface is visited at a different frequency, and the visual weight of each should match that frequency.

Interaction model

What a complete decision looks like end to end.

Each decision in Today's Actions is a full package: signal → reasoning → recommendation → evidence → action. The merchant never has to leave the surface to understand why something is happening. Below is a complete example from detection to execution.

Inventory Guardian Agent checks every 6 hours
→ LED Lamp Pro: 8 units remaining, +18% sales acceleration, 10-day supplier lead time
→ Calculates: stockout in ~3 days · ~$3,400 GMV at risk
→ Triggers a Signal

Conv list: "LED Lamp Pro may stock out in 3 days"  [Needs approval]
→ Detail: POD RECOMMENDS — "Restock 500 units from Victor Lee…"
→ Expected impact: "Prevent stockout and protect ~$3,400 GMV"
→ CTA: Contact supplier  (one action, one click)

If merchant opens View Data Drawer:
→ 3 signal chips: 3 days · +18% · 10 days
→ 4 data source chips: Shopify · TTS · Sales · Action history
→ Confidence: High · 14 / 14 days of data · sparkline
→ Detected by: Inventory Guardian · Recent runs timeline

Key design decisions

Six trade-offs, surfaced deliberately.

Design reviews tend to defend decisions after the fact. This section does the opposite — it names the trade-offs upfront so reviewers can challenge them on their actual merits.

Trust architecture

From observer to collaborator — four stages.

The core design challenge isn't UI — it's trust. A merchant won't let an AI act on their behalf until they believe it won't overstep. The product needs to earn that trust progressively, not demand it upfront.

Design validation

Five questions a merchant asks in two seconds.

Any Recommendation Card in Today's Actions must pass this test before it ships. If any question goes unanswered by the UI, the design is incomplete.

Any of these questions going unanswered is a design failure, not a content gap.

Phased roadmap

Four phases from alpha to GA.

Three categories are permanently out of scope: Workflow Builder (contradicts Pod's AI-first positioning), AI Marketplace (outside AfterShip's data security boundary), and general-purpose chat (collapses the decision-inbox mental model).

Design principles

Six lines that hold all design decisions accountable.