Review System Rebuild

Unblocked · start now

No external dependency

No external dependency. ~4-6 weeks of real value while the API ticket sits in Jeremy's queue.

A · Email deliverability rehab 0/6

A1DNS authentication audit (SPF, DKIM, DMARC) for both sending domains

Pull current state from Google Postmaster Tools.

DNS

A2Decide which sender to keep — ew-emails@ vs service@

Pick the address with better Postmaster reputation.

Decision

A3Set up bounce processing pipeline (Gmail FBL, Yahoo CFL, Microsoft JMRP)

Eng

A4Implement RFC 8058 one-click List-Unsubscribe-Post header

Eng

A5Engaged-only warmup plan (last-90-day openers only for 2 weeks)

Plan

A6Standardise sender — retire the unused address with proper warming

3-week +10%/day ramp on the kept sender.

Eng

B · Email content design 0/7

B1Audit current Day 3 + Day 12 templates from EW-emails inbox

Audit

B2Audit the supplier-self-send template (Leanne's mention)

Used when vendors don't want EW chasing their brides directly.

Audit

B3Rewrite Day 3 subject line

Kill "Congratulations on your wedding, {firstname}" → question form.

Copy

B4Design Day 6-7 nudge email (different angle, same agent link)

Copy

B5Move Day 12 send time off midnight → Tue-Thu 5-7pm AEST

HubSpot

B6Mobile-first responsive design pass on all three templates

Design

B7Multiplicative incentive copy ("5 reviews = 5 tickets, 10 = 15…")

Copy

C · AI agent design 0/9

C1Lock the conversational state machine as JSON spec

Turn-by-turn, inputs/outputs per state.

Spec

C2Write the Claude system prompt for the agent

Formats her words only — never generative.

Prompt

C3Write the validation gate (substring/paraphrase check)

Logic

C4Define the halt list — phrases/signals that route to human review

Death, divorce, cancelled wedding, supplier complaint, distress.

Spec

C5Curate tag library per supplier category (8-12 tags each)

Mix of trait words + outcome words per category.

Content

C6Negative review path workflow — Slack escalation channel + SLA

Ops

C7Confirmation-turn UX (agent shows assembled review + edit/post)

C8Multi-supplier session flow + "Done" exit

C9Tokenised URL spec — signed JWT + 60-day TTL + key rotation

Spec

D · Cloudflare Worker scaffold 0/6

D1Create new Worker repo / folder under EW's CF account

Eng

D2Set up env management (Twilio creds, Claude API key, EW admin key, JWT secret)

Eng

D3Stub the API client — implement contract, swap stubs for real calls when Jeremy ships

Eng

D4Mock data layer for local dev (build + test agent without admin API)

Eng

D5Inbound webhook handler for Twilio (STOP keyword + opt-out)

Eng

D6Hold-out cohort assignment logic (deterministic 50/50 by bride_id hash)

Eng

E · Stakeholder + comms 0/6

E1Send the CEO update DM (Leanne meeting outputs)

E2Send Ian the Twilio Alpha Sender ID Slack message (drafted)

E3Send Vitali the elasticity SQL request

Does supplier review-count past 30/40/50 actually move bookings?

E4Send Vitali the Trigger 3 sizing SQL request

How many brides qualify for the registered-bride channel?

E5Send Matt or Ian the Spam Act consent question

E6Set up weekly check-in cadence with Leanne

Calendar

F · Phase 2 prep (low priority, parallel-safe) 0/2

F1Sketch claim-your-storefront flow (free vs paid tier)

Sketch

F2Identify the upgrade-trigger logic (locked-but-visible enquiry queue)

Sketch

Blocked · waiting on external

Build can continue against stubs

Live operation gated on these. Worker scaffold + stubbed clients (D3) lets us build the rest in parallel.

G · Blocked on Jeremy's API access 0/5

G1Live Worker integration with admin endpoints (1, 2, 3, 4, 5)

Blocked Jeremy's dev queue

Eng

G2End-to-end WedXM rail test

Depends on endpoints 1 + 2 + agent

Eng

G3Trigger 3 channel build

Depends on endpoint 5

Eng

G4Review submission via API

Depends on endpoint 4

Eng

G5WedXM mark-contacted writes

Depends on endpoint 2

Eng

H · Blocked on Twilio Alpha Sender ID 0/2

H1Live SMS sends (Day 10-12 nudge + WedXM rail)

Waiting 3-8 wk AU carrier approval cycle · Ian to submit

SMS

H2End-to-end test with branded sender ID

Workaround long-code fallback for internal pilot

SMS

I · Blocked on Spam Act consent confirmation 0/1

I1Public SMS launch (any cohort)

Waiting Matt or Ian on registration ToS coverage. If not covered → SMS opt-in step adds 2-4 weeks.

Compliance

J · Blocked on Trigger 3 sizing query 0/1

J1Trigger 3 build commitment

Decision gate 500/yr threshold (minor) vs 2K+/yr (major channel)

Decision

Decision points · need a human call

Block downstream until decided

Items where I have a recommendation but need explicit sign-off before proceeding.

AI-generation strictness line

Strict (her words only, validation gate rejects + retries) vs lenient (more polished prose, drift risk).

→ Recommend strict. Validation gate is non-negotiable architecture.

Alec

Negative review path

Auto-publish ≤2 stars vs queue-for-human-review vs delay-and-re-ask.

→ Recommend human review queue for ≤2 stars. Brand/Marketing to confirm.

Brand

Sales team capacity for Phase 2

Before Phase 2 (claim-your-storefront) ships, confirm sales has bandwidth to receive interested-leads from non-advertisers.

→ Gate Phase 2 on this answer. Phase 1 not affected.

Sales

If nothing else unblocks

Even with NO API access, NO Twilio Alpha Sender, NO Spam Act resolution — these three ship standalone:

Email deliverability rehab (Section A) — lifts open rate ~30-50% on its own. ~10K → ~13K reviews/yr.
Static email template fix (Section B) — kill bad subject lines, move send times, multiplicative incentive copy. Layered on (A).
Agent design + Worker scaffold (Sections C + D) — stays in dev until APIs ship. ~2-3 weeks regardless. Unlocks fast launch when Jeremy is ready.

That alone gets us most of the Phase 1 short-term lift. The full system unlocks when API access lands.