KEX — AI Robot Strategy

⚠️ 3 Issues Holding All 3 Robots Back

These affect every bot — fixing them is the single biggest lever.

📝

Script Too Long

Robot talks too much. Customers hang up before responding. Best-performing bots elsewhere use short, direct scripts.

📱

Keypad Input (DTMF)

All 3 bots require pressing phone buttons. Customer pulls phone from ear → confused → no response. #1 cause of wasted calls.

🏷️

"KEX" ≠ "Kerry"

KEX is the new official name after the acquisition — and the company wants to establish it. But Thai customers still only recognize "Kerry." The bot needs to bridge both: say "Kerry Express, now KEX" so customers pick up and start associating the new brand.

📊 Current Performance — COD Bot (Only Active Bot)

115,000+ calls/month — here's what happens:

3 in 10

Confirm

Target: 5 in 10

3.5 in 10

Wasted — no response

Target: <2 in 10

6–7 in 10

Connect

Target: 7.5+

1 round

No follow-up

Best: 4 rounds

Root Cause

Robot is 90% accurate when connected. The problem is setup (wrong data, confusing input) and process (no follow-up strategy).

Case 1: COD Delivery Notification

Confirm customer will receive their COD parcel before delivery

● LIVE — NEEDS IMPROVEMENT

🧩 Two Layers That Determine Robot Success

⚙️

Layer 1 — Bot Configuration

What it says, how customers respond, what data it gets

🎯

Layer 2 — Operational Strategy

When it calls, how often, what happens with results

Fix one → incremental. Fix both → transformational. Below we break down each layer with a 3-phase rollout.

⚙️ Layer 1 — Bot Configuration

What the robot says & what data it gets

❌ Current

Opening

"Parcel from KEX"

Customer doesn't recognize it

Input

Press buttons (DTMF)

Phone away from ear → confused

Order Channel

"KEX"

Wrong — can't personalize

COD Amount

Sometimes blank

Can't verify

Caller ID

Unknown number

Looks like a scam

🔄 Phase 1–2 (Manual Upload)

Opening

"Kerry Express — Shopee parcel, ฿350"

Short, recognizable brand

Input

Voice response ("say yes")

Natural — phone stays on ear

Order Channel

"Shopee" / "TikTok" / "Lazada"

Ops maps data manually

COD Amount

Always filled

Ops enforces mandatory field

Caller ID

Recognized KEX number

Under discussion with telco

✓ Phase 3 (API Integrated)

Opening

"Kerry Express — Shopee parcel, ฿350"

Same script + tracking number

Input

Voice response

+ bilingual Thai/English

Order Channel

Auto-populated from CRM

Zero manual entry

COD Amount

Auto-populated from CRM

Always accurate

Caller ID

Branded KEX number

Integrated with KEX telco

🎯 Layer 2 — Calling Strategy

When & how often — this is where the biggest gains hide

❌ Current

Rounds

1 round — call once, done

Timing

Random

Whenever ops uploads

Redial

5 min apart

Too aggressive

Missed calls

No follow-up

Task creation

Manual every time

🔄 Phase 1–2 (Workflow)

Rounds

Up to 4 rounds/day + retry

Timing

8–10, 12–1, 3–4, 6–8

Confirmed peak windows

Redial

15 min apart

Respectful spacing

Missed calls

Filter → retry next window

Task creation

Workflow template (set once)

✓ Phase 3 (Automated)

Rounds

4 rounds — fully automated

Timing

Same 4 peak windows

System-triggered, zero effort

Redial

Smart spacing per customer

Missed calls

Auto-filtered, auto-retried

Task creation

CRM triggers → auto-call

Multi-Round Calling Flow — 4 confirmed peak windows

8–10 AM

Round 1 — Full List

Morning peak. Before customers leave home.

Filter ↓

No response → Round 2

12–1 PM

Round 2 — Missed Only

Lunch break. People check phones.

Filter ↓

Still nothing → Round 3

3–4 PM

Round 3 — Afternoon

Work break window.

Filter ↓

Final attempt for the day

6–8 PM

Round 4 — Evening

After-work. Best for people busy all day.

Next Day

Retry — Refusals Only

Fresh day, different mindset. 8–10 AM.

📈 Expected Outcome — Both Layers Combined

3 in 10

confirm today

↓

5+ in 10

confirm (target)

3.5 in 10

wasted calls

↓

<2 in 10

wasted (target)

1 round

single attempt

↓

4 rounds

systematic follow-up

Case 2: Pre-Delivery Check

Call before dispatching — confirm customer is available

● READY TO ACTIVATE

📦 Problem & Solution

Without Pre-Check

~2 in 10 deliveries fail first attempt
Wasted fuel, time, truck capacity
Customer not home → return trip

With Pre-Check

Target: <1 in 10 failures
Only dispatch to confirmed addresses
Reschedule before courier leaves

⚙️ Bot Config

Bot

"Kerry Express — your Shopee parcel is scheduled for delivery today. Are you available?"

├─ "Yes" → Dispatch ✅
├─ "Not today" → Reschedule
└─ No answer → Dispatch anyway

🎯 Strategy

6 AM

Upload delivery list

7 AM

Robot calls

Before customers leave home

8:30

Dispatch decisions

Confirmed → go. Rescheduled → reroute.

Activation

Stabilize Case 1 first → Pilot in one delivery zone → Compare failed delivery rate → Expand.

Case 3: Post-Delivery Verification

Call after delivery — confirm parcel was actually received

● READY TO ACTIVATE

🔍 Problem & Solution

Without Verification

Rely on complaints to find issues
Fake proof-of-delivery undetected
Disputes take days to weeks

With Verification

Proactive check within 24 hours
Fraud caught before complaint
Deterrent — couriers know

⚙️ Bot Config

Bot

"Kerry Express — our records show your parcel was delivered yesterday. Did you receive it?"

├─ "Yes" → Verified ✅
├─ "No" → 🔴 Investigate
└─ "Damaged" → Customer service

🎯 Strategy

9 AM

Upload yesterday's deliveries

Focus: COD > ฿1,000

10 AM

Robot calls — verify

1 round, 1 redial at 15 min

11:30

Flag & investigate

"Didn't receive" → immediate action

Activation

After Cases 1 & 2 stable → Start with COD > ฿1,000 (highest risk) → Measure fraud catch rate → Expand.

🗺️ Action Plan — 3 Parts, 3 Months

Organized by who needs to act. Part 1 starts immediately — no waiting.

🔧

Part 1 — Bot Fixes

AI Rudder Alone Weeks 1–3

🎯 Fix all 3 bots so they perform at best practice level

🔧 Shorten all scripts — concise messaging All

🔧 DTMF → Voice response All

🔧 "KEX" → "Kerry Express" All

🔧 "Who is KEX?" handler + Lazada COD

🔧 A/B test: concise vs current script COD

🔧 Add bilingual support All

📋

Part 2 — Ops Transition

AI Rudder Leads Weeks 3–8

🎯 Remove all manual friction — AI Rudder designs & executes, KEX confirms

📋 Fix Order Channel data (Shopee/TikTok/Lazada) KEX Ops

📋 Train team on 4-window calling strategy Joint

📋 Build & standardize Workflow templates — replace all manual Task creation AI Rudder

📋 Activate Pre-Delivery bot — 1 zone pilot Case 2

📋 Activate Follow-Up bot — COD > ฿1K Case 3

📋 Assist daily execution (file upload, monitoring) to reduce KEX friction AI Rudder

⚡

Part 3 — API Integration

AI Rudder + KEX IT Weeks 8–12

🎯 Full automation — zero manual work, system-to-system

⚡ CRM → AICalling API (auto contact push) API

⚡ Webhook result callback API

⚡ Tracking number auto-enrichment API

⚡ Full loop: CRM → call → results → next action API

Key Principle

Part 1: AI Rudder acts alone — no dependency. Part 2: AI Rudder does the thinking and heavy lifting — KEX confirms and we help execute to minimize friction. Part 3: AI Rudder + KEX IT build full automation together.

🚧 Blocker — Order Channel Data

The Order Channel field in contact lists still shows "KEX" instead of the actual platform (Shopee, TikTok, Lazada). This has been unresolved for 3+ weeks across multiple internal KEX teams.

Without this data, the robot can't personalize calls — which directly drives the high "no response" rate.

What We Need

KEX ops to map Order Channel to actual platforms in the contact list upload. AI Rudder will provide the template and assist with the transition.

This is the single biggest unlock for Part 2.

✅ Confirmed Action Items (Carry-Forward)

Confirmed by previous team lead — ready to implement:

✅ AI Rudder Can Act Now

• Clarify "KEX is Kerry Express" in opening

• Add "Who is KEX?" handler

• A/B test concise vs current scripts

• Add Lazada as supported platform

• Peak windows: 8–10, 12–1, 3–4, 6–8

• Bilingual Thai + English

📌 Needs API (Phase 3)

• Tracking number auto-pull from system

Requires KEX IT involvement