Extending OX Group's AI leadership to the front line: 24/7 automated service for every region.
Your internal productivity stack is already running. This proposal covers the other half — the customer-facing layer. A Voice AI Receptionist and a Dealer Portal Chat Widget, built once for ANZ and scaled across every region.
OX Group — a global tool manufacturer already shipping AI in production
OX Group operates across six regions globally, with a centralised global purchasing function and a Philippines support centre. The Australian customer service team, led by Liza Moren, is the proving ground for global customer-service modernisation. This proposal extends an existing AI investment programme that already has live production components.
“We are very passionate about adopting AI in the business. Everyone should be, and we definitely are.”
Ben Truswell · Head of Global Systems & Business Processes
Already investing heavily in AI
This proposal extends an existing AI investment programme, not a pioneering experiment. Your team is already shipping live AI capability across these initiatives:
Oxley — Internal Teams Chatbot
In build on Microsoft AI Foundry — internal productivity layer across NetSuite and freight data.
NetSuite AI Order Scanning
Live in production — dealer PDF orders auto-scanned and populated into NetSuite.
Power BI + Microsoft SQL Data Warehouse
Celigo-fed warehouse powering business intelligence and the Oxley brain.
Microsoft Copilot Premium
Deployed to sales managers and CS leadership for drafting and presentation generation.
Where customer service time is actually going
The top five call drivers cover roughly 73% of inbound customer service volume. Each one is structured, repetitive, and well-suited to an AI agent that reads from a clean, read-only data layer.
Call drivers by volume
Captured by agent, warm-transferred to a rep — agent never quotes price.
ETA read-back from Celigo-fed order data.
Receptionist pattern with door-keeper logic.
Mostly end-users — guide to stockist or dealer portal.
Inbound shipment date lookup against product code.
Five drivers, 73.4% of all inbound calls. Building the agent against these covers the bulk of customer service workload while staying within a tightly scoped data slice.
Operational context
ANZ customer service is a six-person team; three of them take inbound calls. After-hours coverage is currently zero — calls outside business hours bounce off the IVR.
WA customers operate two to three hours behind eastern Australia, meaning the last hours of the trading day in WA may be uncovered by AEST-based staff.
All product, pricing, inventory, customer and order data lives in NetSuite. Celigo is already in place and licensed for data movement.
Existing IVR uses a press-1-press-2 menu structure. Caller experience is functional but unintelligent — no intent detection, no self-serve.
From a messy “today” pile to a clean “tomorrow” split
Intent
The agent cannot make up prices, ETAs, or shipment dates. It reads from a read-only data slice in Airtable — populated by Celigo from NetSuite with only the three to five columns each intent needs. If the data isn't in that slice, the agent says so and warm-transfers. No hallucinated prices, no leaked GP, no competitor exploits.
Where customer service is today, and where it can be
Current state
100% of calls touched by a human
Every caller waits in the IVR menu, navigates to the right team, and consumes a CSR's attention even for low-value tasks like an ETA lookup.
AEST business hours only
Calls outside the operational window go to voicemail or bounce. After-hours demand is invisible to the team.
Known call drivers, no leverage
Liza already pulls clean call-driver data from 3CX. The picture is sharp. The constraint is that knowing the drivers doesn't shrink the time spent on them — every ETA, invoice and back-order call still routes through a human.
Capacity scales with headcount
Growth in any region requires hiring an additional CSR in that market.
Caller experience frustrating and slow with IVR
Press-1-press-2 menu navigation costs every caller time before they reach a human. Tradies hate it, end users abandon, and dealers ring back through the front desk.
Future state
30% of calls fully self-served
AI handles the simple ETA, copy-of-invoice, and back-order calls end-to-end, with verification and email confirmation.
24/7 inbound coverage
After-hours calls are captured, qualified, and either resolved or queued for follow-up. Zero calls bounce.
Every call transcribed and scored
Retell Assure monitors 100% of calls for hallucinations, latency, and resolution quality. Monthly dashboards for Liza.
Capacity scales independently of headcount
Adding a new region adds a Retell minute allocation, not a CSR hire.
Consistent caller experience globally
The same AU female voice, the same handling logic, the same verification flow across all six regions.
Scale without hiring
If OX grows 30% across the next two years, the current operating model requires hiring an additional customer service rep in each region carrying that load. The proposed model absorbs that growth on the same retainer. The agent is a permanent capacity multiplier — once built, every additional dollar of revenue adds zero customer service cost until volume genuinely exceeds the platform's bandwidth.
Two projects, one shared data layer
Each project can be approved standalone or together. They share the same data architecture, so deploying both does not double the integration work.
Voice AI Receptionist
Replaces the current 3CX IVR with a conversational AU female voice that detects intent, self-serves the contained drivers, and warm-transfers everything else with full context. Voice and vocabulary localise per region — Australian for ANZ, mid-Atlantic English for UK/Europe, North American for US/Canada, with regional terminology baked in (e.g. "tracking number" for AU/US vs "waybill" for UK trade).
The agent answers every inbound call on the existing 3CX line via a Twilio SIP trunk. It uses a soft, friendly Australian female voice with personality. It identifies the caller's intent in the first 10 seconds, captures verification details (order number plus name or email), reads from a Celigo-fed Airtable data slice with only the columns required for each intent, and routes appropriately. Customer service calls warm-transfer to Liza's team. Finance routes to the accounts extension. Sales rep enquiries go through a door-keeper gate before any mobile transfer. IT and supplier enquiries get diverted to email. Cold callers are politely re-routed to email, never hung up on.
Intent flows
Dealer Portal Chat Widget with Teams Escalation
An embedded chat agent on the new dealer e-commerce portal. Handles common queries from logged-in dealers, escalates complex cases directly into Microsoft Teams.
An embeddable chat widget sits on the new dealer portal (logged-in dealers only — no public access). The agent reuses the same Celigo-fed Airtable data layer as the voice receptionist, so there is no duplicate integration cost. Common dealer queries get handled in-widget. When escalation is required, the agent fires a webhook through Zapier that posts a card into Liza's Teams channel containing the full chat transcript. A CSR replies inside Teams, and the message relays back into the widget — the dealer's experience stays in the portal, but Liza's team never leaves the tool they already live in.
The dealer never leaves the chat widget. Liza's CSR never leaves Teams. The handover is invisible from both sides.
Intent flows
Sample call recordings
Short demo recordings of the agent handling real-shaped scenarios with fake customer data, grouped by the Australian and New Zealand lines. Built specifically so the management team can hear exactly what this sounds like before approving the build.
Australian calls
Catching a wrong email address
The caller gives an email address with the “.au” left off the end. Layla reads it back, picks up that it’s incomplete, and confirms the correct address before anything is sent.
CEO impersonation attempt
A caller claims to be a senior executive and pushes hard to get information released. Layla stays polite, holds the verification line, and doesn’t cave to the pressure.
New Zealand calls
Invoice request & order ETA
A dealer calls to request a copy of an invoice and check where an order is up to. Layla captures the request, looks the order up against the live data layer, and reads back the status and ETA.
Name already known — collection skipped
The caller’s name is already known at the greeting, so Layla recognises it and skips straight past the name-collection step instead of asking for details it already has.
Investment
Two projects priced as separate engagements so the management team can approve one, both, or sequence them. All figures ex-GST. OX retains full ownership of every platform.
Voice AI Receptionist
Dealer Portal Chat Widget
If your new dealer portal is a single multi-region site, this $9,000 covers the global deployment. If each region has its own portal instance, expect a per-region duplication fee of ~$2,000 to install and localise on each. Confirm at kickoff.
OX Group holds the Retell, Twilio, Airtable and Modal accounts directly and pays each platform at its standard published rate. NxtLayr never sits between you and those bills — no markup, no resale. OX owns the accounts, the rates, the data and the residency. For reference, expected platform cost at AU volume is around $600-$900 per month across both projects.
2 weeks of post-launch support are baked into the build fee. Optional retainer kicks in from week 3 onwards. Liza's team validates the agent inside the workflow before any recurring fee starts.
Capacity for Growth
The objective Ben set is to take the routine, repetitive calls off the customer service team — not to replace it. So the framing here isn't headcount savings. It's capacity returned. A conservative 30% of inbound calls are pure data lookups the agent handles end to end, freeing the team for the work that needs a person and absorbing growth without new hires. Every figure below is built bottom-up from OX's own call-driver report and Australian salary data, and stated conservatively. The optional retainer is excluded from ROI 01-03 and shown separately in ROI 04.
Decouple headcount from revenue.
Instead of hiring to keep up with volume, OX Group can decouple headcount from revenue. 30% deflection means your next three years of growth don't require a 30% increase in CSR payroll. The agent absorbs the load.
Every call answered at once.
A three-person team can take three calls at once — a fourth caller waits. The agent answers every call the moment it lands, so the Monday-morning rush is handled the same as a quiet afternoon. No hold queue, no busy signal, no callers lost at peak.
AU pilot — customer service capacity returned
Dealer portal chat standalone
The dealer-portal chat widget reuses the voice agent's Celigo-fed data layer and the same deflection logic, so there is no duplicate integration cost. OX has not run a portal chat bot before, so there is no query-volume baseline — a precise ROI is sized once the portal has a quarter of live data. Illustratively, every 4 hours per week of dealer queries it absorbs is worth roughly $8,400 a year of CSR capacity at the same $84,000 loaded cost. The firmer case for this build is strategic: it deflects dealer queries on the channel dealers prefer, and supports adoption of the new portal launching this year.
Five-region voice at scale
The AU pilot proves the unit economics. Each further region is a $3,500 configuration that returns capacity on the same basis. Five regions at the AU baseline is roughly $378,000 a year of customer service capacity against a one-time $45,000 build. We treat this as a conservative floor: ANZ is the smallest of OX's major regions, so the US and UK should carry higher volume. Each regional figure should be validated against that region's own call data before its rollout gate.
Economics with the optional retainer
The retainer is optional and excluded from ROI 01-03. It is not a maintenance fee for the build — it is what keeps the 30% deflection rate from drifting as call patterns, products and platforms change: continuous monitoring, monthly tuning, edge-case fixes, and handling Retell, Twilio and model updates. The figures below show the picture with it taken.
Method · Deflection (30%) is built bottom-up from OX's own call-driver report — order-status and backorder lookups, which together are 31.2% of inbound volume. The loaded AU customer service cost ($84,000) is a base salary of ~$60-65k (SEEK, PayScale) plus ~40% standard on-costs: superannuation, payroll tax, workers compensation and leave. ROI 01 and 03 exclude the optional retainer; ROI 04 folds it in. Five-region figures assume each region replicates the AU baseline — ANZ is the smallest of OX's major regions, so the total is a conservative floor. Validate each region against its own call data at the rollout gate.
Built by someone who's spent a decade selling to businesses like yours.
Chris Gulotta
Founder & CEO, NxtLayr AI
I've spent over a decade working with businesses — across sales, operations, and technology — and the one thing that's always been true is this: the businesses that move first win.
Before founding NxtLayr AI, I spent six years in technology — e-commerce and cybersecurity — selling to businesses of every size, sitting inside their operations, and seeing where they were leaving money on the table. The pattern was always the same: too much manual work, too many leads slipping through the cracks, not enough hours in the day.
NxtLayr AI exists because the AI tools that genuinely transform operations weren't reaching the businesses that needed them most — manufacturing, industrial, multi-region distributors, trade-facing companies. The kind of companies where AI can compound across thousands of conversations, orders and customer touchpoints. The kind of company OX Group is.
I work on every engagement personally. The Voice Receptionist build for OX would be designed, scoped, built, scale-tested and handed over by me — not handed off to a junior delivery team. That's deliberate: senior judgement on every routing rule, every escalation path and every line of the agent's script is a different quality bar to a template-driven implementation run by a delivery pool.
Retell account, Twilio account, Airtable workspace, agent JSON — all transferred to OX ownership at handover. No platform lock-in, no markup, no NxtLayr in the middle of your data.
Every flow scale-tested across 50 scenarios via Retell Batch Testing before any live call is taken. Continuous monitoring and QA on 100% of production calls.
Build timeline
Discovery & architecture sign-off
Week 1Confirm Airtable schema, validate Celigo data pipe, lock routing extension map, finalise voice direction.
Middleware + integration
Weeks 2-3MCP server built on Modal. Airtable read paths wired with column-level scoping per intent. Twilio SIP trunk configured against a test DID.
Agent build
Weeks 4-5All five conversational flows built in Retell. Voice tuning. Door-keeper logic on rep transfers.
Scale testing
Week 6Retell Batch Testing runs 50 realistic call scenarios. Edge cases logged and addressed before any live call is taken.
UAT with Liza's team
Weeks 7-8Liza, her three inbound reps and the management team review live calls. Iterate on tone, transfer thresholds, escalation logic.
Soft launch on test DID
Week 9Live calls on one test number. Monitored by Liza and NxtLayr. Performance dashboards spun up.
Full AU cutover + handover
Week 10Main 3CX line cuts over. Full handover documentation delivered. Training session with Liza's team. Optional retainer activates from go-live with the 6-week waiver.
What happens next
Project confirmation & scope
Voice AI Receptionist, Chat Widget, or both? One email back to confirm scope.
50% deposit paid
NxtLayr issues the deposit invoice. Remaining 50% billed on cutover.
Project kickoff
AU pilot live within 10 weeks of kickoff. 2 weeks post-launch support included.