The operational gap
Chat is excellent for ordering.
It is not an order system.
Restaurant customers naturally send shorthand, mixed languages, corrections, delivery notes, and partial product names. Supplier teams must turn every message into accurate operational data—usually by hand.
Information is unstructured
Product, quantity, unit, timing, and special instructions arrive in one conversational message.
Knowledge lives in people
Aliases, preferred products, pack sizes, and client prices depend on individual memory.
Mistakes surface too late
Ambiguity, missed changes, and retyping errors can reach customer service or the warehouse.
The Orderline product
One controlled layer between customer chat and fulfilment.
Orderline keeps the customer’s familiar ordering channel while giving the supplier a structured, traceable workspace. It supports the sales team; it does not replace their commercial judgment.
How it works
From message to confirmed order
A complete operational flow with a human-controlled decision at the point that matters.
-
1
LINE
Receive
Capture the raw customer message first, preserving the original source for auditability.
-
2
AI
Extract
Identify candidate items, quantities, units, intent, delivery hints, and notes.
-
3
SKU
Validate
Match against the real tenant catalogue using deterministic product and alias rules.
-
4
THB
Price
Apply client-specific prices, units, pack constraints, and supplier delivery rules.
-
5
OPS
Review
Surface ready, held, cancellation, change-request, and needs-review work in one queue.
-
6
✓
Confirm
Create the final order and structured internal handoff only after controlled confirmation.
The operator workspace
Every order shows the next safe action.
Sales sees a focused operational queue instead of searching chat history. Clear orders move forward; uncertain orders stay visible and held until a person resolves them.
- Ready Catalogue-matched and safe to confirm
- Needs review Product, pricing, unit, or delivery attention
- Held Customer intent preserved without guessing
- Changes Add, update, remove, and cancel through a controlled path
2 items · Product choice required for “buffalo mozzarella”
4 items · Delivery Thursday · Client pricing applied
Designed for trust
AI reads language.
Orderline controls the order.
The product deliberately separates language understanding from business decisions. That boundary makes automation useful without letting uncertain model output silently become operational truth.
- Candidate item text
- Quantity and unit evidence
- Order or change intent
- Delivery hints and notes
- Language context
- Product and SKU selection
- Client price resolution
- Unit and pack constraints
- Delivery validation
- Confirmation and final order
Ambiguous or unmatched items remain held for review.
Prices come only from deterministic tenant and client rules.
Confirmation is a separate, explicit business action.
Raw intake and key workflow decisions remain traceable.
Product value
Less order administration.
More operational confidence.
Reduce manual retyping
Convert chat evidence into structured draft lines for the team.
Catch uncertainty earlier
Surface product, unit, price, and delivery blockers before fulfilment.
Standardise client knowledge
Keep aliases, preferences, catalogue rules, and pricing in the system.
Create cleaner handoffs
Send confirmed, structured order information to internal teams.
Operating economics
Cost scales with people,
not with servers.
A 170-fold increase in clients raises the monthly running cost by roughly 3.6×. Every figure below is taken from live billing or modelled from measured system behaviour.
Per month for the complete system, running two fully independent environments.
Per month, as a planning figure. The modelled total is $257; the remainder is headroom.
Hosting, transactional email, and AI language processing combined.
Two full environments with backups carry the pilot today and a thousand clients after one upgrade.
Under half a cent of language processing per order, and only for the messages that need it.
Load testing puts throughput headroom several times above projected demand at a thousand clients.
Confirmation is deliberately a person’s decision. That is the cost that grows, and where leverage is planned.
Modelled on 22 ordering days a month, one order per client per ordering day, three AI-parsed messages per order, and one confirmation email per order—about 22,000 orders a month. Structured orders bypass language processing entirely and cost nothing to parse. Under deliberately conservative assumptions—five parsed messages per order, a managed database, and higher email volume—the upper bound is roughly USD 680 a month.
Current product stage
Live production pilot
The full ordering workflow now runs in production. Real restaurant orders arrive through LINE, are reviewed and confirmed by sales, and reach fulfilment as structured orders—while the pilot expands restaurant by restaurant.
Running in production today
- LINE text intake and raw-message storage
- Natural-language candidate extraction
- Deterministic catalogue matching
- Client-specific pricing and rules
- Held drafts, review, and confirmation
- Dashboard queue, audit trail, monitored backups
Live pilot focus
- Onboarding pilot restaurants and their pricing
- Refinements driven by real order traffic
- Operator review workflow polish
- Learning from sales resolutions over time
- Weekly supplier price-list imports
- Controlled path from pilot to full launch
Orderline
Keep the conversation.
Control the order.
Supplier-owned ordering infrastructure for the way restaurants already buy.