Broa answers your customers' questions automatically — on email, web chat, text, web forms and mobile apps — using your own catalogue, pricing and order records. You decide how much goes out without a person checking it first.
Your team handles the exceptions, not the repetition. One pipeline serves every channel, so growing volume does not mean growing headcount.
Web chat northgate-eng.com · 09:12
Do you have the X-40 couplings in stock? We'd need 250, and a price if we go to 500. Also chasing order 4471, was due Friday.
Reply · generated by Broa
Yes — 250 X-40 couplings are in stock and can ship tomorrow. At 500 units your contract price drops to £18.40 each. Order 4471 left our warehouse Thursday and is due with you today; tracking is below.
Grounded in Catalogue Price book · Northgate Order 4471
The problem
Customer service is the same question, asked five different ways.
Where is my order. Do you have it in stock. What does it cost at 500. When will it ship. The answers are sitting in systems you already own — and your team retypes them all day, into a different window depending on where the customer happened to ask.
Every channel is its own island
A helpdesk for email, a widget for chat, an inbox for the contact form. Three tools, three queues, three sets of rules, and an answer that depends on which one the customer found.
Generic AI guesses
A model with no access to your data will produce a confident, fluent answer about a product you do not sell, at a price you never set.
Hiring scales linearly
Twice the volume means twice the service staff, and the fiftieth reply of the morning is no faster to write than the first.
Every question is an order waiting on an answer.
A distributor's inbox is a pipeline of orders nobody has placed yet. Broa answers them in seconds, from your own stock and your own prices — so fewer of them go looking elsewhere while your team works through the queue.
Every channel
Seven ways in. One pipeline.
A message is normalized the moment it arrives, into a record with a channel-neutral core. Everything downstream — classification, retrieval, generation, approval — reads that core and never asks how the message got here.
- De-duplication
- Spam & phishing quarantine
- Attachment sandboxing
- Size & rate limits
- Tenant resolution
Channel-neutral core, plus the envelope its channel needs to reply.
Natural language
No forms. No menus. No keywords.
Your customers write the way they would write to someone on your team — half a part number, three questions at once, no punctuation to speak of. Broa reads it the way a person would, which is why nobody has to be taught how to ask.
Three questions, one message
Asks are separated and answered together, rather than the first being answered and the rest quietly dropped.
"Do you have the X-40 in stock? We'd need 250, and a price if we go to 500. Also chasing order 4471."Their words, not your catalogue's
A description, a nickname, or whatever they called it last time — resolved against your catalogue rather than rejected for not matching it.
"the blue couplings we had on the last order"A part number that is not quite right
Superseded, mistyped, or their code rather than yours. Matched where Broa is confident, asked about where it is not — never guessed at.
"X40-B, or it might be XL40B"Context carries down the thread
A one-line follow-up is read against what came before it, so a customer never has to repeat themselves to be understood.
"and what about 500?"No keyword list to maintain
Nothing to enrol, no phrase table to keep current, and no rule that breaks the first time somebody words a question differently.
In whatever language they used
Detected on the message itself and answered in kind, from the same product data and the same prices.
Language & reach
One catalogue. Every language your customers write in.
Broa detects the language of every incoming message and answers in it — from the same product data, the same prices, the same order records. You maintain your catalogue once, not once per market.
Detected, not configured per sender
Language is identified on the message itself. A customer who writes in Portuguese on Monday and English on Thursday gets both, without anyone setting a preference on their record.
Grounded in the same data
Translation happens at the answer, not to your catalogue. There is no second copy of your product data to keep in step, and no market where your prices are three months out of date.
A language you haven't approved escalates
Broa does not quietly attempt a language you have not switched on. It routes the message to a person instead — the same rule that governs everything else it is unsure of.
Tone rules per market
Formality is not universal. Your German correspondence can be formal and your Australian correspondence relaxed, from one set of underlying facts.
A workspace per region or brand
Regions, brands and individual mailboxes can each carry their own configuration, templates and users — while staying inside your single tenant boundary.
Answers at 3am, without a night shift
The real advantage of selling across time zones is not translation, it is that a customer twelve hours away gets an answer now instead of tomorrow.
Yes — 250 X-40 couplings are in stock and ship tomorrow. At 500 units your price is £18.40 each.
Catalogue · Price book · Northgate
Sim — temos 250 união X-40 em stock, com envio amanhã. Para 500 unidades, o preço é £18,40 cada.
Catalogue · Price book · Northgate
Oui — 250 raccords X-40 sont en stock, expédiés demain. À 500 unités, votre tarif est de 18,40 £ pièce.
Catalogue · Price book · Northgate
One catalogue, one price book, one set of order records. Only the wording changes.
Plausible is the problem.
Which is why a Broa reply has to come from your data and survive validation before it goes anywhere — whether a person reads it first or not.
Autonomy
You decide how much of it runs without you
Autonomy is a setting, per tenant and per intent — not an all-or-nothing switch you flip once and hope about. Start with review on everything, then let go of the questions you have watched it answer correctly a thousand times.
Autonomy is set per question type, not once for the whole business — and the figures beside each one are what justify moving it.
How it works
Grounded in your data, governed by your rules
Broa is not a chatbot bolted onto a mailbox. It is a pipeline, and every stage is inspectable, configurable and recorded.
One message, end to end
Arrives
From any channel. Normalized, de-duplicated, threaded into a conversation.
Classified
Intent identified — stock, quote, order status — and routed to its pipeline.
Grounded
Your catalogue, pricing, order status, FAQs and prior replies are retrieved.
Drafted & validated
Written with your templates and tone rules, then checked before it can go anywhere.
Approved
By your policy, by a reviewer, or by both — you set which, per intent.
Sent & recorded
Back down the channel it came from, tracked, and written to a tamper-evident log.
- It answers from your facts
- Every reply is grounded in tenant-owned data. One that cannot be grounded escalates to a human rather than being invented.
- You set the autonomy
- Per tenant and per intent: review everything, review what falls below a confidence threshold, or review the exceptions only.
- Your tenant is sealed off
- Isolation is the platform's highest-priority quality attribute, ranked ahead of performance and cost.
- Everything is on the record
- Actions and reads alike, in a tamper-evident trail you can query and export for your tenant alone.
- New behaviour is configuration
- Adding an intent, a channel or a whole new business is a staged config change with rollback — not a release.
Straight answers
What it does, and what it deliberately does not
Every platform has a boundary. Ours is written down, rather than discovered halfway through your evaluation.
In scope today
- Email, web chat, text, web forms, mobile apps and API
- Intent classification and triage across every channel
- Retrieval over your catalogue, orders, FAQs and prior threads
- Reply generation from your templates and tone rules
- Autonomy per intent — reviewed, threshold-based or fully automatic
- Sending, delivery tracking and a full audit trail
- Language detection, and replies in your customer's language
- Many independent businesses on one platform
Not in this release
- Live spoken conversation — in development, and phased (see above)
- Outbound calling and dialling campaigns
- Sending with no approval policy defined at all — a policy is always required, even an automatic one
- Fine-tuning or training foundation models on your data
- Replacing your CRM or ERP — Broa integrates with them
Try it
Ask Broa about Broa
A customer service agent that cannot answer questions about its own company would be a poor advert. Ask it anything on this page — and notice what happens when you ask something it cannot ground.
Pricing
You pay for answers, not for seats
We have not published a price list yet. We have published the model, because it is the part that should decide whether Broa is a fit for you.
The unit
Per message answered
You are billed for messages Broa handles, not for how many people you have. A quiet month costs less than a busy one, which is how a cost that tracks your business ought to behave.
Not per seat
Charging per user would be a contradiction
Broa exists so that answering more does not mean hiring more. Billing you per person would mean profiting when that fails. Add every reviewer you like — the price does not move.
Getting started
A cheap way in
Sorting and routing your inbound, with no drafting, is a smaller commitment and the fastest way to see whether the classification is right about your business before anything is written on your behalf.
Point it at a real conversation
The fastest way to judge Broa is to give it a week of your actual customer questions — and to look at what it refuses to answer.