Skip to content
Omnichannel communication · Built and run by MxD

Every customer conversation. One platform. Nothing dropped.

Your customers message you on email, chat, social DMs, SMS and WhatsApp. This is the platform that catches every one of those conversations in one governed place, answers the routine ones with AI that can only say what you have approved, and hands the sensitive ones to a human with the full story attached. Built for your business by MxD, foundation first.

Message flowExample
One ordered stream

  • Order query

    A. Rivera · Email

    Captured

  • Booking change

    J. Nakamura · WhatsApp

    Deduplicated

  • Opening hours

    M. Okafor · Instagram DM

    Delivered

  • Refund request

    P. Lindqvist · SMS

    Human alerted
Delivery confirmedHuman alerted
  • Every channel in one auditable place
  • AI answers only from approved knowledge
  • Sensitive topics always reach a human
  • You own every account and key
Why this exists

You cannot trust AI you cannot audit. And you cannot audit six inboxes.

Right now your customer conversations are scattered: email here, Instagram there, WhatsApp on someone’s phone, website chat somewhere else again. Scatter is not just slow, it is unmeasurable. What cannot be audited cannot be improved, or trusted. Bolt an AI onto that mess and you get the worst of both worlds: people and AI agents replying in random places, and you logging into all of them to find out what was said. So the order of work is deliberate: one cohesive, reliable communication platform first. Intelligence second, once it has something solid to stand on.

Foundation first. Then intelligence.

Scatter, resolvedExample
  • Email inbox
  • Instagram DMs
  • WhatsApp on a phone
  • Website chat
  • Shared team inbox
  • SMS replies

One ordered stream

How it works

Three pillars, modelled on how your best person actually works.

Reliable like your most dependable employee. Knowledgeable like your most experienced one. Improving like your most ambitious one.

The Communication Bridge

never drops a message

Its one job is reliability. Every inbound message is written down the instant it arrives, before anything else happens. Duplicates are caught, so nobody gets answered twice. Every reply is tracked until it actually lands, and anything that fails is retried and then raised to a human, never lost. More than one delivery provider sits behind the platform, so a single outage never takes a channel dark. Think of a switchboard operator who never drops a call.

Message capture, dedupe, ordered queue, AI or human handling, delivery tracking and confirmation, with failover and human-alert branches.
Capture on arrival
Dedupe
Ordered queue
AI or human
Delivery tracking
Confirmed
Retry, then a human is alerted
Failover: a second delivery provider

The Organisational Brain

knows your business, says only what you approve

Most AI knowledge bases are a filing cabinet: flat documents the model rummages through. A brain is a network. Your products, policies, events and content are connected the way your most experienced person connects them, so a question about one thing pulls in the context around it. The exact facts, prices, dates, refund windows, are never paraphrased; they are retrieved as records. And the AI answers only from knowledge you have approved. If the approved answer does not exist, it hands off to a human rather than improvising.

A flat filing cabinet of documents contrasted with a connected network of approved knowledge nodes lighting up around a query.

Filing cabinet

  • Document
  • Document
  • Document

Connected brain

  • Products
  • Policies
  • Events
  • Memberships
  • Content
  • Prices as records

A question pulls in the context around the answer. No approved answer, hand off to a human.

The Dream Cycle

gets better every night, under your control

Every night the system reviews the day: every conversation, every reply, every place a human stepped in to correct it. It grades itself, finds the gaps, flags stale or conflicting knowledge, and proposes improvements. Then the crucial part: a human approves anything that changes what customers are told. You get the compounding intelligence of a self-improving system with none of the drift of one that rewrites itself unsupervised. And you can watch it work: delivery reliability, guardrail activity and the improvement trend are dashboard metrics, not promises.

Nightly cycle: replay, grade, detect, clean, propose, a human approves, then apply and measure.
  • Replay
  • Grade
  • Detect
  • Clean
  • Propose
A human approves
Apply and measure, then back to tonight’s replay
The invisible 80%

Any developer can wire a chatbot to your inbox. That is the easy 20%.

A typical low-code automation build demos well: messages flow, the bot replies, everyone claps. Then reality arrives. Social messaging enforces 24-hour reply windows. WhatsApp splits pre-approved templates from free-form replies, each with its own rules. SMS compliance changes country by country, with opt-outs that must be honoured automatically. US and EU obligations apply at the same time. Providers send duplicate webhooks, drop sessions, and rate-limit you mid-conversation. A platform absorbs all of that by construction: dedupe, ordered delivery, retries, a dead-letter queue with a human alert, consent checked centrally on every send. A workflow tool makes you rebuild those guarantees by hand, and you find out which ones are missing one dropped customer at a time.

A cheap build can be made to work, with enough custom engineering bolted on. The gap in price is the cost of the engineering that has not been scoped yet.

The economics of AI-handled conversations are already public: Intercom prices its Fin agent at $0.99 per outcome and publishes a 76% average resolution rate across more than 12,000 customers. The models work. What separates outcomes is everything underneath: whether messages reliably arrive, whether the AI is allowed to improvise, and whether anyone approves what it learns.

Source: Intercom’s published pricing and figures, checked August 2026.

The easy 20%

A demo chatbot wired to one inbox

The invisible 80%

  • Reply windows
  • Template rules
  • Country-by-country opt-outs
  • Dual jurisdictions
  • Duplicate webhooks
  • Rate limits
Built for businesses where a wrong answer costs real money

Your company owns every word its AI says. This platform is built backwards from that fact.

  • Approved knowledge only.

    The AI cannot answer from the open internet or its own imagination. No approved answer, no improvised reply: it hands off.

  • Rules decide, not vibes.

    AI classifies the message; deterministic rules make the routing call. Refunds, complaints, health, legal and anything emotionally sensitive hard-route to a human, no matter how confident the model feels.

  • Guardrails outside the model.

    Blocked claims and required disclaimers are enforced by code a clever prompt cannot talk its way around.

  • A human approves every change.

    The nightly learning loop proposes; your team approves anything that alters what customers are told. Every version is kept, so you can prove what the AI knew at any moment.

No hostage infrastructure

Every account, every key, every workflow: opened in your name.

The platform accounts, the AI services, your CRM: all provisioned as yours, not rented back to you. Workflows, prompts, knowledge and configuration are delivered as documented, exportable assets. Conversation and customer data lives in your systems. A team and full documentation stand behind the build, so it survives any one person. If we part ways, the system keeps running and every piece stays with you.

  • Your accounts
  • Your data
  • Your exportable assets
Take these into every sales call

Six questions that separate a platform from a demo.

Ask these of anyone proposing to put AI in front of your customers. Ask us too; answering them on this page is the point.

  1. 01

    What happens to a message when a provider goes down or sends a duplicate?

    If the answer is not dedupe, retries, and an alerted human, messages will vanish silently.

  2. 02

    Can your AI say something you have not approved?

    If it retrieves from anywhere beyond governed, versioned knowledge, the honest answer is yes.

  3. 03

    Who approves what the AI learns?

    “It improves automatically” means it changes what it tells customers with nobody checking. That is how the public failures happened.

  4. 04

    What decides when a human takes over?

    If it is only the model's own confidence, the dangerous case, confident and wrong, has no gate.

  5. 05

    Where do the messaging-window, template and opt-out rules live?

    Every channel has them. Someone is enforcing them centrally, or breaking them quietly.

  6. 06

    Who owns the accounts, and can you export everything?

    If leaving the vendor means losing the system, you are renting, not buying.

What comes later, on purpose

Voice can wait. The foundation cannot.

  • Voice

    Joins as a channel once the foundation and the AI layer have proven themselves in writing first; a live caller is the wrong place to discover a gap.

  • Member and customer journeys

    Proactive follow-ups and reactivation plug into the same foundation later.

  • A second brand

    The platform is built brand-aware from day one, so another brand is added as configuration, not a rebuild.

Each step lands on the proven floor, which is the whole point of building the floor first.

Questions

Frequently asked questions

Is this a chatbot?

No. A chatbot is one channel and one script. This is the communication layer for your whole business: every channel captured in one governed place, AI resolving the routine conversations from approved knowledge, humans handling the sensitive ones with full context, and a nightly, human-approved improvement loop. The AI is one part; the platform is the product.

Which channels does it cover?

Email, website chat, SMS, and the social messaging channels your customers already use, including WhatsApp, Messenger and Instagram DMs. Voice is deliberately on the roadmap rather than in the launch scope: it joins once the foundation has proven itself in writing.

Does it replace our CRM?

No. Your CRM stays the system of record for the relationship: contacts, deals, tickets, timeline. The platform integrates with it and keeps the messaging machinery separate, so replies never queue behind a CRM rate limit. If you change CRM later, the platform is built so that is a configuration change, not a rebuild.

What stops the AI saying something it should not?

Three layers. It can only answer from knowledge you have approved, so there is nothing else for it to say. Deterministic guardrails outside the model block disallowed claims and inject required disclaimers. And sensitive topics, money, health, complaints, distress, hard-route to a human regardless of the model's confidence.

How does it get better without going off-script?

Every night it reviews the day's conversations, grades itself, and proposes improvements: better answers, filled gaps, retired stale knowledge. Your team approves anything customer-facing before it goes live. Improvement is the system's job; authorising it stays yours.

What do we own at the end?

Everything. Accounts and keys are opened in your name, workflows and knowledge are delivered as documented exportable assets, and your data lives in your systems. If the engagement ends, the platform keeps running and stays yours.

How long does it take?

The foundation, every channel consolidated onto one reliable platform, comes first and is visible early. The AI layer follows once the foundation is proven. Exact timelines depend on your channel mix and knowledge; that is the first thing we scope in a conversation.

What does it cost?

Every engagement is scoped to your channels, volume and knowledge, so there is no rate card on this page. The platform and AI services carry their own usage-based licensing, billed in your own accounts, so you own them directly. A conversation gets you a number; bring the six vendor questions above to it.

Put AI in front of your customers without betting the brand.

A conversation about your channels, your volume and your knowledge. No demo of a generic bot, no obligation, and you leave with the vendor questions answered either way.