Three AI questions are worth your money. Do the assistants name you when a buyer asks one? Does the repetitive half of your week still need a person? Is the screen your team lives in the wrong shape? We work on all three.
250+ active client engagements across five countries · An eight-module finance platform with an AI co-pilot, delivered to user acceptance testing in 14 weeks for a major Victorian council · A marketplace operator's three SaaS tools replaced by one custom dashboard · We publish an llms.txt and a full-text corpus on this site so an assistant can read us without guessing
Client case studies
AI product imagery and a working room visualiser.
We rebuilt this Melbourne tile and bathware store on Shopify, with product descriptions, AI-assisted imagery, a room visualiser and connected catalogue workflows. Open the full story to see the work.
Showing 1 of 1 client projects
1/4 · A storefront organised around the renovation
Website development
Boutique Home Centre
Retail & productsWebsite designShopify
Melbourne, Australia
3,592
published product pages in the Shopify catalogue
Public sitemap review · 22 September 2026
Shopify website design for Boutique Home Centre in Melbourne, with a detailed product catalogue, AI-assisted content and a room visualiser.
Boutique Home Centre’s brief was to move its home-improvement catalogue into a better organised Shopify store and fill gaps in product information.
Delivered: Product pages with useful descriptions, specifications and coverage details. Collections with filtering, sorting and product comparison.
Outcome: Delivered a store where customers can find a product, compare its details, plan an order and ask about freight before buying.
No jargon and no talk of revolution. Every one of these is something you can go and check this afternoon — on your own website, or by watching your own team for an hour.
The answer now sits above the links
Google writes a summary at the top of the page and names a few sources inside it. If your buyer's question gets answered up there, the click you used to earn is never made.
Where it happens: Google AI Overviews and AI Mode
What changes: the answer arrives before the list of websites
Check it today: search your own top question and look at the top of the page
People ask an assistant the way they'd ask a person
Whole sentences instead of three keywords. That is a different shape of question, and most websites were written for the old shape.
Old question: “blocked drain brunswick”
New question: “who's good for blocked drains near Brunswick, and roughly what does it cost?”
What it means for you: the answer has to be on your page, in plain words
There are several front doors now, not one
Each assistant reads differently and some read Bing rather than Google. Being missing from one removes a whole assistant, not a slice of traffic.
Google AI Overviews and AI Mode
ChatGPT search · Google Gemini · Perplexity
Microsoft Copilot — which reads Bing, an index most businesses have never checked
The AI crawlers are switches of their own
A crawler is the software that reads your website on a company's behalf. The AI ones are separate switches from Google's, and plenty of sites have them turned off because a plugin decided, not a person.
The names: GPTBot, Google-Extended, PerplexityBot, ClaudeBot
Where the switch lives: one file on your site called robots.txt
Who usually set it: a security or SEO plugin, silently
How long it takes to change: minutes, once someone looks
The copying-between-screens work became automatable
Not the judgement and not the relationship. The parts where somebody moves the same detail out of one system and into another.
An enquiry re-typed into the CRM by hand
A quote re-typed again into the invoice
Chasing a quote nobody replied to
Filing, tagging, transcribing, and the Monday report
The tool your team lives in can now be built for you
A dashboard or an internal tool used to be a twelve-month project. It is now weeks — so the spreadsheet holding your operation together is no longer the only realistic option.
A job board instead of a shared spreadsheet
One dashboard instead of three separate logins
A quoting screen that raises the invoice itself
Reporting shaped like the question your executive actually asks
The tools got cheap and the discipline did not
Anyone can wire something together in an afternoon. Making it survive week forty, on real customer data, is still engineering.
Cheap: a demo that works once, in front of you
Expensive: something still running correctly in ten months
The difference: testing, logging, permissions and a named owner
None of it reports itself
No platform tells you how often an assistant mentioned your business. No automation tells you it quietly stopped working three weeks ago.
There is no impressions report for an AI answer
Automations fail silently, not loudly
So measurement gets built at the start, or it never gets built
Three different jobs
One word, three separate jobs.
Almost every conversation that starts with “we should do something with AI” is really one of these three. They share a principle and almost no tasks. Working out which one you mean is usually the most useful ten minutes of the first call.
01
Being found
The assistant answers, and somebody gets named. It should be you.
Getting your business named and linked when a buyer asks an assistant instead of a search engine. It is a discipline of its own, so it has its own page.
Say plainly what your business is — in the same words, everywhere
Answer the real question in the first sixty words of the page
Decide on purpose which AI crawlers can read you
Get other sources — directories, reviews, press — to back you up
The repetitive half of the week stops needing a person.
Automation inside your own business, built into the systems you already pay for — with a person in the loop everywhere a mistake would cost you a customer.
An enquiry arrives and files itself in the CRM
A quote goes out, then chases itself if nobody replies
The phone gets answered at 8pm and the job gets booked
The screen your team lives in gets built around your job.
Dashboards, internal tools and full platforms, with AI as a feature inside them rather than the reason they exist. Councils buy this. So do eight-person businesses.
One operations dashboard instead of three separate logins
A job board, a quoting tool or an approval queue that fits your process
An assistant that answers from your own price lists and documents
Reporting shaped like the question your executive actually asks
A few specific, unglamorous workflows and — where it earns its place — one screen built around them. Everything runs inside systems you own.
One named owner per automation
A written rule for what it may do without asking
A log of everything it did
Reviewed every month by someone on your account
It is not a product you buy once. Someone can tell you every month what it ran, what it got wrong and what changed since last time.
Who is doing the work
SoudCoh · client record
250+
Active client engagements
Australia, the United Kingdom, Saudi Arabia, the UAE and New Zealand
SoudCoh Compound™
1,093
Account changes pushed through our approval queue
Proposed by our tools or our team, and pushed only once approved
Published on this site
2
Machine-readable corpora for AI crawlers
An llms.txt and a full-text file — we do on our own site what we ask of yours
These describe SoudCoh's own practice, not AI outcomes. There is no honest industry benchmark for what automation or AI visibility will do for a particular business, so no outcome figure appears on this page. The delivered engagement figures further down are dated per case study and are not a forecast.
Ten things happen on an AI engagement. Here they are, in order.
Not every engagement uses all ten. Automation clients usually take steps one to five, build clients take six, and visibility clients take nine. The drafting, searching, watching and filing run on software. The decisions about your business are made by someone you can ring.
01
Follow the work before automating any of it
We sit with your team and follow one job the whole way through. Most of what gets sold as an AI problem is a form in the wrong place or a field somebody types twice.
The job, end to end: enquiry, quote, schedule, do, invoice, chase
What we write down: who touches it, what they open, what they re-type
What you get: a map of where the hours go — yours to keep
02
Connect the tools you already pay for
Most businesses do not need another platform. They need the five they already own to stop needing a person in the middle re-typing the same customer's details.
Enquiry form into the CRM, once
CRM into the quote, then the quote into the invoice
All of it into one report nobody rebuilds by hand
03
Draft with software, publish with a person
Software gets you to a rough draft in seconds. Somebody who knows your business decides what goes out with your name on it, and that order never reverses.
A first draft of an ad
A reply to an enquiry you answer forty times a month
A summary of a long phone call
The skeleton of this month's report, ready to be corrected
04
Answer the calls you are currently missing
A missed call after hours is a job that went to somebody else. A voice assistant picks up, takes the details and books the job.
It says what it is, in the first sentence
It takes the name, the address and the problem
It books into your calendar and texts a confirmation
It hands anything unusual or upset straight to a person
05
Put your own documents behind the answers
Rather than answering from general knowledge, the assistant reads your material and tells you which document the answer came from. The industry calls this retrieval.
What it reads: price lists, SOPs, warranty terms, past jobs, supplier specs
What it returns: the answer, plus the file and the line it came from
Why that matters: a person can check it in one click
06
Build the tool when the spreadsheet runs out
Job boards, quoting tools, approval queues and dashboards. These are custom platforms first and AI second — the model is a feature inside the tool, never the reason it exists.
A dashboard that replaces three logins and a weekly export
A quoting screen that raises the invoice from the same record
An approval queue, so nothing goes out unreviewed
A mobile app for the field team and a console for the office
07
Write down what the software is never allowed to do
Agreed in writing before anything touches a live account, because it is a business decision rather than a technical one.
Which actions need a human signature
Which data never leaves your environment
What gets hidden or removed before a model sees it
What happens when the software is not sure
08
Keep it private where privacy is the real requirement
Most businesses do not need this level of isolation and we will say so. The ones that do need it built properly rather than promised.
Models run on hardware you control, not a third party's
Encrypted storage and role-based access
An audit log of who saw what, and when
Written in the contract, not described on a call
09
Make the business readable to the assistants
The visibility half. One clear description of what you are, facts marked up so nothing has to be guessed, and pages that answer the question at the top.
One description of your business, matched everywhere it appears
Structured data — facts written in a format machines read directly
Pages shaped like the question, answered in the opening lines
A deliberate decision about which AI crawlers get in
10
Measure it, then keep watching it
Automations fail quietly and mentions disappear quietly. Both need watching on a schedule rather than when somebody happens to notice.
Logging on every automation, and an alert when one stops
A monthly check on what the assistants say about you
A report a non-technical person can read without a translator
Walk the job first. Automate the expensive part second.
Week 1
The walkthrough
We follow the work rather than the org chart. The output is a written map of where the hours go, and you keep it whether or not you hire us.
Who touches a job, and in what order
What they open, and what they type twice
Where a job sits waiting for somebody
Week 2
The shortlist, with the boring options left in
A ranked list of fixes, each with a reason and a rough effort. It deliberately includes the ones that need no AI at all.
Sometimes the fix is a mandatory field, not a model
Sometimes it is one integration nobody ever built
You see the order, and the reasoning, before work starts
Weeks 2–6
One workflow, built narrow and built properly
We finish a single job end to end inside your existing systems instead of demonstrating six half-things. Narrow and finished is the only version that survives a busy week.
One job, chosen because it is the expensive one
Built into the tools you already own
Working on real data, not a sample file
Weeks 4–10
The person in the loop, placed on purpose
Nobody should have to guess who to ring when something looks wrong, so it is decided and written down before go-live.
Which steps stop and wait for an approval
Who gets told when something escalates
What happens when the software is unsure
One named owner per automation
Monthly
The report, and the failures in it
On a fixed rhythm, good news or bad. If something went backwards you hear it from us first.
What ran, and how often
What broke, and what a person had to correct
What the assistants said about you this month
What we are changing next
Ongoing
Re-reading the tools
Models, prices and limits change without a changelog and sometimes without notice. Part of the retainer is somebody whose job is to notice.
A model gets retired, or gets better
A price changes, or a limit moves
We tell you when something we relied on stopped being true
Nothing we automate goes live without a named owner and a written rule for what it may do on its own. In Compound, that rule is written down with its reason, so the next build starts from it.
The stack, named
Engineered, not prompted.
You should know what your business is running on. Each row is a layer, what it is for in plain words, and the products we build with — which change more often than any of them like to admit.
Models
The engines that read and write. We pick per job, not per fashion.
OpenAI · Anthropic Claude · Google Gemini · xAI Grok · Meta Llama
Workflow
The wiring that makes step two happen after step one, reliably.
LangGraph · n8n · Make · Temporal
Retrieval
How an assistant finds the right page of your own documents.
Pinecone · Weaviate · pgvector
Voice and telephony
Answering the phone, transcribing the call, sending the text back.
Vapi · Retell · ElevenLabs · Twilio
Interfaces and platforms
The dashboards, consoles and mobile apps people actually click on.
React · TypeScript · .NET · Postgres · Supabase
Hosting
Where it runs. Australian hosting where the data requires it.
AWS · Google Cloud · Supabase · Cloudflare
Safeguards
Testing it, watching it, hiding personal data, and keeping a record.
Product marks identify the model providers our tooling is built against. They are not partnership badges and no affiliation or endorsement is claimed. Nothing we build is designed so that it cannot be moved to a different provider.
What runs on its own, and what carries a signature.
Eight jobs make up an AI engagement. Four run on their own once built. Four are decisions, and a decision goes out with a name against it. The line between them is drawn in writing before anything goes live.
Runs on its own
Writing the first version of something
Quick, cheap, and always read by somebody before it leaves the building.
A first draft of an ad
A reply to an enquiry you answer every week
A summary of an hour-long call
The skeleton of a report, ready for a person to correct
Runs on its own
Finding the answer inside your own documents
The answer comes back with the document it came from, so a person can check it in one click.
Price lists and rate cards
Procedures, warranty terms and contracts
Notes from jobs you did two years ago
Runs on its own
Doing the same small thing ten thousand times
The work nobody was ever glad to be doing.
Copying a detail from one system into another
Tagging, filing and transcribing
Chasing a quote nobody replied to
Runs on its own
Watching for the thing that broke at 3am
It watches continuously, because the failures that cost you money are the quiet ones.
A data feed stops arriving
A form starts failing silently
Ad spend spikes overnight
An automation stops running and says nothing
Signed before it ships
Deciding what your business actually is
Written down once, then every page, ad and assistant answer is held to it.
What you sell, and what you do not
Who you are for
What is true enough to publish
Signed before it ships
Deciding what an automation may do on its own
Where a mistake would cost you a customer, a person signs. It is a business decision rather than a technical one.
Drawn before anything goes live
Written down, not agreed on a call
One named owner per automation
Signed before it ships
Talking to your customers when it matters
A voice assistant can take an after-hours booking. It hands over the moment a call stops being routine.
It tells the caller what it is
It hands over when somebody is upset or unusual
It leaves a transcript, so nothing rests on memory
Signed before it ships
Reading the report and changing the plan
Specialists who sit together in Melbourne, on your account. The report is where the plan changes.
Automation is only safe where accountability already exists.
Compound runs every account as a ladder: a footing, six rungs and one engine. Automated work leans hardest on these two, because they decide what an unattended system counts and what it may repeat.
Bidding should learn from calls and forms, not page views or taps on a phone number. Getting that right is the first thing we check, because every rung above it is only as honest as what it counts.
Automation is only as honest as what it counts. We check that calls and forms, not page views or taps on a number, are what any automated system learns from.
Your account does not start from zero. It opens with the negatives, checks and platform rules the accounts before it already paid to learn.
Automation you can trust follows written rules. On our accounts, a near-miss becomes an automatic check and every rejected ad becomes a written rule with its reason.
Four platforms we built. Two are government engagements, named by sector and jurisdiction only — a policy, not a hedge. In all four the AI is a feature inside a tool people use all day. Never the product, and never the thing making the decision.
Cost, timing, contracts, who owns what, whether we can build you a dashboard, what happens when it does not work, and whether any of this is coming for your staff.
It starts with a walkthrough, not a tool. We follow one job the whole way through and write down where the hours go. Then you get a ranked list of fixes, including the ones that need no AI at all. From there we build the most valuable one properly, inside the systems you already own. A named person stays in the loop wherever a mistake would cost you a customer. Every month you get a report on what ran, what broke and what a person had to correct.
No. There are three separate jobs on this page and most clients buy one or two of them. The first is being found: getting named when a buyer asks an assistant instead of a search engine. The second is being run: automating the repetitive half of the week inside your own business. The third is being built: a custom dashboard, internal tool or platform shaped around how your team actually works, with AI as a feature inside it. They share a principle and almost no tasks.
Yes. A marketplace operator replaced three disconnected tools with one workspace for vendors, listings and enquiries, supported by an AI assistant the team uses daily. A property maintenance business brought customer updates, mobile job access and administration together in eight weeks. At a larger scale, a major Victorian council's eight-module finance platform reached user acceptance testing in fourteen weeks. The custom platform page explains how we scope this work.
The walkthrough and the written map can be bought on their own, so you can see the shortlist before committing to a build. After that it is scoped on the actual work: how many systems have to talk to each other, whether your data can sit with a mainstream provider or has to be hosted privately, whether we are automating inside your existing tools or building a new one, and how much sign-off is required. You get a real number for your situation in writing before anything starts. The first call is free.
The walkthrough takes a week and the shortlist lands the week after. A first narrow automation — one job, end to end — is usually live between four and eight weeks, depending on how cooperative your existing systems are. A custom dashboard is typically two to four weeks; a fuller platform is longer and gets phased. We would rather ship one finished thing than demo six half-things, because the half-things are the ones nobody opens again after the first busy week.
There is no long-term lock-in. What we ask for is enough runway to be fair to the work, because judging an automation three weeks in is judging it during setup. The terms are put in front of you in plain English before anything is signed, and if the honest answer at any point is that a workflow is not worth automating, we would rather say so than keep a retainer running quietly.
A marketing and technology team — specialists who sit together in Melbourne, working on your account. Our tools draft, search and transcribe; a person reads what comes back. The judgement goes out with a name against it: what your business is, what an automation may do on its own, what is true enough to publish. That name is on the monthly report too.
That is not the pitch and it is not what we sell. The work we automate is the copying between screens: re-typing a customer's details into a third system, chasing an unanswered quote, filing, tagging, transcribing. What is left over is the part your team is actually good at. If a business genuinely wants to cut headcount, an automation project is a slow and unreliable way to do it, and we would say that in the first meeting.
You do. Accounts are created in your name where the platform allows it, integrations are built against your credentials, and the prompts, workflows and configuration are documented and handed over. For custom platforms we ship source-code escrow and a defined data exit — that is how the government engagements are structured, and there is no reason a smaller business should get worse terms.
We tell you in the monthly report and change what we are doing. Usually the failure is visible and boring: an integration broke, a step needed a person after all, or the workflow we automated was not the expensive one. Where the honest answer is that a process should not be automated, we say so and stop billing for it. What we will not do is let something run unattended and hope you do not check.
A monthly document in plain English: what each automation ran, what failed, what a person had to correct, what changed, and what we are doing next — alongside what the assistants said about your business this month if that work is in scope. On a fixed rhythm, good news or bad. If something went backwards you hear it from us first, not from a customer.
SEO wins you a position on a page of links. Generative engine optimisation makes you eligible to be named inside an AI-written answer — a discipline in its own right, with its own page. This page covers that visibility work plus two other jobs: automating work inside your business, and building the tools your team uses. They share a principle and almost no tasks. Most businesses start with the visibility half because it is the cheapest to fix.
Yes, and it is one of the most useful places to start for a service business, because a missed call after hours is a job that went to somebody else. A voice assistant picks up, takes the details, books into your calendar and texts a confirmation. It is set up to say what it is, to hand anything unusual or upset straight to a person, and to leave a transcript so nothing depends on anyone's memory.
It can. Where the sector or the data requires it we use self-hosted models, encrypted stores, role-based access and audit logs, and we set out in writing what leaves your environment and what never does. Most businesses do not need that level of isolation and we will say so rather than sell it. What every engagement gets is a written rule for what is redacted, what is retained and who can see it.
Mainstream providers — OpenAI, Anthropic, Google, and open models where hosting them ourselves is the right call — with orchestration, retrieval, voice and monitoring layers around them. The tools are listed on this page because you should know what your business is running on. They also change constantly, so nothing we build is designed to be impossible to move, and re-reading the landscape is part of the retainer rather than an extra.
— 12 · Nearby
AI is a layer on other work, not a replacement for it.
Nothing on this page stands on its own. Most engagements run two or three of these together, and the visibility work, the automation work and the build work usually start at different times.
A free 30-minute session. We work out what can genuinely be automated, what needs a person in the middle, what is worth building and what is a quick win — before you commit to anything. You keep the map either way.
