In the last six months we have sat with a property business that runs a large portfolio on Excel, a retail group running its procurement through a project-management subscription, and an accounting practice assembling every client’s document pack by hand. None of them asked for AI. They asked for the work to stop.
That distinction matters more than any tool we could name. Earlier this month on this page we wrote that CCG will not sell you the robot, and that the half of an automation programme nobody demonstrates is the half that decides whether it works. This is the other half. It is what the technology looks like when it is built around a business rather than sold to one, and what we have learnt building it.
Where the work hides
Nobody’s org chart has a box marked “manual work that should be automated”. It hides in three places.
It hides in spreadsheets. A month-end reconciliation someone has run by hand for years. A property portfolio managed across Excel, Word documents and a shared drive for two decades. A document pack retyped from the same client information every time a new client signs.
It hides in one employee’s head. The person who knows how the stock is counted, which supplier gets called when, what the exception is and why. When they are on leave, the process is on leave.
And it hides in WhatsApp. Enquiries arrive there, quotes go out from there, and the thread is the system of record until the phone is lost. Almost every operator we speak to, from plumbers to dog groomers, runs part of the business this way.
The technology industry sells automation for the first of these. The other two are where most of the value sits, and they only give it up if you go and look.
What it looks like when it is built
Some of the work from the past six months, with the names removed.
An accounting practice came to us with one problem. Reconciling bank transactions against its retail clients’ point-of-sale records was slow and manual. We built a reconciliation platform and it went into daily use. The first automation in a business is rarely the last. Once people watch a process disappear they start pointing at the next one, and the same practice has since commissioned four more. Two of them are below.
The first was document preparation. Every new client needs a pack of documents, and every pack was being assembled by hand from information the practice already held. Now a client’s details are captured once, with AI extracting and validating them from the documents the client sends in. The pack is generated from a template library, downloaded in a batch, and delivered by email or WhatsApp. The next phase adds signing on a phone, with the signed PDF filed straight back into the client’s record.
The second was HR on WhatsApp. The practice runs payroll for its clients, which means those clients’ employees phone and email all month for payslips, tax certificates and leave balances. Now an employee enrols once on a WhatsApp line and asks for what they need in the chat. Payslip, IRP5, leave balance, back within seconds, every request logged. The next phase routes leave requests to the manager for approval in the same chat, writes the answer back into the payroll system, and sends payday notifications. No app, no login, no training, because everyone already knows how to use WhatsApp.
A retailer with an online store could not see its stock. We built an inventory system where a purchase order is raised with the last month’s and the last six months’ sales sitting beside it, and the goods-received note is pre-populated by AI from the supplier’s invoice. Low-stock alerts are predicted from how fast each product sells rather than a fixed minimum, and every stock take produces a variance report against the last one. Sales sync in from the store, inventory value and receipts push out to the accounting system, and a sync log shows whether either side has stopped talking.
The same retail group ran its procurement business on a generic project-management subscription. A request comes in from the website, suppliers are asked to quote, a price is worked out with live exchange rates and margin, the client approves, the purchase is placed, dispatch is tracked, delivery is confirmed. Six steps, none of which the tool was designed for. We are replacing their R200k subscription with one system built around exactly that lifecycle. The subscription goes, and every request is visible at the step it is stuck on.
A tutoring business had a thousand contacts in its CRM, two salespeople to close them, and follow-up that went cold the moment those salespeople got busy. Getting leads was never the problem. Closing them was. We are building agents on WhatsApp that qualify each enquiry, follow up on the cadence a good salesperson would keep and never does, and carry the conversation through to the close, with a person stepping in only where the conversation needs one. It is the case we use to show how far agents can be pushed into a sales cycle.
An e-commerce business asked for a new website and online store. Enquiries went from twenty a month to twenty a week. Their problem is now too many customers, and the next build is a CRM that automates the selling one pipeline stage at a time. Automation does not remove a bottleneck. It moves it, and you have to be ready to follow.
The pattern underneath
Every one of those started with a process, not a product. We did not arrive with a platform to install. We sat in the business, found the specific thing costing time or money, and built around the way that business already runs, rather than asking it to adapt to software designed for someone else.
The rest of the pattern is consistent enough that we now treat it as method.
The first build is a door. Scope it tightly, ship it, and let the business point at what comes next.
The bottleneck moves. Plan for the problem your automation creates, not only the one it solves.
Someone has to own it. A process nobody runs still needs a person accountable for its output, and a monitor that looks at the business outcome rather than the server.
Meet people where they are. If the workforce lives on WhatsApp, the software lives on WhatsApp. If the data cannot leave the building, the AI runs inside it.
How it gets built now
The part that has changed most is speed. A written brief on a Monday can be a working build on the Tuesday. Every project now starts on an AI-driven build pipeline, with our own standards and components reused across clients, and a person takes over wherever the machine stops. Before the first line of code there is a scoped and signed document. Before handover there is a break session, where the team tries to destroy the build the way real users will, and security testing on what ships.
Weeks rather than quarters is not a slogan. It is the difference between a business waiting a year for a system and a business using one before the next month-end.
But the build is the easy half, and here we agree entirely with our colleagues on the change side. Shipping the software is where the real work starts. The exceptions, the ownership, the person who keeps the old spreadsheet to be safe. That is why we stay on after a build rather than handing over and stepping away, and why the technology and the change discipline sit under one roof.
Find the inefficiency. Build the fix. Make it work. Then do it again.
If you would like help with Doing Change Better, we would welcome the conversation. Get in touch with us.
