The work between the systems. Own the layer.

Introduction · The automation context

Automation is a global force. The question is where it pays.

Advances in robotics, artificial intelligence and machine learning are opening a new age of automation, as machines match or outperform people across a widening range of work, including cognitive tasks. The McKinsey Global Institute’s study of more than 2,000 work activities across 800 occupations set the scale of the shift.

The point is not that jobs vanish. Automation moves activities rather than whole roles: more occupations change than disappear, much as earlier shifts reshaped agriculture and manufacturing. Value accrues to the organisations that redesign the work around the new tools.

Five factors set the pace of adoption

Technical feasibility · Cost of solutions · Labour-market dynamics · Economic benefits · Regulatory and social acceptance

For a large business, that redesign rarely happens inside any one system. It happens in the work between them, which is where this paper begins.

01 · The problem

Most manual work does not happen inside your systems. It happens between them.

A large business has an ERP, a payroll system, a CRM and many other tools. Each one works. The trouble is the work that moves between them.

A driver sends a photo of a signed delivery note to a WhatsApp group. Someone downloads it, opens the invoice, compares the two and records the difference in a spreadsheet. A buyer retypes a supplier’s quote. A regional manager keeps the real outlet list in a workbook because the master data is wrong.

None of this is on the system diagram. It still costs salaried hours, errors found a month late, and numbers nobody fully trusts. We call it seam work. Bain reached the same view this month: the opportunity lies in the expensive human labor that connects systems, not in the systems themselves.

Two things have changed

AI can read. It can pull data from a photographed delivery note, a supplier PDF, an email or a voice note. When every result is checked against records you already hold, it is accurate enough to run a process on.

Software is cheap to build. What took a year now takes weeks. McKinsey found that nearly a third of companies have dropped a software purchase because they could build it themselves.

Keep your core systems. Buy what is the same in every company. Own the layer that connects your systems and runs your processes.

02 · What we have built

Six businesses, one pattern.

Clients are not named. Each platform started with one job and sits over the systems the client already had.

What they have in common

• None replaced the core system. The ERP, payroll and bank stayed. The platform reads from them and writes back through them.
• AI reads and drafts. Code decides. Money, prices and approvals follow fixed rules. When the AI is unsure, a person decides.
• Staff use what they already have. Usually WhatsApp.
• Every process has an owner. Software runs it. A named person answers for it.
• The first build leads to the next. The second module is cheaper because the data and connections already exist.

03 · Why SaaS falls short

SaaS is built for departments. Your business runs across them.

• Handoffs belong to nobody. Sales, warehouse, logistics and finance each buy their own tool. No vendor owns the steps between them, so people and spreadsheets do.
• You adopt the vendor's view of your business. When its data model does not fit, your staff fill the gap by hand.
• Per-seat pricing shuts out the frontline. Drivers, merchandisers and plant staff rarely get a licence, so their work stays on paper and WhatsApp.
• AI needs joined-up data. Settling a short-delivery claim needs the order, delivery note, invoice and credit terms together. Each vendor's AI sees only its own part.
• Building is no longer slow. A new module takes weeks, on your timetable, not the vendor's.

SaaS still makes sense for the general ledger, payroll calculation and email. They are the same everywhere, and rebuilding them would be waste.

Owning has to mean owning. Gartner expects 70 percent of enterprises to abandon AI systems built by vendors’ embedded engineers by 2028, because they cannot afford to run or change them. The code and data should sit in your accounts, and your team should be able to change them.

What an owned platform is

A layer over your existing systems, not a replacement for them.

04 · How we build it

Map it. Model the data. Then automate.

Most failed automation projects skip a step. They automate a process nobody mapped, on data nobody cleaned, and hand it to nobody in particular. We work in this order.

The technology is the smaller part. BCG found that companies getting value from AI put 70 percent of their resources into people and processes. That is why the change work runs through every step, not only at the end.

Where we would look first in a large distribution business

We do not yet know which of these apply to you. Mapping would tell us.

• Proof of delivery and claims. Delivery notes, short deliveries and returns matched to orders and invoices automatically.
• Trade promotions. Customer claims checked against agreed terms before any credit note is raised.
• Outlet execution. Rep and merchandiser checks captured on WhatsApp with photos, and turned into data by outlet and product.
• Coolers and assets. Placement, service and recovery tied to the outlet record.
• Frontline staff. Payslips, leave and incident reports on WhatsApp for drivers and plant staff.

A first step

A four-to-six-week mapping of one or two processes you choose. You get a map of the work, a ranked list of what to automate, and a costed first module. Then you decide.

If you would like help with Doing Change Better, we would welcome the conversation.