In 2009, I was at M&S running a programme to introduce in-store collection.
Today, click and collect feels completely normal. Back then, it didn’t.
We were trying to connect ecommerce, distribution centres, stores, carriers, customer service, payments and the Amazon platform into a single customer proposition.
The customer experience sounded simple:
- Order online.
- Choose a store.
- Collect your parcel.
The operation behind it was anything but.
I remember spending an enormous amount of time in Excel, mapping out dependencies and interdependencies across the programme:
- If volumes moved, what happened to capacity in the distribution centre?
- If the carrier missed its delivery window, what happened in store?
- Could stores physically handle the volume of parcels?
- Would the solution that worked for five stores still work for fifty?
- Would it still work at peak?
We had assumptions around volumes, capacity, transport, systems, training, store processes, infrastructure and service recovery.
And every assumption seemed to connect to another one.
At the time, I thought I was managing a complicated programme.
Looking back now, I think I misunderstood the complexity I was actually dealing with.
I was managing decisions. More importantly, I was managing decisions that didn’t really exist inside any single system.
The systems were there: Order-management systems. Warehouse systems. Payment systems. Carrier systems. Store systems. Customer-service systems.
Each had a very clear role.
But the decisions that made the whole proposition work sat somewhere in between them.
- How much capacity did we need?
- Where was the risk?
- What happened if demand changed?
- Could the operation still deliver the customer promise?
- What needed to change if one assumption stopped being true?
Those decisions lived in project meetings, spreadsheets, operating processes, governance forums and, critically, in the experience of the people around the table.
I didn’t think that was unusual.
I thought that was simply how complex operations worked.
We launched.
It held up – mostly because a relatively small number of people involved in the programme understood how all the different parts fitted together and could re-cut the plan when something didn’t go as expected.
The programme succeeded.
But much of the capability that made it succeed never made it into a system.
It remained in the heads of the people who had built it.
Fast-forward seventeen years
A lot has changed.
The technology inside operations is dramatically more sophisticated.
WMS platforms have become incredibly capable. Workforce systems have improved. Automation and robotics have transformed warehouse execution. Organisations have access to vastly more data and significantly better analytical tools.
And yet, as we have spent the last few years working deeply with warehouse and logistics operations, I have found myself looking at something remarkably familiar.
Not the same systems.
Not the same processes.
But the same structural problem.
There is often a gap between the systems that understand what is happening and the systems that execute what has been decided.
And in that gap sits a human being making a judgement. Very often with a spreadsheet.
Take labour planning.
An operation may have a forecast, historical performance, productivity data, labour availability, shifts, skills, service commitments and operational constraints.
It may have a WMS capable of orchestrating thousands of tasks, a workforce-management platform capable of scheduling thousands of people and millions of pounds of automation inside the building.
But someone still has to answer a very basic question:
Given what we expect to happen, what labour do we actually need?
And then, when reality changes:
What should we do instead?
That decision frequently sits outside the systems designed to execute it.
This is the part of modern operations I think we have underestimated.
We have digitised execution far faster than we have digitised judgement
For several decades, operational technology has rightly focused on execution.
Moving orders more efficiently. Scheduling people. Managing inventory. Automating repetitive activities. Tracking performance.
The execution stack we have built is extraordinarily capable.
But increasingly I think we are reaching a point where adding more execution capability alone creates diminishing returns.
Because perfect execution of the wrong decision is still the wrong outcome.
A warehouse can execute a labour plan brilliantly.
But if the organisation committed too much labour three weeks earlier, the cost has already been incurred.
If it committed too little, the service risk may already exist.
If the wrong assumption was made about productivity, execution systems can only work with the capacity they have been given.
The economically important decision happened earlier.
That is a fundamentally different problem.
Operations has moved closer to the customer
This matters because operations is no longer simply a back-office capability.
For many businesses, the operation is now an integral part of the customer proposition.
In 2009, the promise we were trying to create at M&S — order online and collect conveniently from your local store — depended on dozens of operational decisions being made correctly behind the scenes.
Today, that relationship is even tighter.
Customers expect next-day delivery, same-day fulfilment, accurate availability and increasingly precise promises.
There is much less buffer between an operational decision and the customer experiencing its consequence.
At the same time, demand is more volatile, labour is harder and more expensive to flex, margins are under pressure and operations are becoming more complex as people work alongside automation.
That makes the quality of the decision increasingly important.
Automation does not remove the decision
There is a tendency to talk about automation as if the long-term direction of operations is simply towards removing labour.
That isn’t what I see. As warehouses automate, labour doesn’t disappear.
The decision changes.
The operation now has to understand the capacity of the automation, the workload entering the building, the productivity being achieved, the labour available, the service commitment and where people are still required around the automated environment.
I’ve sat with operations directors who can tell you their automation’s rated throughput to the unit, and in the same breath admit they’re still estimating how many people they will need around it on a Tuesday afternoon six weeks from now.
The machine’s capacity can be precisely known. The labour decision sitting next to it often isn’t. And that gap becomes more important as automation increases.
Because now the quality of that decision isn’t only determining the productivity of the people in the operation. It is helping determine whether millions of pounds of capital investment is being used effectively too.
So the question changes:
Given what the automation can do, the workload coming in, the labour available and the promise we’ve made to the customer, how much labour do we actually need around the automation, where do we need it, and when?
Automation can execute exceptionally well.
But something still needs to decide what the operation should do.
The real constraint is often not data
For a long time, the answer to better operational decision-making was assumed to be more data.
I’m not convinced that is the constraint anymore.
Most large organisations already have huge amounts of operational data.
The harder problem is turning it into a connected decision.
What does the forecast mean for capacity? What does capacity mean for labour? What happens to cost if productivity moves? What happens to service if the plan changes?
These questions cut across traditional system and functional boundaries.
Historically, making sense of them has been the job of experienced operational people.
And those people are incredibly good at it.
The problem is that their judgement is difficult to scale.
It lives in experience, instinct, relationships between teams and knowledge built over many years.
That takes me back to the M&S programme.
The people in that room weren’t valuable simply because they knew their individual part of the process.
They understood the consequences of their part on somebody else’s.
They could see the dependencies.
They knew when an assumption had changed enough to matter.
They knew which constraint could be worked around and which couldn’t.
The organisation had built an extraordinary amount of operational intelligence.
It just hadn’t built a way of retaining and scaling it.
That is not really a technology problem.
It is an organisational capability problem.
This is where I think AI becomes genuinely important
There is a lot of noise around AI in operations.
The use of AI that interests me most is not another chatbot sitting over an operational dashboard.
It is the ability to begin codifying and augmenting operational judgement.
To continuously bring together demand, capacity, performance, availability and constraints.
To understand the trade-offs.
To recommend what should happen next.
To explain why.
And to learn from the outcome.
Not to remove the operator.
Quite the opposite.
The opportunity is to give operators a capability that previously existed only through years of experience, fragmented information and manual analysis.
To allow the machine to deal with the complexity, while the human retains the judgement, context and accountability that matter.
We went from systems of record.
To systems of execution.
The next layer is systems that help organisations decide.
The timing of a decision matters as much as the decision itself
One of the things we have learned in labour is that the same decision looks completely different depending on when you make it.
Eight weeks out, uncertainty is high.
But the organisation still has choices.
It can recruit. Change agency commitments. Move work. Adjust shifts. Challenge assumptions.
A few hours before execution, uncertainty is much lower.
But most of those options have disappeared.
The labour may already have been bought. The shift committed. The capacity fixed. The service risk embedded.
That creates a simple but important principle:
The earlier you can make a high-quality decision, the more valuable that decision becomes.
This is why simply making execution faster is no longer enough.
The opportunity is to move intelligence earlier.
Before cost has been committed.
Before service risk has become unavoidable.
Before the only remaining option is to firefight.
Making operational judgement repeatable
We have spent decades making operational execution more repeatable.
The next challenge is making operational judgement repeatable in the same way.
Can an organisation capture the reasoning of its best operators, not just their output?
Can it connect decisions across functions and planning horizons, rather than optimising each in isolation?
Can it understand the consequence of a decision before committing to it?
Can it recognise when reality has changed enough that the decision needs revisiting?
And can that capability be deployed across an entire network, rather than living with individual people at individual sites?
I think the organisations that answer yes to those questions will have a genuinely different kind of operational advantage.
Not because they execute faster.
Because they decide better.
And because that capability stays in the business after the person who built it has moved on.
We now call part of this Labour Decision Intelligence
Labour is where we chose to start because it exposes the problem so clearly.
It’s expensive. It’s constrained. It has to be committed ahead of execution.
And getting it wrong shows up quickly in both cost and service.
There’s nowhere for a bad labour decision to hide.
Labour Decision Intelligence is our name for the capability of continuously turning operational demand, capacity, performance and constraints into better labour decisions across different planning horizons.
But the bigger idea isn’t really the terminology.
It’s recognising that there’s a layer of operational capability that has historically been largely invisible.
The decision layer.
In 2009, I was trying to manage it with spreadsheets, meetings, governance and a lot of very experienced people.
I just didn’t have a name for it.
Nearly seventeen years later, I think we finally have both the technology and the need to treat that capability differently.
The last generation of operational technology transformed execution.
I believe the next competitive advantage will come from transforming the decisions that happen before it.