I have been called in to rescue enough technology programmes to have noticed a pattern in when they were commissioned.
It is almost never a technology decision. It is a moment of accumulated frustration, usually after a bad quarter, where somebody senior says the systems are holding us back and nobody disagrees. Six weeks later there is a scope, a supplier, an eighteen month plan and a number with six figures in it.
The systems probably are holding the business back. That part is usually true. What follows tends not to work, and the reason has very little to do with the software.
01Why do big programmes fail in businesses this size?
Three reasons, and they compound.
The business outruns the plan. A company growing at twenty per cent a year is a materially different organisation eighteen months later. More people, different customers, a changed operating model. A programme designed for the business as it was at kickoff delivers into a business that has moved, which is why so many go-lives are followed immediately by a change request queue longer than the original scope.
There is nobody whose job this is. Large organisations run programmes with people who do nothing else. A £15m business does not have those people. So the work is done by the finance director, the operations manager and whoever is good with systems, on top of their existing jobs, for eighteen months. Day jobs have customers attached. The programme loses, gradually, in ways nobody reports until it is obvious.
The decision that mattered was never made. Most system replacements are commissioned to fix a process that different parts of the business perform differently and disagree about. The new system forces a single way of working, which means the disagreement has to be resolved. It is usually resolved in configuration workshops, by whoever attends, without anybody realising a significant operating decision has just been taken by default.
02What works instead?
Sequencing by reversibility. Do the things you can undo first.
Rank every candidate change by how hard it would be to reverse if it turned out to be wrong. A reporting layer over existing data is easy to reverse. Replacing the system of record for customer transactions is not. Then work from the reversible end.
This feels backwards to anyone used to programme thinking, where you establish the foundation first and build on it. In an environment where you cannot be confident about what the business needs in eighteen months, the value of learning cheaply exceeds the value of architectural tidiness. You find out what people actually do, as opposed to what they said in a workshop, and you find out for a fraction of the cost.
In practice a sequence looks like this. First, get the data out of the existing systems and into one place where it can be reported on, without changing any of them. That is usually four to eight weeks and it produces immediate value, because half the complaints about a system are actually complaints about not being able to see anything.
Second, fix the one process where the pain is measurable and the fix does not require replacing anything. Quoting, handover, invoicing, whatever your version is. Do it with the systems you have, even if the result is inelegant.
Third, and only third, replace a system, if the first two have shown that the system genuinely is the constraint. About half the time they show it is not.
Nobody has ever regretted spending six weeks finding out what people actually do before spending six figures on software to support what they said they do.
03What should you buy, and what should you leave alone?
Buy the boring middle. Finance, payroll, email, storage, the customer record. There is no competitive advantage available in any of these and no version of your business where doing them differently wins you a customer. Take the standard product, adopt its way of working, and resist configuration.
Build, or heavily customise, only where you genuinely differ from everyone else in your market. For most businesses that is one thing, occasionally two. It is usually somewhere in how work is specified, priced or delivered, and it is usually the thing customers actually mention. That is where bespoke effort earns a return.
Leave alone anything that works and is not the constraint. This is the hardest instruction to follow, because once a programme has momentum it acquires scope, and there is always a system somebody dislikes. A system being old is not a reason. A system being disliked is not a reason. Being the measurable constraint on something that matters is the only reason.
04What about the people running it?
The person who owns the change should be the person who owns the outcome it is supposed to improve.
If quoting is the target, the commercial director owns it, not IT and not a project manager. This sounds like a detail and it decides everything, because when the work competes with day jobs, only the person accountable for the result will keep choosing it.
They will need help. Buy the specialist skills you lack, and buy them by the day rather than by the programme. What you should not buy is somebody to own the outcome on your behalf, because outcomes cannot be outsourced to people who leave when the invoice is paid.
05What do you measure?
One number: elapsed time for a real transaction, end to end, measured from the customer's perspective.
How long from enquiry to quote. How long from order to delivery. How long from delivery to cash. Pick the one that matters, measure it before you start, and measure it monthly.
It cannot be gamed, it does not care what the project plan says, and it moves when something real improves. Compare that with the metrics technology work usually reports, which are milestones delivered and budget consumed, both of which can look excellent while nothing about the business has changed.
If elapsed time has not moved after three months, stop and find out why. That question, asked at month three, costs a fortnight. Asked at month fifteen, it costs the programme.
06The version that actually happens
A business I worked with had convinced itself it needed to replace its core system. Eighteen months, a large number, three suppliers shortlisted.
We spent five weeks reporting on the data it already held. That surfaced that eighty per cent of the delay everybody complained about sat in a single handover between sales and operations, where two teams maintained separate versions of the same information and reconciled them by email.
Fixing that handover took nine weeks and no new software. Elapsed time from order to delivery fell by a third. They eventually did replace the system, two years later, for genuine reasons, and the replacement went smoothly because by then they knew exactly what they needed it to do.
The programme they nearly bought would have delivered the same handover problem, on a newer platform, eighteen months later.
