28 September 2026
In short
Which processes are worth automating first in Northampton?
The processes worth automating first in Northampton are the dull, frequent ones that need no judgement, not the complicated one everybody complains about. GTX Digital counts what the job costs now before building, so the saving can be proved rather than claimed.
- Start with the dull, frequent job. Not the complicated one.
- Count what it costs now, or you will never prove it worked.
- The exceptions are the project. The normal case is the demo.
- Automating a bad process gets you a faster bad process.
- Twenty minutes a shift across forty people is the real business case.
Think this applies to your business?
We will tell you straight whether it is worth doing, what it would take, and what we would do instead if it is not.
Got a process in mind already? Book a technology review. Half a day watching a shift tells us more than any requirements document.
There is a version of automation that sells well and works badly. It starts with the most complicated process in the business, because that one is obviously the most painful. It is also full of judgement calls and exceptions, and it is the worst place to start.
Where the hours actually go
In a distribution week the expensive jobs are rarely the interesting ones.
| The job | How often | Worth automating? |
|---|---|---|
| Booking a consignment into a carrier portal by hand | Dozens a day | Yes. Dull, frequent, no judgement |
| Keying a customer order from an emailed spreadsheet | Daily, per client | Yes, and it removes errors as well as time |
| Producing the Monday stock report | Weekly | Yes, and it is usually an afternoon |
| Deciding how to split a short delivery | A few times a week | No. That is a judgement about a relationship |
| Chasing a supplier for a late ASN | Constant | Partly. Automate the chase, not the conversation |
Nobody is proud of the first three. They happen dozens of times a week and need no thinking. That is exactly why they are worth doing first.
The complicated process can wait. It gets easier after two simple ones have landed and the floor has seen that automation does not mean chaos.
Size it against peak, not against today
Counting what the job costs now is the usual advice and it is right, but in this business the average week is the wrong unit.
A process that takes twenty minutes a shift in June takes the same twenty minutes in December, except there are three times as many shifts and the people doing them started last week. The saving is not linear, and neither is the cost of not doing it.
So take the measurement twice: once now, and once from last December's figures. The second number is the one that justifies the project, because it is the week when you were paying agency rates for people doing work a system could have done.
It also tells you the deadline. Anything meant to help at peak has to be live, tested and familiar by September, which usually means starting in spring rather than in autumn. We stage workflow automation engagements around that, and the Northampton version of that page covers the shape of the work.
## The Northampton version of the problem
The recurring one here is the gap between a customer's system and yours.
A brand sends orders as a spreadsheet, because that is what their system exports. Another uses a portal you log into. A third sends a file at six in the morning that is sometimes late.
Somebody sits in each of those gaps with a keyboard. That person is the automation, and they are expensive, accurate on a good day, and unavailable in December.
The gap is where the errors live too. The same number typed twice is only the same number if nobody is interrupted, and at goods-in everybody is interrupted.
Agency staff are the real test of a process
A process that only works when experienced people run it is not a process. It is a habit, and in a business that triples its headcount for eight weeks that distinction becomes expensive.
So judge each candidate by a different question: could somebody on their second day do this correctly, with the system guiding them and nobody standing over them?
Where the answer is no, automating it as it stands will encode the tribal knowledge rather than remove the need for it. You will have built something that still requires the experienced person, and you will find that out in November.
Where the answer is yes, you have found something worth building, because the saving arrives exactly when you have least capacity to absorb the work by hand.
That reframing also kills a few candidates honestly. Some jobs genuinely need judgement, and the right answer there is to make the judgement faster, not to pretend it is a rule.
## The exceptions are the project
Here is where these projects die.
The normal case is easy. The order arrives, the fields are there, it goes in. Everyone demos that happily.
| The exception | What a good system does |
|---|---|
| An order with no valid postcode | Stops, flags it, queues it for a person |
| A SKU the client has not told you about | Accepts the order, quarantines the line, alerts |
| A duplicate file sent twice | Detects it, does nothing, says so |
| A quantity ten times the usual | Holds it for confirmation rather than picking it |
| The file arriving at 09:00 instead of 06:00 | Processes it late and reports the delay |
Build for those first. Decide what the system does when it is unsure, and the answer is nearly always: stop, flag, queue.
A system that handles ninety per cent and hands the rest over cleanly is a success. One that attempts everything and gets five wrong silently is worse than the clerk it replaced.
What should stay manual
- Anything involving a judgement about a customer relationship.
- Anything that happens twice a year. The build cost will never come back.
- Anything where the rules change every few months, unless you automate the ability to change rules rather than the rules.
- Anything where the real problem is that two departments disagree. Software will not settle that, it will encode one side of it.
What it looks like when it works
Quiet, mostly.
The orders appear without anyone typing them. The Monday report is already in the inbox on Monday. The exceptions queue has four things in it rather than four hundred, and someone clears it before the first break.
Nobody mentions it after the first month, which is the point. Good automation is boring.
Where to start this week
Go and stand at goods-in for an hour, at the busiest part of the morning. Then do the same at despatch.
Write down every time somebody types something that already exists somewhere else, and every time somebody waits for a screen. That list is usually shorter than people expect and it is almost always the project.
Then check one thing before committing: is the awkward format coming from a client who would change it if asked? A conversation sometimes removes the problem for nothing, and no supplier will suggest it.
For the jobs that need reading rather than rules, AI automation for Northampton businesses covers the difference. Where the records themselves disagree, CRM integration in Northampton comes first.
We work across logistics and 3PL and retail and e-commerce operations in the town, and the Northampton page has the wider picture. When you have the log, talk to us.
Common questions
Which process should we automate first?
The dull, frequent one that needs no judgement. Twenty minutes a shift across a large team beats a complicated job that happens twice a month.
How do we know whether it worked?
Count the job before you build: how long it takes, how often, how often it goes wrong. Without that number the review meeting is an argument about impressions.
What should stay manual?
Anything involving a judgement about a customer relationship, anything that happens twice a year, and anything where the real problem is that two departments disagree.
Next step
Start with a technology review, not a proposal.
A short, structured conversation about where your systems are costing you time, margin or growth — and what the realistic options are. No pitch deck.