The government's bill keeps rising. Your overheads don't have to.
Another line on the P&L you didn't choose
If you run a small or mid-sized business in the UK, you didn't need a budget headline to feel the squeeze. It's been building for a while: minimum wage steps, pension auto-enrolment, compliance paperwork that grows every year, business rates that bear no relation to your actual month, supply chains that wobble, and customers who expect Amazon speed on a wholesaler's margin.
The latest Employers' National Insurance increase is just the most recent item on a list that keeps getting longer. Whether you employ five people or fifty, the direction of travel is the same: the cost of carrying staff went up again, and nobody asked whether your pricing, your processes, or your systems got any better to absorb it.
We're not writing a political piece. Westminster will do what Westminster does. What we're interested in is the lever you do control: how much of your payroll is spent on work that actually needs a human, versus work that's only human because nobody fixed the plumbing.
The overhead hiding in plain sight
When owners talk about overheads, they usually mean rent, stock, energy, software subscriptions, the van fleet. All real. All worth watching.
But the biggest overhead in most of the businesses we sit with isn't on a supplier invoice. It's people doing repetitive work because the systems don't talk to each other.
Same pattern, different postcode:
- Copying order details from email into the ERP
- Re-keying invoice lines from PDFs into Xero
- Pulling status from three tabs to answer "where's my delivery?"
- Building the same report every Monday because nothing auto-reconciles
- Chasing approvals that live in someone's inbox instead of a workflow
- Emailing partners the latest price list, then rebuilding a CSV when stock changes overnight — your team acting as the human API for every retailer or reseller you supply
Nobody hired your best people for that. They're doing it because "that's how we've always done it" survived the last three cost crises unchanged.
We wrote about this directly in If what you're doing is repetitive, there is a better way. The headline still holds. What's changed since then is the urgency: when every employee costs more to carry, you can't afford to burn their hours on work a machine should have taken years ago.
Modernisation isn't a luxury project anymore
"Digital transformation" used to sound like something for companies with a transformation budget and a consulting firm on retainer. For a lot of UK operators, it sounded optional — something for when things settle down.
Things aren't settling down. The tax and regulatory environment is telling you, bluntly, that running the same operation with the same manual overhead is a bet you will lose slowly.
Modernisation, in the sense we mean it, isn't a rip-and-replace of everything you use. It's usually narrower and more practical:
- Find the repetitive load — where skilled people act as the glue between systems
- Decide what "good" looks like — data in the right place, exceptions in front of a human, audit trail intact
- Build the missing layer — often a bespoke application that sits between the tools you already pay for, not instead of them
- Add AI where judgement is pattern-matching — reading documents, drafting first-pass replies, flagging anomalies — with humans on the sharp edges
That's not a five-year programme. Done properly, it's phased: fix the worst bottleneck first, prove the saving, fund the next piece from the time you got back.
And sometimes step three isn't bespoke at all — it's buying a focused tool for a problem that's already well understood. More on that below.
Why now is different: code got cheap
For a long time, "bespoke software" sounded expensive for a reason. It was. Custom build meant months of specification, development, and painful change requests every time the business shifted.
That economics has moved. We wrote about it in Agentic development is making code cheap: AI-assisted, agentic workflows have compressed the cost of building, fixing, and iterating on software. With proper guardrails — tests, staging, human review on money and permissions — the implementation is increasingly disposable; the specification is the asset.
For a small business owner, the practical translation is this:
- A well-scoped internal tool that used to feel like a £80k gamble may now be a £15–30k phase with a clear ROI
- Changes when HMRC, a supplier, or your own process shifts aren't month-long ordeals — they're days, sometimes hours
- The reason to say no to bespoke is no longer "we can't afford the code". The reason to say yes is still "we've understood the work first"
The risk hasn't disappeared. Building the wrong thing quickly is still building the wrong thing. That's why discovery before you quote matters more, not less, when generation is cheap.
What a well-designed, AI-enabled application actually does
Strip away the buzzwords. In the businesses we work with, a good bespoke layer usually does four jobs:
1. Moves data so people don't have to
Integrations with the systems you already own — CRM, accounting, warehouse, portal — so the same fact isn't typed three times. We've covered what that involves in plain language in A business owner's guide to APIs.
2. Handles the predictable 80%
Rules-based routing, scheduled reconciliations, standard acknowledgements, status updates that follow a template. The stuff that makes a Tuesday feel like a Tuesday.
3. Uses AI on the fuzzy edges — carefully
Reading a PO and proposing line items for review. Summarising a thread before a human picks up the phone. Flagging a margin that looks wrong. Draft, don't decide — especially where a customer will quote you back in an email.
4. Leaves exceptions with people who can resolve them
The awkward customer, the one-off credit note, the supplier who's being difficult. Software should make those moments faster to handle, not try to eliminate them.
That's the shape of overhead reduction that sticks. Not "replace your team with AI". Stop paying your team like robots for robot work, so you can afford to keep them — and afford the NI bill — while the business grows.
When the answer is Tandem, not a six-month build
Not every overhead problem needs a custom project. Some patterns are so common — and so expensive in staff time — that the right move is a product built specifically for them.
If you're a manufacturer, importer, or distributor still keeping B2B partners aligned through shared folders, email threads, and parallel spreadsheets, you're paying skilled wages to do unskilled data distribution. Partners chase you for updates. Someone hand-builds a CSV per account. Product images live in Dropbox with no stable link to SKUs. A retailer lists something you discontinued last month because their copy of your catalog was already stale.
That's the workflow we built Tandem for — a trading name of InsideData, aimed squarely at this problem. One shared catalog for products, images, trade pricing, and stock. Partners log into a private portal at their contract prices, export CSV when they need a file, or connect via API and webhooks so their Shopify, WooCommerce, or ERP stays in sync with what you've actually published. Governed publishing means partners only see complete, live products — not the half-finished draft someone forgot to hide.
It's the buy path for a repetitive overhead that wholesaler and distribution teams know too well — and it lands in days, not quarters. Starter is free while you prove the model; you don't need developers for day-to-day catalog work, only when a partner wants a deeper integration.
Tandem doesn't replace bespoke work when your edge is in how you operate — unusual fulfilment rules, a proprietary quoting engine, AI on your inbound POs, something that doesn't look like anyone else's business. It does remove a whole category of manual glue work that has no business being on someone's desk in 2026. Same philosophy as this article: update once, stop being the human API, free your team for the work that actually needs them.
The maths owners actually run
Nobody signs off on software because it sounds modern. They sign off because the alternative hurts more.
A rough pattern we see:
| Before | After |
|---|---|
| Two people spend half their week on data bridge work | One person handles exceptions; the rest is automated |
| You'd need three new hires to absorb next year's volume | Same team absorbs it; software cost < one salary |
| Errors from manual re-keying cause credits and firefighting | Single source of truth; audit trail when something's wrong |
| Owner is the integration layer — everything routes through you | Workflows run; you see dashboards instead of forwarding emails |
When employer costs rise, the comparison isn't "software vs nothing". It's "software vs another hire" — or software vs burning out the people you've got, which is the option too many owners take until someone hands in their notice.
"But we're not a tech company"
Good. Neither are most of our clients.
You don't need an CTO. You need someone who will stand next to your team, watch how the work actually happens, and build software that fits the Tuesday morning — not a demo that looks finished on a Sunday afternoon.
The build vs buy question still applies: if focused SaaS solves it, buy. Tandem is our own example — product data for B2B suppliers and partners, not a generic PIM with a two-year implementation. Bespoke earns its place when your edge is in how you operate, when the integration tax has become unbearable, or when off-the-shelf forces you to run like everyone else when you can't afford to.
For a growing number of UK operators in 2026, at least one of those is true — and the cost of building the fix has come down at exactly the moment the cost of standing still has gone up.
What to do this quarter
You don't need a transformation roadmap. You need one honest week of attention:
- Pick the most expensive repetitive workflow — the one that makes your best people sigh when it lands in their inbox
- Count it — hours per week, error rate, what happens when that person is on holiday
- Ask whether the rules are mostly predictable — if yes, it's a candidate for automation
- Talk to someone who'll tell you the truth — including "don't build this yet" or "fix X in your existing system first"
The NI increase isn't the first blow, and it won't be the last. The businesses that come through aren't the ones that got lucky on margins. They're the ones that stopped treating manual overhead as fixed — and realised that, for the first time in a long while, custom software isn't just for companies with deep pockets.
If you're a B2B supplier drowning in catalog email and spreadsheet exports, try Tandem — it's free to start and built for exactly that overhead.
If your problem is broader — integrations, AI on inbound work, something that doesn't fit a catalog hub — say hello to InsideData. We'll look at the work before we talk about code, and we'll tell you honestly whether you need us, Tandem, or neither yet.
InsideData
InsideData
More from the workshop
Why we spend a whole week on discovery before quoting
The two-week discovery sprint isn't a sales tactic. It's the only honest way to estimate a bespoke software pr...
Build vs buy: the question we get asked the most
After 20 years of these conversations, here's the honest framework we use to help businesses decide whether to...
AI in customer services: a draft on every reply
How AI changes the shape of a customer-services inbox: reading inbound messages, looking up the customer and o...
Let’s talk about your back office
Start with a free 30-minute discovery call. No slides, no sales pitch; just a real conversation about where your business is and where it could be.