Blog · Software
Building business software: 18 keys to success
Business software succeeds on understanding, not code. Follow your process, build it to last, and choose a partner who stays. 18 keys, updated for AI.
Published Updated
Business software succeeds or fails on understanding, not code. It has to follow how you actually work. The people building it have to understand the problem well enough to challenge you. And someone who understands it has to keep it secure, supported and improving long after launch.
We first wrote these keys in 2020. AI-assisted development has changed a lot since then: a working first version that would once have taken weeks can now often be ready in days. What it hasn’t changed is the hard part. When code is quicker to produce, knowing what to build, and building it so it lasts, matter more than ever.
The keys apply whether your software is built in-house or by a partner. The original list had 21. A few said the same thing twice, so we’ve merged them.
1. Business software should follow your process, not the other way round
This is the main reason to build custom software at all. Off-the-shelf products ask you to work their way. Software built for you should fit the way your team already works, or the better way you’ve decided to work.
If you’re still deciding whether to build or buy, start with custom software vs off-the-shelf.
2. Build what works
Before automating a process, have the people who do the work run it by hand for a while and confirm it actually works.
Once a project starts, it’s tempting to redesign processes as you automate them. Resist using software to discover that a process you imagined doesn’t work in practice. It’s much cheaper to find that out with a checklist, and to keep improving a process once it’s at least partly automated.
AI has changed one part of this. A rough working prototype is now cheap enough to be a fair way of testing a process. Treat it as exactly that: a test. A prototype that works in a demo is a long way from software you can rely on (see key 9).
3. Breadth beats brilliance
Hundreds of decisions get taken while software is built. The best ones come from people who understand the consequences from top to bottom: what the firm is trying to achieve, what frustrates the users, how the data is structured, how it connects to your other systems, how it’s secured and what people see on screen.
A brilliant specialist who only sees one layer can’t make those calls well, however talented they are.
AI has sharpened this. AI coding tools can now write a lot of the code a developer used to type by hand, so typing code matters less. Understanding the whole system well enough to judge what the AI has produced, and to spot when a plausible answer is wrong, matters more.
That understanding shouldn’t live only in one person’s head. Ask how it’s written down, how the software is documented and tested, and what happens if the person who built it is unavailable.
4. Complicated is easy, simple is harder
The best business software is simple to use, and the code, connections and data underneath are simple too. Getting there takes a great deal of thought.
Building something complicated is relatively easy. The result is often inflexible, creates duplicate or conflicting data, or simply goes unused.
AI makes this easier to get wrong. It can generate a lot of code very quickly, so complexity can pile up faster than it used to. Deciding what not to build is a bigger part of the job than it was.
5. Allow access, at all levels
Replacing the way something is done means asking why it’s done that way: why this step, this screen, this spreadsheet column. The people doing the work aren’t always best placed to ask those questions, but they’re usually best placed to answer them.
Give whoever builds your software access to the people who’ll use it. That means staff at every level, and clients or suppliers if they’ll use it too. In an accountancy firm, the person who chases clients for their records every January knows things about your workflow that the partners may not.
6. Use new tools, but don’t be the test case
Good builders keep looking for tools that make them faster and cut running costs. Managed cloud services, where you pay for what you use rather than for servers someone has to patch, are one example. AI coding tools are the latest.
They’re a real change, but not a magic one. The 2025 DORA report from Google Cloud, based on survey responses from nearly 5,000 technology professionals, found that AI “doesn’t fix a team; it amplifies what’s already there”. Strong teams got better. Struggling teams found their existing problems intensified. It also found AI adoption still went hand in hand with less stable software delivery.
So the rule from 2020 still holds: new technology should be proven on the builder’s own systems before it reaches yours. Ask how they use AI tools, how they check what those tools produce, and where your data goes when they do.
7. Change is inevitable, embrace it
Software, like the processes it supports, is never finished. Rules change, clients expect more, and the other systems your software connects to get updated. With very few exceptions, it can and should keep improving.
That means choosing someone you can work with not just for the build but for years afterwards. A builder who hands over the software and moves on leaves you with something nobody is looking after.
8. You can’t predict the future, so plan lightly
Six months is a long time in business, and longer in software. Fixing every requirement up front, then building for months before anyone sees anything, rarely survives contact with real users.
A little planning goes a long way. Plan enough to agree the problem, what success looks like and what to build first. Then build in short cycles, look at working software regularly and adjust as you learn. The better your builder, the more decisions you can safely leave until you know more.
AI has shortened those cycles. Use that: steer by looking at real software rather than by reading documents about it.
9. Good architecture beats a fancy interface
This matters more in business software than on a public website. Business software is usually more complex, it does more, and its users often have no choice but to use it.
Good design matters, but don’t be won over by a polished screen. What you can’t see decides how much the software costs you over its life: how the data is structured, how it connects to your other systems and how easily it can change.
That’s worth repeating now that AI tools can produce a convincing-looking app in an afternoon. A good demo tells you very little about what’s underneath. Choose a builder you trust not to take expensive, inflexible shortcuts. It’ll cost you less in the long run.
10. Start with the problem, not the answer
Don’t trust a builder who asks what you want and doesn’t challenge the answer. You need someone who asks the right questions and steers you towards what will actually work, even when it isn’t what you came in with.
A good partner shouldn’t only fix the problems you’ve already spotted. They should know your firm well enough to suggest the feature, the connection to your practice software or the automation you hadn’t thought of.
When building gets cheaper, the expensive mistake is building the wrong thing well.
This is where AI has moved the scarce skill. Capacity to write code is no longer the bottleneck. Understanding the problem is, and it’s the most valuable thing a builder brings.
11. Then start small
Build a minimum viable product (MVP) first: only the core features that make the software useful, so you can learn from real users before investing further. Keep some background steps manual at this stage (see key 2). That limits what it costs you when an assumption turns out to be wrong.
When extra features are cheap to add, the temptation is to build them all at once. Resist it. Starting small was never only about saving build time. It’s about learning what your people actually need before you commit to it.
12. Big bang roll-outs can be just that
Roll out new systems cautiously. Test in the real world, with real users, just not with all of them at once. Start with one team or a handful of clients, fix what they find, then widen it.
And pick your moment. For an accountancy or tax firm, the run-up to 31st January is not the time to change how anyone works.
13. Testing is hard, but essential
Part of your job as the customer, whether your builder is in-house or a partner, is to test what’s been built. It’s your chance to check that the trust you put in them was well placed. Don’t skimp on it.
Test with the awkward cases your team knows about, not just the tidy ones. There shouldn’t be big surprises if you’ve been involved throughout, but there will always be bugs.
14. There will always be bugs, so insist on automated tests
Every piece of software of any size has bugs. That’s as true of code an AI wrote as of code a person wrote. AI produces plausible code quickly, and plausible isn’t the same as correct.
Automated tests are checks written alongside the software that run every time it changes, confirming it still does what it should. They take time to write, and they work: they catch bugs early, before they become slower and costlier to diagnose and fix. AI has made tests quicker to write too, so there’s less excuse than ever for skipping them.
Ask any builder how their tests work and how often they run. If the answer is vague, treat that as a warning.
15. Better to have support and not need it
Even after a smooth launch, there will be:
- edge cases to handle
- user questions to answer
- issues to diagnose
- security patches to apply
- updates to the code libraries the software depends on
- infrastructure to look after
Your IT provider may cover some of that, but rarely all of it, especially for bespoke software. Agree how support works before launch, not after the first problem.
That split is why we build and run software on the IT we manage: the person you speak to understands both the software and the environment it runs in.
16. Not all support is equal
Make sure the people who built your software can support it, at least while it beds in. Don’t be passed to a support desk that doesn’t understand the software or how you work.
If something is stopping your clients reaching their documents the week before a filing deadline, you want someone who knows the system working on it straight away, not a ticket waiting in a queue.
17. Security is fundamental
You only get one chance at this. IBM’s 2026 Cost of a Data Breach Report put the average UK breach at £3.13 million. Whatever the figure would be for you, the harder cost to recover is your clients’ trust.
If whoever builds your software doesn’t understand application security thoroughly, find someone who does. And don’t let them build their own sign-in. On Microsoft 365, staff should sign in to business software with their existing Microsoft account, so multi-factor authentication applies and a leaver loses access once their account is disabled.
AI adds two questions to ask. Where does client data go when the software, or the people building it, use an AI service? And who has reviewed the security of code an AI wrote? Get clear answers to both before anything touches client data.
18. Communication, communication, communication
This is hard to get right, and critical. No-one wants a builder who needs constant guidance, but you should always know where things stand. Finding that balance is your responsibility as much as theirs.
Agree how and when you want to hear from them: written or spoken, as it happens or in a regular summary. Set up regular check-ins and don’t let them slip.
Faster building makes this more important, not less. When software changes week to week, a short, regular look at the working software tells you more than any status report.
In summary
Most of these keys come back to the same point. Business software succeeds when the people building it understand the problem, build it to last and stay to run it.
AI has made the building faster. It hasn’t changed what makes software worth having. Choose a partner you can trust, and who trusts you.
If you’re weighing up software for your firm, a Technology Review is a good place to start. We look at your setup, your security and the workflows that take up your team’s time, and tell you plainly what’s worth building.
Know someone who’d find this useful?