Software is sold on growth and mostly delivers work. Not because the software is bad — because most of what a business buys automates a step in a process without changing what the business can do.
The distinction is worth being precise about, because it predicts which purchases pay back and which become another system to maintain.
Three things software can actually do
| Effect | What it looks like | Typical return |
|---|---|---|
| Removes a cost | Nobody retypes invoices any more | Reliable, modest, easy to measure |
| Removes a limit | You can now serve customers in a time zone you could not staff | Large, and rare |
| Removes a delay | Quotes go out in an hour instead of three days | Large, and consistently underrated |
Speed of response is the quiet one. In most competitive situations the business that replies first wins a disproportionate share, and that has nothing to do with being better — only with being faster to answer.
The question to ask before any purchase
What will we be able to do after this that we cannot do now?
If the answer is “the same thing, faster” — that is a cost saving. Fine, and worth doing, but price it as a cost saving rather than as growth, and the maths usually looks different.
If the answer is “serve a customer we currently turn away”, “answer in an hour instead of a week”, or “know something we currently guess at” — that is a limit or a delay, and those are the purchases that change a trajectory.
Where the work quietly accumulates
- Integrations. Two systems that must agree need someone to maintain that agreement, forever, and that person is rarely budgeted.
- Per-seat pricing at scale. Cheap for six people, a real line item at four hundred. Model it at the size you expect to be.
- The specialist. One person who understands the configuration, and a business that cannot change it when they leave.
- Data that must be kept clean. A system is only as useful as the discipline feeding it, and that discipline is a permanent cost.
None of these appear in a sales conversation. All of them appear in year two.
What tends to actually work for smaller businesses
- Remove the retyping. Anywhere a person moves data from one place to another by hand, there is a measurable saving with an obvious verification step.
- Answer faster. Whatever shortens the time from enquiry to response — templates, routing, a shared inbox that makes ownership obvious.
- Know one number you currently guess at. Not a dashboard of forty metrics. One number, watched, that tells you whether the month is going well.
- Write down how things are done. Unglamorous, no licence fee, and it is what lets you hire without every new person needing the founder’s attention for a month.
Two of those four cost nothing. That ratio is normal and it is worth internalising before signing anything.
The trap of buying capability you cannot use
Businesses routinely buy the enterprise tier of something because it has features the enterprise tier of a competitor has, then use 15% of it. The unused 85% is not free — it makes the interface more complex for every employee, every day.
Buy for the problem you have. The upgrade will still be available when the problem arrives, and by then you will know which features you actually need.
Common questions
How do we measure whether a tool paid off?
Decide the number before buying, and put a date in the calendar to look. Retrospective justification always finds a reason, which is precisely why it is worthless.
Is it better to build or buy?
Buy anything that is not what makes you distinct. Build only where an off-the-shelf tool would force you to work like everyone else at exactly the point where you do not.
What about AI specifically?
Same test. It removes cost where verification is cheap, and adds work where it is not. The categories that pay back are in AI tools for business.
How do we avoid ending up with fifteen tools?
A register of what exists and who owns each, plus the rule that adding one means naming the one it replaces. The four jobs framework works for company software as well as team tools.
Leave a Reply