Most systems that fail do not fail loudly. Nobody cancels them. They go live, they get used for a month, and then the spreadsheet comes back out — first for the awkward cases, then for everything.

The reason is almost never a missing feature. It is that putting something into the system takes longer than remembering it.

The twenty-second rule

A salesperson records a sale on a phone, between two calls, standing in a doorway. If it takes twenty seconds, it happens there. If it takes two minutes, it waits.

What waits until the evening gets entered from memory. At that point you have paid for a system and kept the original problem: the numbers are memory again, just stored somewhere more expensive.

One sale, entered on a phone
  1. Partner
  2. Product
  3. Quantity
  4. Price
  5. Save
20 s

A typical entry, with room left

Five steps is the whole transaction. Every field beyond these five has to earn its place against the chance that the entry stops happening at the counter.

The same sale, two forms

A form built from the database

~2 minutes · entered this evening

  1. Sign in again
  2. Scroll an alphabetical list of 900 partners
  3. Type the product name exactly as it was imported
  4. Fill eleven fields, four of them required and unknown
  5. Save, then check whether it actually saved
  6. Entered from memory at the end of the day

A form built from the job

Under 20 seconds · entered on the spot

  1. Recent partners first: one tap
  2. Three letters find the product
  3. Quantity, price, save

The dashed steps are the ones that push an entry to the end of the day.

Where the seconds go

  1. Signing in. A session that expires every hour costs more than it protects. On a shop floor, the login screen is where a system loses its first users, and they do not come back on their own.

  2. Finding the customer. An alphabetical list of 900 partners is not a picker. The five you dealt with this week are the ones you want, and they belong at the top of the screen.

  3. Finding the product. If search needs the exact catalogue name, people will type the exact name wrong. Search has to match how the warehouse talks, not how the import file was written.

  4. Typing the numbers. A keyboard that offers letters in a quantity field costs three seconds and one mistake. The field should know what kind of thing it holds.

  5. Being sure it saved. An entry with no clear confirmation gets done twice. The duplicate costs far more than the second you saved by leaving the confirmation out.

Fewer fields, better data

The instinct is to collect everything while you have the person’s attention. It backfires in a specific way.

A required field that someone cannot answer does not stay empty. It gets filled with whatever passes validation: a zero, a dash, today’s date, the first option in the list. The field is now worse than missing. It is wrong, and it looks exactly like data.

Ask for what is known at that moment. The rest can be added later by someone who knows it, or worked out from what you already hold.

Free text is where data goes to die

One product written six ways is six products to every report you will ever run. No amount of training fixes this, because the person typing is not doing anything wrong — they are typing what the thing is called where they are standing.

The fix is a catalogue with no free-text escape. A supplier’s own name for a product is mapped to yours once and remembered on every purchase after that. The person entering it never thinks about it again, and the report is right because it cannot be anything else.

Test it where the work happens

A form tested at a desk, on a laptop, with good light and a fast connection, has not been tested.

Test it standing up, on the phone the person actually owns rather than the newest one in the office, with one hand because the other is holding something. In a warehouse where the light is bad and the signal drops in the far corner. With the gloves on.

Almost everything that makes an entry slow in real life only appears there, and none of it appears in a requirements document.

Measure it, do not assume it

Speed is a feature, and like any feature it rots unless something defends it.

In the trading system we built for Techizat, an automated test enters a sale on an emulated phone on every release and times it against a hard limit. If a change pushes the entry over that limit, the build fails. Nobody has to notice it, argue about it, or remember that it mattered.

That is the part most teams skip, and it is why a fast form built in month one is a slow form by month nine. Every new requirement adds one more field, and every one of them is individually reasonable.

When this does not matter

Not every screen needs to be fast, and treating them all the same wastes the effort.

Work done sitting down, once a month, by someone whose job it is — closing a period, running payroll, reconciling a bank statement — can take as long as it needs to. Those screens should be careful rather than quick. Shaving seconds off them buys nothing.

The rule applies to the handful of actions that happen constantly. In most companies there are three or four, and they are the ones that decide whether the system gets used at all. Find those first.

If you already have a system and are not sure which kind you have, there is a cheap way to find out. Stand behind someone while they enter a real record, hold a stopwatch, and say nothing. The number will tell you most of it. The apologies they make while doing it will tell you the rest.

Projects behind this article

What we build in this area

  • ERP & business management systems

    ERP for trade, distribution and service businesses — sales, stock, finance, payroll and approvals, connected to 1C and e-qaimə.

  • SaaS platform development

    Multi-tenant SaaS products engineered end to end — tenancy, billing, mobile PWAs, AI features and the console to run it all.

All articles