The ROI of keyboard-first data entry: 1.8 seconds a cell, times a day

The most expensive office motion in an ERP is a hand leaving the keyboard. Each instance is too small to complain about — until it is multiplied by eight thousand cells a day.

The most expensive office motion in an ERP is not a feature. It is a hand leaving the keyboard. On multi-line forms — engineering quotes, bills of quantities, multi-item orders, bills of materials — whoever is keying repeats that motion thousands of times a day, and no report in the system records it once.

Adding a shortcut layer that reproduces the spreadsheet experience is not interface polish. It is an ergonomic change, and it can be measured in money.

1. Why nobody measures this motion

Three reasons, and all three are why it survives.

1. Each instance is too small to complain about. Leave the keyboard, travel to the next cell or to the "Add row" button, click, come back. The Keystroke-Level Model puts that switch at 1.5 to 3 seconds on average. Nobody files a ticket over two seconds.
2. It breaks concentration rather than data. Across a 30–100 line table, switching hands constantly interrupts the rhythm of thinking. People read the wrong row, miss a unit price, put a specification on the wrong line. Those errors surface later, under a different name, and nobody traces them back to the mouse.
3. Users already have a workaround, and it hides the problem. They build a spreadsheet outside the system, finish the work there, and paste it in when someone chases them. From the management side the data still arrives. It just arrives late, and after one hand-copy.

The third is a familiar shape: the dispatch-board piece met exactly this in delivery planning. When a screen is slower than a spreadsheet, users do not object — they leave, quietly, and take the data with them.

2. What the shortcut layer actually is

Not a list of key combinations. Two different kinds of work.

Group one — keep both hands on the keyboard. Enter moves down and creates a new row at the end of the table. Shift + Enter moves up. Tab / Shift + Tab move across. ↑ ↓ move within a column. The only goal is taking cell-to-cell time from around two seconds to under 0.2 seconds.

Group two — what a spreadsheet does and a web form usually does not.

• Ctrl + D copies the cell above: same material group, same unit, same supplier.
• Alt + ↓ duplicates a whole row and increments its code — right for a bill of quantities where items differ only in size.
• Ctrl + V pastes a whole block from Excel: the customer or the engineering office sends the quantities as a spreadsheet, and it goes straight into the grid instead of being retyped.
• Ctrl + Backspace deletes the current row, Esc reverts a cell to its previous value — without hunting for a button.
• Ctrl + / toggles the shortcut reference in place.

Group one is what a stopwatch can measure. Group two is what stops people opening Excel first. Two different kinds of value, and only the first is what this piece puts a number on.

3. Three lines of money, and which of them to trust

The assumptions, taken from the source: a team of ten (estimators and quote desk), five documents each per day, forty lines per document, four cells to key per line — item code, specification, quantity, unit price.

A. Hours spent reaching for the mouse

This is the firmest line, because it rests only on time a stopwatch can measure. One document gives a baseline; ten give a distribution.

B. Gross profit from quoting sooner

Saving 20–30 minutes per document means the customer has the quote the same day instead of the next one. That is worth something real. But it is the softest line, and here something needs saying plainly about the source.

The source puts two numbers side by side: a win rate up 3%, and roughly 80 million VND a year of extra gross profit. Those two numbers do not belong to each other. At the volume the source itself assumes — 12,500 documents a year — a 3 percentage-point lift means 375 extra deals won. At any contract value worth calling a bill of quantities, 375 extra deals is not 80 million; it is billions. The 80 million was never derived from the 3% — it is a free-floating estimate placed next to a plausible-sounding rate.

What is interesting is that the money is about right and only the mechanism is wrong. Build the chain out — documents × win-rate lift × contract value × gross margin — and a lift of just 0.1 percentage points already produces about 75 million a year.

The gap between 0.1 and 3 percentage points is thirty-fold. A model that states which assumption it needs is one a reader can argue with; a free-floating number can only be believed or disbelieved.

If you have never measured win rate against response time, set this line to zero and read the total again.

C. Pricing errors avoided

Esc reverting a cell and Ctrl + V pasting a whole block both remove one specific family of mistakes: an extra zero, a misplaced decimal, a row slipped while travelling with the mouse. This is the kind of error that goes straight into a contract.

4. The total, and where the model collapses

A 6.6% gap, almost all of it from two deliberate choices: this piece uses 250 working days a year rather than the source's 312 (26 days × 12 months), and rebuilds line B from a chain instead of taking the free-floating figure. Against a one-off build cost of 20,000,000 VND, it pays back in about 42 days, a first-year return of 775%.

But the number that matters most is the zero. In the estimator at the top of this piece, set "seconds per cell with shortcuts" equal to "seconds per cell with the mouse", set the win-rate lift to zero, set the error rate to zero. The total goes to zero — not to some small remainder.

A model that cannot collapse when its assumptions are taken away is not a model. It is a prejudice in arithmetic costume.

5. What to do before believing any of this

Three steps, in this order.

1. Time a real document with a stopwatch. Not an estimate, not a question to the team. Sit next to somebody keying a forty-line table and time it from opening the form to saving. Do it with three different people — the spread matters more than the median.
2. Count cells, not rows. A four-cell row and a twelve-cell row are different problems. The input this model needs is cells, and it is usually larger than people expect.
3. Re-measure later, on the same group and the same season. If the before-baseline came from a quiet month and the after-measurement from a busy one, most of the improvement you are reading is the season.

In a factory, Kaizen starts by removing the times a worker has to bend or reach. In office software, making somebody lift their hands off the keyboard a thousand times a day is that same kind of waste — except that nobody sees it, because it leaves no trace anywhere in the system.