erpkaizen

Measuring the value

A site diary generated from one plan: 30 pages say "Sunny", the month had 18 rainy days

Has a measurement By 30 Oct 2026 About 10 minutes
Three jobs on a plan joined by arrows: foundation concrete, foundation masonry, backfill. The third starts earlier than its predecessor allows and is flagged. An arrow leads to a pre-filled diary page: light rain, the day's job. No real data. Plan Footing concrete Footing masonry Backfill ⚠ Diary 15/09 Morning: light rain Eng 2 · Crew 12 · Footing masonry One plan, joined to the quantities diary, acceptances, payment come from it
One plan in a standard order, joined to the quantity lines: a job that starts early is flagged, and the diary page fills in the day’s work and the real weather.

On a real construction-management workbook, 132 of 330 schedule rows were copies of the price annex and every diary page said "Sunny". When the diary, the acceptance records and the payment quantities come from one plan joined to the quantities, how many hours a month does an engineer get back?

How many hours a month does a site engineer spend copying the plan into paperwork?

All it needs is the number of sites, the jobs in each schedule, the months of construction and an hourly cost. Pages, records and payment periods follow from those. Everything is editable.

That one engineer keeps the records for
Rows in the job list / programme
From ground-breaking to completion
Salary plus social insurance, per hour

Derived from the four numbers above (per month)

Diary pages–
Acceptance records–
Payment quantity statements–
Documents sent back for correction–
Assumptions — open to see and edit each one

Counts

Diary

Acceptance records

Payment quantity statement

Documents sent back

Setup

Hours a month for each kind of document Left: today · right: generated from one plan · hover a line for its numbers
Hours a month for each kind of document

A. Diary–
B. Acceptance records–
C. Payment quantity statements–
D. Documents sent back–
E. Setup hours, spread over the build–
Hours given back each month, valued–

An estimate, not a promise. Line D is the softest — set both return shares equal and read again. Set the pre-filled minutes equal to today's: every line goes to zero except setup. An honest model has to be able to do that.

Open the construction-management workbook of a site engineer on a publicly funded restoration project and turn to the diary sheet. Every page records the weather as "Sunny", two site engineers, nine workers — every day exactly like the day before. Against the archived readings of the weather service for that very area, September had 18 days with rain during working hours, 9 of them with 2.5 mm or more.

Nobody set out to write it wrong. The diary is made by copying yesterday's page, changing the date, editing a few lines of work, then printing page by page. On the next sheet, 132 of the 330 rows in the job list are near-verbatim copies (95% alike or more) of a line in the contract price annex; 277 rows are at least 60% alike. The same job is typed at least twice, and the two copies are not joined: change a quantity in the annex and the schedule never hears of it, while the acceptance record is typed a third time.

This piece asks: if the diary, the acceptance records and the payment quantity statement were all generated from one plan, joined to the price annex, how many hours a month would an engineer get back?

1. What changes

The engineer imports the very Excel file already in use — the price annex and the job list. Columns are recognised by their header text, so a workbook that differs from site to site still reads. Then:

  1. Every schedule row has its own code, joined to the code of a quantity line. The machine suggests the most similar annex line for each job; the engineer confirms it. Until confirmed, the quantity reaches neither an acceptance record nor a payment.
  2. Jobs have a standard order. Excavation before blinding, foundation concrete before foundation masonry, structural concrete must reach its age before the cube test. A job starting earlier than its predecessor allows is marked ⚠ — the machine only marks it, it does not move it. On the workbook above, about a quarter of the jobs were marked.
  3. The diary fills itself: the jobs in progress and the day's acceptances come from the schedule, the workforce from the site's defaults, the morning and afternoon weather from real readings at the site's coordinates — and all of it can be edited. Once a day is locked, its content stays put even if the schedule changes later.
  4. Acceptance records and the statement of completed quantities print pre-filled with the record number, quantities and signatories; the payment statement sums the records accepted within the period.
Diary calendar for September: each day shows its weather, the jobs in progress and the acceptance records
One month of the diary on a sample site: every day already carries the real weather at the site's coordinates, the jobs in progress and the acceptance records, before anyone types a word. Demo data, from no real project.

Vietnamese rules set no mandatory template for the construction diary, but it must record the weather, temperature, labour, equipment, the work done and the acceptances of each day (Decree 207/2026/ND-CP, Appendix IIa). So the print templates are data anyone can edit, not code.

2. First stream: diary pages

Hours saved/month = sites × working days × (minutes per page today − minutes once pre-filled) ÷ 60
Example One engineer keeps the records of 2 sites, 26 working days: 52 pages a month. Today each page takes 20 minutes (copy the old page, edit the work, change the date, print). Once pre-filled, what is left is to read, fix what is wrong and lock: 3 minutes. 52 × (20 − 3) ÷ 60 ≈ 14.7 hours a month.

How to measure it. Time one week of diary as it is done now, printing included. The "after" minutes are measured by the tool itself: the moment a day is opened and the moment it is locked are both recorded.

How far to trust it: high. The pages are counted, the minutes measured over a week.

3. Second stream: acceptance records

Hours saved/month = sites × jobs in the schedule × records per job ÷ months of construction × (minutes today − minutes once pre-printed) ÷ 60
Example 300 jobs per site, 10 months of construction. On the real workbook there were 288 records for 333 jobs — about 0.86 records per job. Two sites make about 52 records a month. Typing the record number, job name, date, quantities and signatories by hand: 25 minutes. Pre-printed, then checked: 4 minutes. 51.6 × (25 − 4) ÷ 60 ≈ 18.1 hours a month.

How to measure it. Take last month's list of records, count them and ask the person who wrote them how long one takes. Records the supervisor sent back for a wrong figure belong to the fourth stream — do not count them twice.

How far to trust it: fair. The records are counted; the minutes vary a lot between a record with one quantity line and one with many.

4. Third stream: the statement of completed quantities

Hours saved/month = sites × payment periods per month × (hours per period today − hours once summed from records)
Example One period per site per month. Today the statement of completed quantities for one period takes 6 hours: dig out the records, add them up by hand, check them against the annex. Once it is summed from the records accepted in the period, what is left is checking: 1 hour. 2 × (6 − 1) = 10 hours a month.

How to measure it. Next time a payment file is prepared, note when it starts and when it is submitted. Do the next one on the tool and compare like with like — a period full of new work items is not a period of finishing work.

How far to trust it: fair. One period a month is an assumption; many publicly funded projects pay less often — change that field.

5. Fourth stream: documents sent back

Hours saved/month = (pages + records + periods) × (share sent back today − share once the machine checks) × minutes to fix a page ÷ 60
Example About 106 pages a month. Today 5% come back because an acceptance is dated before the work finished, a record's quantity does not match the annex, or concrete was accepted for quality before it reached its age; once the machine checks order and quantities, 1% remain. Fixing and reprinting a page takes 40 minutes. 106 × (5% − 1%) × 40 ÷ 60 ≈ 2.8 hours a month.

How to measure it. Count the pages the supervisor or the owner sent back over the last three months, and the reason for each. Only the reasons a machine can check (dates, order, quantities) belong here.

How far to trust it: low — the softest line. Set both shares equal and the calculator still shows 42.0 hours a month.

6. To subtract: setup

Each site needs its file imported, the suggested quantity links and order confirmed, the signatories and coordinates entered: about 4 hours. Spread over 10 months of construction, 2 sites cost 0.8 hours a month. None of this uses AI, so there is no model cost.

7. Adding it up

StreamHours/monthValue/month (60,000 VND/h)
Diary14.70.88 M
Acceptance records18.11.08 M
Statement of completed quantities10.00.60 M
Documents sent back2.80.17 M
Setup−0.8−0.05 M
Given back44.82.69 M

About 45 hours a month, close to six working days, for one engineer looking after two sites. Today the paperwork takes about 54 hours; the machine takes over about 84% of them.

Count it once. Those 45 hours are money — 2.7 million VND a month, 32 million a year at the salary in the example — or the capacity for the same engineer to take on another site. Never both. For a small contractor the real value often lies in the second, but that is the company's decision, not the calculator's.

8. What this ROI stands on

The plan has to be kept true. That assumption sits under all four streams. A diary generated from the schedule is only right when the schedule is; if the real work drifts away from it and nobody updates it, the machine will neatly pre-fill wrong pages. That is why the diary stays a draft until the engineer locks the day — and locking it is where a person still has to work.

Automatic weather is a station reading, not the site's. Each grid cell is several kilometres wide; a local shower may not be in the reading. The weather field can be edited, and an engineer's own entry is never overwritten.

The standard order is a suggestion, not a law. Some jobs overlap for good reasons. A ⚠ only says the job starts earlier than the order allows; the engineer still decides.

Plan table: each job has a code, dates, its predecessor and how many quantity lines are joined; jobs starting out of order carry a red bar and a ⚠
The plan of a sample site after import: each job has its own code, its predecessor in the standard order, and the number of quantity lines joined to it (yellow cells are suggestions not yet confirmed). A red bar at the start of a row marks a job that starts earlier than the order allows. Demo data, from no real project.

Machine-suggested quantity links must be confirmed. They are suggested by how alike the wording is; a wrong line joined to a record is a wrong line printed beautifully. Until confirmed, it reaches neither a record nor a payment.

9. Measuring it in your own company

  1. Take a baseline. One week: time the diary, count the acceptance records and the hours to write them; the hours spent on the last quantity statement; the pages sent back over three months.
  2. Run one site for one month. Note the setup hours. Each week, note the days locked, the quantity links confirmed, the jobs marked ⚠ and how many of those overlap on purpose.
  3. Compare like with like. A foundation month has more records than a finishing month; compare foundations with foundations.

10. What to remember

A job should be typed once. When the price annex, the schedule, the diary and the acceptance records are four hand-typed copies of the same event, each looks right and none checks the others. Join them with one code, and a "Sunny" diary page on a rainy day becomes the hard thing to produce instead of the easiest.

A site diary is not valuable because it is thick. It is valuable when every line in it says what actually happened.

The share of schedule rows copied from the price annex, the records per job and the "Sunny" diary on every day were measured on a real construction-management workbook, anonymised. The rainy days come from archived weather readings for that area and month. The number of sites, the minutes, the salary and every assumption in the calculator are illustrative. Replace them with your own.

Read next