Savings calculator
What would AI engineers take off your team?
Enter the tags on your project and the stages you run. See the engineering hours each Deskely product takes over, from reading the drawings to assembling the handover dossiers.
Modelled estimates: the engineering hours each product takes over on a 100,000-tag project, scaled to your tag count. Review time stays with your engineers. Actual savings depend on the scope you hand over.
Hand work the AI engineers take over: reading drawings, comparing revisions, checking records, assembling dossiers.
Errors caught before they reach the record, instead of found at completion, commissioning or handover.
- 01Engineering125,000 h25,000 h
- Extract40,000 h8,000 h
- Delta22,000 h4,400 h
- Clause27,000 h5,400 h
- Submittal27,000 h5,400 h
- Conform9,000 h1,800 h
- 02Completions111,800 h22,400 h
- Core3,750 h750 h
- Verify40,000 h8,000 h
- Walkdown23,000 h4,600 h
- Testpack45,000 h9,000 h
- 03Commissioning32,000 h6,400 h
- Sequence20,000 h4,000 h
- Horizon12,000 h2,400 h
- 04Handover36,700 h7,300 h
- Redline11,000 h2,200 h
- Dossier3,700 h740 h
- Asset22,000 h4,400 h
Labour is the work taken over now; rework is time saved later. They do not overlap, so they add up. The same accuracy rates apply to every product's hand work.
Where these numbers come from
- Hours per product. Deskely's modelled engineering hours per 100,000-tag project, the same figures as the product pages, scaled in proportion to your tag count. Reading a drawing by hand takes about 0.4 hours per tag on our measured case.
- Hand work that goes wrong: 3%. Applied the same way to every product's hand work. No published rate exists for completions records. In measured studies, typing from a clean form gets about 0.3% of fields wrong, copying values into another system 3.7% of entries, and reading a complex document and writing out what is in it about 6.6% of fields (meta-analysis of data entry methods, JAMIA transcription study). A tag, an ITR or a test pack has several fields, so 3% is a conservative start.
- Left after AI engineers and review: 1%. Extract reads tags at about 99% accuracy. Every AI engineer checks its work against the register and the source documents, and sends what it is unsure of to an engineer for review.
- Cost to fix later: 10 times the work. The 1-10-100 rule of data quality: an error costs about ten times more to fix downstream than at entry. For a tag read by hand in about 0.4 hours, that is about 4 hours: correcting the register, reissuing ITRs, sometimes a retest. Rework avoided is the hours each product takes over, times the 2% difference, times ten. Industry studies put total rework at around 5% of project cost, with bad or inaccurate information behind roughly 14 to 22% of it (summary of rework studies).
Want a measured number instead of a modelled one? Request a free scope test on one package of your drawings.