For anyone who has spent years in the field, Excel is far more than a spreadsheet program. It holds design criteria, it carries the trial and error a company has lived through, and it quietly encodes who reviews what, and in which order. The problem is not that Excel is old. It is that the knowledge inside it moves file by file, disconnected from drawings, models, reports and specialist software.

The question you hear most often in the meeting room

A weight review meeting at a shipyard, a design coordination meeting on a construction project, and a P&ID review on a plant project share a strangely similar scene. The latest model is on the screen, and every participant has a different Excel file open in front of them. One person argues from a PDF received yesterday, another from a file on the shared drive, and a third from an email attachment.

The argument rarely begins with a technical question. It begins with verification: “Is this value current?”, “Which drawing did that number come from?”, “Has the change been reflected in the estimate as well?” Engineers have to establish the reliability of the data before they can even start calculating. This is less a problem of design complexity than a cost created by broken links between pieces of information.

Drawings and modelsGeometry and location are managed reasonably well, but calculation inputs and approval context are easily lost.
Excel calculation sheetsProven formulas and in-house know-how live here, but the meaning is trapped in cell addresses and file versions.
PDF reportsRich in standards, assumptions and test results, yet not data you can recalculate from.
Specialist analysis softwareProduces precise results, but has weak links to documents from other disciplines.
ERP and scheduling systemsManage cost and schedule, but are poorly connected to the engineering rationale behind design changes.
Human experienceFills the gaps of exception and judgment, but becomes hard to trace when the organization changes.
Engineering Data BackboneA shared spine connecting source, version, object, formula, standard, change and approval
Figure 1. The industry already has more than enough information. What it needs is not to copy everything into one place, but to build trustworthy connections between different originals.

Why Excel does not simply go away

The claim that digital transformation is complete only once Excel is gone takes far too simple a view of the field. Excel is fast and flexible, and it lets engineers see the calculation as it happens. It adapts easily to the client-specific formats and exceptional conditions that differ from project to project. Above all, it already contains formulas and report layouts validated over decades.

A single calculation sheet holds more than formulas. It encodes which values must be entered first, which ranges should raise a warning, and which table layout reassures the reviewer. That is why one aging Excel file often behaves like a small piece of specialist software. Rebuild all of it as a new web screen while ignoring that, and you end up with a product that looks modern but that practitioners do not trust.

The point is not to discard Excel, but to stop Excel from single-handedly serving as database, calculator and report generator all at once.The starting point of “Between Drawings and Excel”

What has to change, then, is not the tool but its role. Authoritative data is managed in a controlled data layer, while Excel is freed to focus on calculation, analysis and reporting. Users keep the calculation sheets they know, but the inputs arrive from drawings, models and reports carrying their source and version with them.

When the file is the truth

The same input is re-entered into several calculation sheets
Copies in email and shared folders disagree with each other
Only the author knows what cell B17 actually means
After a design change, people rely on memory to find the impact
The result survives, but the basis for it fades

When data flows and files present it

Authoritative inputs are managed as objects and properties
Excel keeps its proven calculation logic and formatting
Cells are explicitly linked to their engineering meaning
A change event finds the affected calculations and drawings
Every result carries source, standard, version and approval
Figure 2. The goal is not to remove Excel, but to connect the knowledge locked inside it to the wider engineering flow.

One giant database is not the answer

Shipbuilding, construction and heavy industry use different systems, and different people own the data. Hull geometry sits in 3D CAD, structural analysis results in specialist software, ground parameters in geotechnical investigation reports, and equipment procurement status in ERP. Forcing all of it into a single system looks tidy in theory, but on real projects it can make ownership of the original and its update cycle even less clear.

A more realistic structure is a federated engineering data backbone. Each system keeps doing what it does well. What gets connected in a shared layer is the object ID, the meaning of each property, the unit, the source document, the revision number, the calculation dependency and the approval status. What matters is less “where it was stored” than “what this number is, where it came from, and what it was used for”.

01Respect the originalPreserve the authority and scope of responsibility held by CAD, BIM, analysis software, Excel and ERP.
02Assign meaningMap cells and drawing elements to objects such as ship equipment, structural members, soil strata and piping tags.
03Reproduce the calculationPin inputs, formulas, units and standard versions so the same conditions produce the same result again.
04Trace the changeExplore, as a graph, how one change propagates through calculation sheets, drawings, quantities, cost and schedule.
05Let people approveAn authorized engineer reviews the AI's proposal and its evidence before anything is finally applied.
Figure 3. The five stages of a federated data backbone. The purpose of integration is not to monopolize data, but to build a path of trust.

Should AI do the calculating, or the connecting?

Since generative AI arrived, many products have promised to “design in natural language and calculate automatically”. But what matters in engineering is not a plausible answer; it is a reproducible one. A calculation whose result shifts for the same input, or that hides its units and applicable standards, creates new risk rather than speed.

The safer arrangement is to leave the actual numerical work to validated Excel formulas, Python modules, analysis software and Rule-as-Code. LLMs and VLMs deliver far greater value elsewhere: finding candidate inputs in drawings and PDFs, grouping different names into a single object, explaining what a formula means and which standard applies, and showing how far a changed value propagates.

Deterministic calculation and rule engine

Produces the same result for the same input, with standards and units under explicit control.

Runs Excel formulas and engineering calculation modules repeatedly under identical conditionsExplicitly checks unit conversions, allowable ranges, required inputs and design criteriaCompares results against the original Excel and reference examples as a regression testStops rather than applying anything automatically once an error is confirmed

LLM and VLM layer

Reads the context of documents and drawings, and connects evidence and change impact in terms people can follow.

Searches drawings, reports and tables for input values and their evidence regionsLinks a cell address such as B17 to meaning: a design parameter, a member dimension, an equipment weightExplains in plain language what changed and which calculation sheets are affectedWhen uncertain, states why review is needed instead of manufacturing confidence
Figure 4. Separating the roles of AI and the calculation engine widens the scope of automation while preserving reproducibility and a clear chain of responsibility.

What happens across an industry when a single number changes

The value of this philosophy becomes clear not in a search screen but in a change situation. When the weight of a ship's pump changes, the equipment list, lightship weight, center of gravity, purchase specification and production plan are all affected. When the groundwater level on a construction site shifts, earth pressure, uplift, bearing capacity, liquefaction, the dewatering plan and the construction cost all have to be revisited. When an equipment tag on a plant P&ID is changed, the equipment list, datasheets, purchase documents, fabrication drawings and inspection records all need checking.

Today, experienced engineers hold these connections in their heads. The next generation of systems has to record a change as an event and follow object relationships and calculation dependencies to find the scope of impact. Automatically correcting every item is not the objective. Missing nothing that needs a second look comes first.

Drawings and modelsWhether geometry, annotation and detail references have been updated
Engineering calculationsInput and formula dependencies, and whether recalculation is required
Technical standardsWhether the applicable clauses and in-house rules need re-review
One changeA revision to a value, geometry, specification or status
Quantities and costShifts in quantity, unit price, allowance rate and estimate
Schedule and procurementImpact on order lead time and the sequence of fabrication and construction
Inspection and approvalWhether re-approval, samples, or class and supervision review are required
ReportsConsistency across tables, body text and conclusions
Operational recordWho approved what, when, and on what basis
Figure 5. Engineering Change Intelligence is not an auto-correction feature. It is a control mechanism that shows the radius of a change first.

Seven principles to settle before the product

When this concept is turned into a real product, certain principles matter more than a spectacular AI demo. First, do not make an enemy of existing Excel files and specialist software. Second, give every value a meaning and a unit. Third, distinguish the original from the copy. Fourth, calculations must be reproducible under identical conditions. Fifth, trace the impact of changes. Sixth, never hide the AI's evidence or its uncertainty. And finally, at the end of every automated path there must be a point of approval by an accountable person.

Follow these principles and the product itself looks different. The center is not a chatbot but the connections between objects and calculations. The source and version of an input matter more than a beautiful dashboard. Difference comparison and rollback are designed before automatic generation. And a scheme for distinguishing which items can be auto-approved from which must always be seen by a person matters more than a single number like “99% accuracy”.

0File searchDocuments and calculation sheets are found quickly, but people still have to interpret the relationship between values and calculations.
1Structured extractionInputs and results are extracted from tables, drawings and cells, with units and formats normalized.
2Semantic linkingObjects, properties, cells, standard clauses and source regions are bound into a single network of relationships.
3Reproducible calculationThe same inputs and formulas are verified repeatedly across the web and Excel.
4Change impact analysisThe system proposes automatically how far a revision reaches into calculations, drawings, cost and schedule.
5Controlled executionAfter verification and approval, a limited change is applied and the full history is recorded.
Figure 6. Real development has to climb step by step from search to controlled execution. Skip a stage, and the faster the AI becomes, the faster errors spread.

Design the flow of knowledge, not the future of the tool

The engineering industry already owns plenty of software. What is missing is not another screen. What is missing is a common language that lets the geometry of a drawing, the calculation in Excel, the assumptions in a report, the sentences of a standard and the changes on site understand one another.

This series examines that common language across shipbuilding, construction and heavy industry. The first installment, on shipbuilding, starts with why shipyards still run on Excel. It means treating Excel not as an aging legacy but as an executable asset that carries ship design knowledge.

doAZ Point of View

Beyond an AI that reads drawings, what is needed is a system that explains which calculation sheets and standards a value passed through, and which decision it led to. What doAZ is building is not a general-purpose chatbot but an operating system for verifiable engineering data and calculation.

YT
Youngtae Kim · CEO · doAZ

Moving between construction and engineering practice, R&D, and the industrial AI business, he works on how to connect the knowledge scattered across drawings, documents and calculation sheets. This series is an open record of the product philosophy and validation direction doAZ is pursuing.