Case study — HP Smart Digitization
80%
reduction in architectural drafting time — from pilot concept to shipped HP product
Smart Digitisation
Experience the future of transforming architectural drawings to CAD with our innovative solution using our powerful ML engine
02 — The problem worth solving
More than half of all construction work today involves existing buildings — renovation, rehabilitation, retrofitting. Every one of those projects starts the same way: dig out the original drawings, then spend days or weeks converting paper into something editable. It was a universal bottleneck with no elegant solution.
Three existing approaches — all broken
Manual redraw
Retracing every line by hand in CAD. Accurate but brutally slow — hours per drawing.
Pain: timeSemi-automated SW
Tools like Scan2CAD assist the process but require extensive manual cleanup after.
Pain: accuracyOutsourced drafting
Sending files to CAD service providers offloads time but is expensive and opaque.
Pain: costTarget users
03 — My role and mandate
I joined HP's Large Format Printing division via Nacar as Customer Experience Lead — embedded within the product team responsible for taking Smart Digitization from internal concept to market. The product existed as an AI engine with no user-facing layer, no validated market, and no pricing model. My mandate was to build the evidence that would determine whether it shipped at all.
What I owned
Agile pilot program
Designed and ran 6 sequential pilots — from smoke test to live HP engine deployment
UX + service design
Full upload → convert → review → download flow across PlantoCAD pilot and HP Beta
User research + GTM validation
28 AEC customer sessions · willingness-to-pay interviews · segmentation strategy
ML benchmark database
Built and maintained the drawing database used to train and evaluate the AI engine
HP Beta UX system
Designed 3 interconnected interfaces — customer, admin, and reseller dashboards
HP Build Workspace contribution
Parallel involvement in HP's broader construction platform as Smart Digitization's anchor feature
My background as a licensed architect — having led major rehabilitation projects including Vienna Airport infrastructure — gave me direct domain fluency. I understood the pain of re-digitizing old drawings not as a user researcher, but as someone who had lived it.
04 — The system I built
Rather than designing a product and then testing it, I designed a validation system first. Each pilot had a single objective, a specific experimentation method, and clear success criteria. The findings from each one shaped the next. Nothing was assumed.
The data infrastructure I built alongside UX
05 — Results and impact
The evidence gathered across six pilots fed directly into HP's product decision. HP AI Vectorization launched publicly as part of HP Build Workspace — the first cloud-based AI vectorization tool built specifically for architects and AEC professionals, trained on over 1,000 real customer drawings from the benchmark database I built.
Willingness-to-pay — 13 mid-large AEC firms interviewed
06 — What I learned
The most important moment in this project was reporting the 2/10 result honestly — and watching it become the catalyst for a better product roadmap. These are the insights that shaped both the product and my approach as a designer.
Accuracy is a prerequisite, not a differentiator
When 6/10 users said they'd rather redo a drawing manually than clean up our output, accuracy became the table stakes. Speed and price only matter once the output is usable. The product roadmap shifted accordingly — accuracy first, everything else second.
The feedback loop had a fundamental design flaw
Happy users thanked us without saying why. Unhappy users selected every improvement option without explaining what was wrong. Neither gave the AI team actionable data. Feedback UI must extract signal, not just sentiment. Pilot 5 redesigned this from scratch.
Scanned and digital drawings are different products
We treated them as one problem for too long. Scanned drawings have fundamentally different noise, resolution, and recognition challenges than digital PDFs. Splitting them into two separate AI processing pipes in Pilot 5 was the right decision — it should have happened earlier.
The decision maker and the user are not the same person
Project Managers — not the architects doing the work — controlled purchasing decisions. Our UX was optimised for the person drawing. Our pricing and contracts needed to be built for the person approving the budget. This reshaped the HP Beta dashboard design entirely.
What I would do differently
Establish the scanned-vs-digital pipe split at Pilot 3, not Pilot 5. Instrument the feedback UI with forced-choice questions earlier. Push for HP-standard branding sooner — the stealth PlantoCAD approach created a perception gap when the product transitioned to HP identity.
customers would use and pay for our Pilot 3 output under best-case drawing conditions. Reporting this number honestly — rather than softening it — is what triggered the engine accuracy sprint that made the final product viable.
07 — The systems architect frame
This project sits at the exact intersection of my two domains. As a trained architect who led complex rehabilitation projects, I understood intimately why the restoration market bottleneck existed. As a UX lead at HP, I built the system that began to solve it. The connection is not metaphorical — it is the reason the product worked.
Physical infrastructure
Architect · 12 years
Digital ecosystems
CX Lead / UX Lead · HP, JAK, Stageflow
What this project enables next