Japanese purchasing teams use CADDi's drawing search to catch duplicate parts and check prices

A buyer searches the drawing archive and finds the same part sitting under two numbers at two prices. Consolidating to the lower one is a saving with a purchase order behind it. SiliconANGLE reports that purchasing teams at CADDi's customers use the drawing search for that, and to check what suppliers charge ^[3]. The search itself was built for engineers, who look up archived designs by shape or specification ^[2]. The report does not include a price for the software or a per-customer saving figure ^[20].

The adoption number is the firmest evidence in the material: more than half of Japan's 100 largest manufacturers are customers ^[4]. Depth of use is thinner on the record. CADDi says sales have been more than doubling each year since March 2025, when Atomico led a $38 million Series C extension at a $470 million valuation ^[7][8]. Headcount has grown by about half in that time, to roughly 900, which puts the company near 600 people at the earlier round ^[1]. The new valuation is about 2.6 times the old one, set some 18 months later ^[2][4]. Of $234 million raised in total, about $120 million came in before this round ^[3].

Ken Hara, a partner at new investor Coreline Ventures, stated the precondition. "AI only delivers where a data layer and a semantic layer already exist," he said ^[11]. He added that in manufacturing "that layer is buried in information fragmented across departments and in records that were never digitized at all, which is why horizontal tooling never reaches it" ^[12]. Design files, purchase orders and inspection results tend to stay with whichever team created them ^[16].

Chief Executive Yushiro Kato is direct about why he thinks a general model will not do this job. "I've never seen anybody who uses LLMs to do design reviews because it doesn't understand drawings or CAD," Kato told Fortune ^[13]. In an essay on the company's website he wrote that over three decades "we made the world of bits more than a hundred times faster" while the world of atoms held roughly its pace in most industries ^[14]. He argues that general-purpose AI cannot handle manufacturing data without domain-specific models ^[15].

CADDi now calls Explorer the successor to Drawer, alongside a new agent product and six workflow products spanning the value chain ^[6]. Luca Eisenstecken, a partner at Atomico, said the company has moved past fixing individual workflows and is building "the intelligence layer for the entire manufacturing lifecycle" ^[17]. A purchasing team trained on Drawer's search now has a different product name in front of it.

Does purchasing hold a price history keyed to part number, and are the archived drawings machine-readable or scanned paper? If both hold, a duplicate-part audit is already possible in a spreadsheet, and what you are buying is speed and coverage across an archive nobody has time to read. Price history plus paper drawings means the first months go to ingestion, and time-to-value tracks the scanning backlog rather than the model. Machine-readable drawings with no price history gives engineering a better lookup and gives procurement nothing it can put on an invoice. If neither holds, this is a digitisation programme, and it should be budgeted as one.