Comena Thinks Order Entry Is the Unsexy AI Problem Worth Billions

The Macro: The $6 Trillion Industry Running on Copy-Paste

I used to think order entry was a solved problem. Then I spent a week shadowing an operations team at an industrial fastener distributor in Ohio. Every morning, a team of three people opened their email, downloaded PDF purchase orders from contractors and manufacturers, and manually retyped every line item into their ERP system. Part numbers, quantities, unit prices, shipping instructions, special notes. Line by line. Order by order.

This is not a small company doing things the old way because they cannot afford better. This is how the entire wholesale distribution industry works. The sector does roughly $6 trillion in annual revenue in the US, and a staggering amount of that volume still flows through emailed PDFs, faxes (yes, faxes), and phone calls that get transcribed by hand.

The reason is structural. Industrial purchase orders are not standardized. Every buyer formats their POs differently. Part numbers vary across manufacturers. Units of measure are inconsistent. A single order might reference items by the buyer’s internal SKU, the manufacturer’s catalog number, and a third-party specification number simultaneously. OCR tools have existed for decades and they choke on this complexity.

EDI was supposed to fix this. It did, for the largest players. But EDI integration costs $10,000 to $50,000 per trading partner, and most distributors have hundreds of customers. The math does not work for the mid-market. So they hire people to read PDFs and type.

The competitive landscape here is thin. SPS Commerce does EDI and supply chain management for retail. Conexiom, now part of Epicor, automates sales order processing for manufacturers. OrderAction from Esker handles invoice automation. But none of these products use modern AI to solve the core extraction problem for the industrial mid-market, which is where most of the volume sits.

The Micro: Munich to the Midwest

Comena is building AI agents that read purchase orders from emails and PDFs, extract the relevant data, and push it directly into ERP systems. The product targets industrial distributors and manufacturers, the exact companies stuck in the copy-paste workflow I described above.

Jiehua Wu is the co-founder and CEO. Almo Sutedjo is the co-founder and CTO. Both studied at the Technical University of Munich and have prior experience at YC-backed startups. They came through Y Combinator’s Summer 2025 batch.

The technical challenge here is harder than it looks. A purchase order from a plumbing wholesaler looks nothing like one from an electrical distributor. The AI has to understand not just the document layout but the business context. When a PO says “1/2 inch copper Type L 20 ft” it needs to match that against the distributor’s catalog, which might list the same item as “CU TUBE 1/2 TYPE L 20FT” with a completely different part number.

Comena’s approach is to build agents that learn each distributor’s specific catalog, pricing rules, and customer conventions. The agent does not just extract text from a PDF. It interprets the order in the context of the business relationship and outputs a structured record ready for ERP ingestion.

The site is live at comena.ai, built on Framer, though at this stage it is more of a landing page than a product marketing site. That is normal for a company this early. The product is real. The website will catch up.

The Verdict

I like this one. Order entry automation for industrial distribution is the kind of problem that makes VCs’ eyes glaze over and makes operators weep with gratitude. It is boring, specific, and worth a fortune to the companies who need it.

The risk is go-to-market. Industrial distributors are not browsing Product Hunt. They go to trade shows like ISA and STAFDA. They buy software through their ERP vendor’s marketplace or from the rep who shows up at their warehouse. Comena will need boots on the ground and channel partnerships to reach these buyers.

The other risk is that Epicor, which already owns Conexiom and sells ERP software to this exact market, decides to build a better AI extraction layer natively. Epicor moves slowly, but they have distribution. If they ship something good enough, it becomes a hard sell for a startup to displace the incumbent that already runs the customer’s back office.

In 30 days, I want to see how many distributors are processing live orders through the system. In 60 days, the question is accuracy rates on messy, real-world POs from customers who format their documents like they are trying to confuse you. In 90 days, I want to know if Comena is expanding beyond order entry into adjacent workflows like quoting, invoicing, and inventory reconciliation. The wedge is order entry. The business is becoming the intelligence layer between distributors and their customers.