Orders rarely start where the automation lives. For manufacturers and distributors, they arrive through CRMs, order-management platforms, portals, internal apps, emailed PDFs and, increasingly, AI agents. Each new channel lands on IT's desk as another integration project.

The instinct is to build. A model can read a purchase order in a pilot. Production is a different story. Order automation becomes a product that engineers must maintain indefinitely. Model and cloud bills grow with every document. Underlying models change on someone else's schedule. And when revenue depends on what the system enters, IT owns every miss. It's no surprise Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, citing cost, unclear value and weak risk controls.

So the better question isn't whether to build or buy. It's what to build. Build what makes you distinctive: the portal customers prefer, the CRM workflow reps live in, the agents that fit your operations. Turning a messy document into the order the customer actually meant is rarely where that engineering time pays off because of edge cases, cost and impact of failures.

Put the intelligence where the orders already are, not in one more screen. It should resolve what orders leave out, like the real SKU, pack size, customer price and ship-to. It should check pricing and inventory as it works. And it should show the reason behind every decision.

That last point matters most as agents arrive. An agent can route the work, but it shouldn't be the only thing standing between a PDF and the ERP. Whatever decides how an order lands should leave evidence a person can audit, not just a confidence score.

The future of order automation isn't one more application. It's trusted intelligence wherever the order begins.

Cliff Barrett, Conexiom chief product officer