AI agents that chase suppliers by email are the wrong fix

Summary. Several companies now sell AI agents that read your open purchase orders and email your suppliers to confirm orders and explain late deliveries. This is a superficial fix for a deeper problem. Every reminder is another email that a person at your supplier has to read, and every reply is another email that a person on your team has to check. At the scale of thousands of open orders, and with every customer running its own agent, the load lands on human teams on both sides. The information these agents ask for already exists in each company's system of record. The durable fix is to give both companies the same data structure, so an update in one system is visible in the other as soon as it happens.

The agents that chase suppliers

Purchasing teams spend a large part of the week asking suppliers the same questions: is this PO confirmed, when will it ship, why is it late. Software vendors have noticed, and the most common answer in 2026 is an AI agent that sends those questions for you.

• Microsoft. The Procurement Agent in Dynamics 365 Supply Chain Management can generate emails that remind vendors to confirm a purchase order or ask why a delivery is late. It can send them automatically or save them as drafts for review, and it reads vendor replies to suggest changes to the PO. It is billed per run, plus a variable cost for each email it writes and reads.
• SourceDay. Its Open Order Chaser automatically sends reminder emails for open and late orders on a cadence that the customer configures.
• Leverage AI. It reads open PO data from the ERP, sends supplier follow-ups, and writes confirmed dates, partial shipments and exceptions back to the ERP. Replies it cannot classify go to a human review queue.
• Didero. It raised a $30 million Series A in February 2026. Coverage of the round describes agents that read supplier communication from email, WeChat and phone calls and update the customer's ERP.

Microsoft's own documentation states the problem clearly: communication with vendors is "often unstructured and conducted via email, even for companies that implement electronic data interchange." Everyone agrees on the diagnosis. The disagreement is about the treatment.

Email follow-up does not scale

An agent makes it cheap to send a follow-up. It does not make it cheap to receive one. Microsoft's 2025 Work Trend Index found that the average worker already receives 117 emails every workday. Automated follow-up adds to that number on both sides of every order.

Consider a mid-size OEM with 2,000 open PO lines and a common rule: send a reminder when a PO is unconfirmed after three business days, or when a line is late. If one line in four trips that rule in a given week, the agent sends 500 emails that week. Each reply must be read, matched to the right PO line and checked before the ERP changes. The vendors above keep a person in that loop for good reason: a wrong date written into the ERP is worse than no date.

Now look at the same rule from the supplier's side. A supplier with 100 customers receives the same reminders from 100 agents, on 100 schedules, in 100 formats. Its customer service team does not get smaller. It gets busier, answering questions whose answers are already in its own ERP. When suppliers deploy their own agents to answer, you get agents writing prose to agents that parse prose, with people on both sides reviewing the exceptions. That is the email relay, run faster.

(The OEM figures above are an illustrative example, not measured data. Run the same arithmetic with your own open PO count and reminder rule.)

A faster pigeon is still a pigeon

Look at the path a single promise date takes in this model. It starts as a structured field in the supplier's ERP. A person, or an agent, turns it into a sentence in an email. Your agent reads the sentence, guesses which PO and line it refers to, and turns it back into a structured field in your ERP. The data was structured at the start and structured at the end. Everything in between is translation, and every translation is a chance to lose the date, the quantity or the reason it changed.

This is the lazy solution. It automates the symptom, the email, and leaves the cause untouched: two companies keep two separate records of the same order and reconcile them by message. It also creates a business that grows with the problem. When an agent is priced per email written and read, more email means more revenue for the vendor and more noise for your team.

Gartner predicted in June 2025 that over 40% of agentic AI projects will be canceled by the end of 2027, citing rising costs, unclear business value and weak risk controls. Agents that generate more work for people than they remove are a likely place for that to happen.

Start from first principles

Ask what a buyer actually needs when they chase a supplier. It is not an email. It is a fact: the confirmed quantity, the promise date, the reason it moved, the ship notice, the certificate.

1. The facts already exist. Your PO is in your ERP. The supplier's sales order, promise date and shipment are in its ERP. Nobody needs to write them again.
2. The gap is structure, not effort. Your system calls it PO 4471. The supplier's system calls it SO-10883. The two records describe the same order in different shapes, so neither side can read the other.
3. Fix the structure once. When both companies map their records to the same data structure, PO 4471 and SO-10883 become one shared order. Each company decides what the other can see.
4. Updates become live. When the supplier changes a promise date in its ERP, the change is visible on your side immediately, with its source. There is nothing to chase, because nothing was hidden.

This does not remove AI from procurement. It gives AI something reliable to read. An agent that queries a shared record answers in seconds and cites the source. An agent that emails a supplier waits days and guesses.

What this looks like with Indax

Indax connects to each company's email, CRM and ERP and maps them to one shared data model. You and your supplier connect once. From then on, the PO, the confirmation, the promise date, the shipment and the certificates are one record on both sides, and each company controls its own permissions.

• For your buyers: order status is read from the supplier's system, not requested from a person.
• For your suppliers: no portal to log in to and no reminders to answer. Their ERP already holds the answer.
• For your agents: Claude, ChatGPT, Copilot or any agent that speaks MCP can query the shared record directly and show where each answer came from.

Agents that write better emails will make the relay faster for a while. A shared record removes the need for the relay.

Sources: Microsoft Learn, "Supplier communications features of the Procurement Agent" (Dynamics 365 Supply Chain Management, production-ready preview). SourceDay, "AI Agents for Procurement". Leverage AI, "Supplier follow-up automation for ERP". SiliconANGLE, "Didero raises $30M Series A" (February 12, 2026). Microsoft, 2025 Work Trend Index special report. Gartner press release, June 25, 2025, "Gartner predicts over 40% of agentic AI projects will be canceled by end of 2027".