Case Study: A Saudi manufacturer managing more than 200 active purchase orders with over 60 international suppliers
A supplier can change a shipment date with a simple 12-word WhatsApp message, quietly affecting a manufacturer’s production schedule, inventory levels, and cash requirements for weeks to come.
That’s why supplier ETD tracking isn’t really a spreadsheet problem. It’s a visibility problem caused by fragmented communication. The purchase order might sit in the ERP, while a supplier sends an update by email or WhatsApp. The latest shipment plan may be in a spreadsheet, and the production team could still be working with yesterday’s information. A small communication gap may seem harmless at first, but repeated across hundreds of orders, it can quickly turn into a major operational challenge.
Agentic AI can be valuable here when it works as a controlled coordination layer. It monitors approved communication channels, identifies the purchase order or batch mentioned by the supplier, and extracts and standardizes the shipment departure details. The agent then compares the update with the latest confirmed plan. If the information is unclear, it sends the case to a human for review instead of making assumptions. Once approved, it updates the system of record and alerts the relevant team when a change could affect production or inventory.
This article explores how a supplier ETD agent works, the operational problems it can address, where human oversight remains important, and what we learned from implementing one for a Saudi manufacturer with a large international supplier network. The case-study figures come from client-reported results after the workflow was implemented. They represent outcomes from a specific operating environment and should be viewed as an example rather than a guarantee of results for other businesses.
An AI agent for supplier ETD tracking continuously turns scattered supplier communications into structured and reviewable supply-chain updates. Instead of requiring procurement coordinators to search through email conversations, WhatsApp chats, WeChat messages, and spreadsheets for shipping updates, the agent finds relevant information and converts it into usable purchase-order line or batch-level data.
A production-ready agent does more than extract information. It understands the context of each conversation, distinguishes between an original date and a revised one, and interprets approximate phrases such as “mid-April” or “after the holiday.” It can also match each update to the correct purchase order and batch, preserve the source as evidence, assign a confidence score, send uncertain cases to a human for review, and sync approved updates with the ERP or planning system.
The keyword is approved. In most manufacturing environments, the first practical version is not a fully autonomous procurement agent. It is an exception-management system that handles repetitive monitoring and reconciliation while people remain in control of ambiguous situations and decisions that could have a significant financial impact.
Enterprise systems are usually good at storing a confirmed date once it has been entered. The real challenge is getting the latest date from the supplier and confirming that it belongs to the right order, item, and shipment.
Modern supply chains still rely on a mix of structured and unstructured communication. Large suppliers may provide advance shipping notice (ASN) data or structured order confirmations, while others rely on spreadsheets. Many simply send updates by email. In the Middle East and Asia, suppliers and buyers also frequently use WhatsApp or WeChat for operational updates because these are the communication channels they already use every day.
This creates a gap between supplier communication and the systems that manage official supply-chain data. SAP identifies poor visibility, disconnected processes, weak communication, and manual data handling as common barriers to supply-chain visibility. DCSA also highlights how shipping data is often manual, inconsistent, and delayed. As supply-chain technology improves, manufacturers still face a key challenge: connecting accurate system data with fragmented supplier communications across email, messaging platforms, spreadsheets, and other channels.
A supplier might say, “first batch next week,” because the buyer already knows what they are referring to from the conversation. But if an AI system sees only that message, it cannot tell which PO, line item, or batch is being discussed. To understand the update accurately, the agent needs to consider the supplier’s identity, the conversation history, current open orders, and sometimes earlier quantities or delivery dates.
A single email can include updates for three purchase orders, two batches, and several different dates. For example, one batch might ship around March 15 and another at the end of April, while a third could be delayed because a component is still pending. An ETD agent needs to separate these updates and link each date to the correct purchase order or batch, rather than treating the entire email as a single shipment record.
Supply-chain communication often includes phrases like “around the 15th,” “second week of May,” “after Chinese New Year,” “production complete by month end,” or “vessel booking expected next week.” These phrases do not all carry the same meaning. Some indicate a planned departure date, others refer to a production milestone, while some simply express an expectation. An ETD agent needs to understand that difference before turning these messages into structured shipment data.
A good system should preserve the uncertainty in supplier updates. If a supplier says “end of April,” the agent can interpret that as a planning window, such as April 25–30. But it should not turn that estimate into a firm April 28 commitment. The original wording, interpreted date range, and confidence level should remain visible to the reviewer so they can see both what the supplier said and how the system interpreted it.
Estimated Time of Departure (ETD) is the expected date or time when a shipment leaves its origin or shipping point. Estimated Time of Arrival (ETA) is when the shipment is expected to reach its destination or the next major milestone. Manufacturers may also use an internal planning date that accounts for factors such as transit time, port handling, customs clearance, and inland transportation. These dates can be different, so an ETD agent needs to understand which date is being referenced before updating the supply-chain record.
These dates should never be merged into a single value without making the distinction clear. If a supplier provides an ETD and the system calculates an ETA based on historical transit times, the interface and audit trail should clearly show that the ETA is a calculated estimate, not a date provided by the supplier. Microsoft’s current Dynamics 365 Procurement Agent takes a similar approach. It can identify the ETD supplied by the vendor, use agreed transit-time assumptions to estimate the ETA, and then show the calculated date to the buyer for review and validation.
A new supplier message does not always mean the latest information is the correct one. It may conflict with a formal supplier confirmation, an ERP record, or a date discussed during a later phone call. In some cases, a supplier may repeat an old date in a new email thread, while two people from the same company may give different delivery estimates. A reliable workflow needs to look at more than just the newest date mentioned. It should consider the source, timestamp, previous confirmations, and which information takes priority. When the data still conflicts, the issue should be flagged for human review rather than automatically replacing the ERP record with the latest date found in a message.
A five-day delay may not matter if the item is a non-critical consumable. But even a two-day delay can put production at risk when the raw material is in short supply. This is where an AI agent becomes more valuable. Instead of looking at the supplier update on its own, it can consider the wider business context, including material criticality, days of stock coverage, production requirements, available inventory, alternative supply options, and customer commitments. This allows the agent to distinguish between delays that can be safely monitored and those that could have a significant impact on operations.
Email rules can automatically route supplier messages, while RPA can copy information when the data follows a consistent format. A traditional text parser can extract a purchase order (PO) number when the supplier always provides it in a predictable location. These tools are still useful for routine tasks, but they have limitations. The bigger challenge in supplier ETD tracking is understanding the context behind each message and determining what the update actually means for production, inventory, and operations. That goes beyond simply finding and copying information.
Agentic AI becomes useful when a process involves multiple connected steps that require both reasoning and action. In supplier ETD tracking, the agent can identify the supplier, find the relevant open orders, interpret the message, and link each update to the correct order line or batch. It can then compare the new delivery date with the previous commitment and assess whether the information is reliable enough to act on. If the details are unclear or confidence is low, the agent can send the case to a person for review. Once approved, it can update the business system and trigger an escalation when the change creates a significant planning risk.
This move from simply generating information to taking controlled action is central to how enterprises are approaching agentic AI. Deloitte describes agentic supply chains as systems that can detect changing conditions, reason across multiple data sources, and take action within established governance rules. In this model, people spend less time on repetitive execution and more time on decision-making, orchestration, and oversight.
The system should only connect to communication channels that the business has explicitly approved and can manage, such as a shared procurement mailbox, authorized WhatsApp Business accounts, supplier portals, automated data connections, or other approved interfaces. Personal messaging accounts should not be accessed or have their content collected without proper authorization. Before implementation, the business should clearly define access permissions, data retention periods, and employee privacy requirements.
The agent should not rely on the language model to make a guess based only on the message. It should first pull the relevant supplier details, open purchase orders, line items, batch schedules, quantities, previous ETDs, and recent conversation history. With this context in place, the agent has a much clearer picture of what the supplier is referring to and is less likely to link the update to the wrong order.
Supplier messages can cover a wide range of topics, including quotations, technical questions, payment discussions, quality issues, production updates, and general conversations. The agent needs to identify which messages actually contain an ETD update and which do not. For example, if a supplier says, “raw material should be ready by Friday,” that does not necessarily mean the shipment will leave on Friday. The agent must understand the context before treating the statement as a confirmed estimated departure date.
The useful output is more than just a date. The system needs to show exactly what that date refers to. At a minimum, the record should include the supplier, purchase order, relevant line or batch when available, quantity or shipment reference if mentioned, the original date stated by the supplier, the standardized date, and the source of the message.
The agent can convert natural language into a structured date or planning range, but the record should always keep the supplier’s original wording and show how precise the information is. For example, “15 March” is a specific date, while “around 15 March” is only an estimate. “Mid-March” indicates a broader date range. A phrase such as “after Chinese New Year” needs the correct calendar and information about the supplier’s operating schedule before it can be turned into even an estimated planning range.
The review screen should make the key operational questions easy to answer: Is the new ETD earlier, unchanged, or later than the last approved date? If it is later, how many days has it moved? Has the quantity changed too? And does the revised date still meet the production requirement? This gives the coordinator the information they need in one place instead of making them compare multiple messages with the ERP system manually.
Before accepting an update, the system should run several checks. It can confirm that the purchase order exists, verify that it belongs to the supplier, check whether the batch quantity is reasonable, and make sure the dates follow a logical sequence. It should also check for conflicts with any other recently confirmed shipment updates. These validation steps make the language model’s output more reliable and turn it into a safer recommendation for the business to review and act on.
Low-confidence cases should be handled safely rather than automatically accepted. If two purchase orders appear to match the message, the agent should show both options and ask the coordinator to select the correct one. If the date cannot be interpreted with confidence, the system should display the supplier’s original wording and request confirmation. Likewise, if a supplier update conflicts with an earlier formal confirmation, the agent should flag it for review instead of automatically replacing the existing information.
The real value comes from getting verified supplier updates into the systems that planners use every day. Once an ETD change is reviewed and approved, the agent can update Oracle, SAP, Dynamics, a custom ERP, or a planning database through controlled APIs. It should also maintain a clear audit trail showing who approved the change, when the update was made, and which supplier message supported the decision. This makes every update easier to track, verify, and review when needed.
A large part of procurement coordination involves following up with suppliers when delivery updates are missing. The agent can monitor purchase orders and batches that are due for an update, then prepare or send a polite reminder through the supplier’s approved communication channel. The follow-up process should take into account supplier preferences, time zones, holidays, and escalation rules. This allows the agent to stay proactive without sending unnecessary reminders or overwhelming suppliers with automated messages.
The goal is not to generate more alerts, but to make each alert more meaningful. An effective agent should assess ETD changes alongside the wider planning context and flag only the situations that could create a real operational risk. For example, it might flag a material that is likely to fall below safety stock, a batch that could affect a production schedule, a supplier that repeatedly changes its delivery commitment, or a shipment that has not moved forward as expected. This helps procurement teams focus on the issues that genuinely require attention.
Most ERP systems focus on the latest supplier delivery date, but that does not always tell the full story. When analyzing supplier performance, the history of promised dates and revisions can be far more useful. Tracking each commitment and change helps procurement teams measure delivery date volatility, average slippage, supplier response time, confirmation reliability, and the gap between stated ETD and actual departure patterns.
Procurement coordinators no longer need to spend hours searching across emails, messaging platforms, spreadsheets, and ERP systems for supplier updates or manually copying delivery dates into trackers. Instead, they can focus on reviewing exceptions, following up with high-risk suppliers, and resolving delivery issues that could impact production or customer commitments.
When reviewed supplier updates reach the ERP within minutes, planners can rely less on outdated spreadsheets and manual trackers. This gives the business a clearer, shared view of the latest supplier commitments, helping teams make faster decisions when delivery schedules change.
A delayed ETD is valuable to know about only if the business can act before it causes a shortage. By connecting supplier updates with inventory levels and production schedules, the system can identify potential supply risks earlier. Instead of simply recording a changed delivery date, teams can understand the likely impact and take action before the delay disrupts operations.
Automated reminders help ensure that unresponsive suppliers and overdue confirmations are followed up on consistently. Procurement teams can still manage supplier relationships directly, but they no longer have to rely on individual memory, spreadsheets, or personal to-do lists to track every follow-up and status request.
A manufacturer can keep the supplier’s original ETD while also using an internal, risk-adjusted date for planning based on the supplier’s past performance. The important thing is to keep the distinction clear: the adjusted date is a forecast based on historical behavior, not a replacement for the supplier’s actual commitment.
Historical ETD changes give supplier reviews a much more objective foundation. Instead of relying on statements like “this supplier is always late,” the team can look at the actual history, how often delivery dates changed, how much notice the supplier gave, how quickly they responded, and whether actual departures regularly differed from the ETDs they originally provided.
Supplier communication often includes business context, relationship dynamics, and details that may not be fully clear. A responsible system should keep a human involved whenever it cannot confidently determine that an action is safe and can be reversed if needed.
· Ambiguous PO or batch references: when the message could belong to more than one open order.
· Material changes with production or customer impact: when a date change may trigger expediting, alternate sourcing, or schedule changes.
· Conflicting supplier commitments: when two messages or two supplier contacts provide incompatible information.
· Forecast overrides: when the agent proposes a risk-adjusted planning date that differs materially from the supplier’s stated ETD.
· Commercial or relationship-sensitive follow-ups: when escalation wording could affect an important supplier relationship.
· High-impact ERP changes: when an update could trigger downstream planning, purchasing, or financial actions beyond simply recording status.
The goal of a mature system is not to eliminate human involvement. It is to reduce the workload to a smaller set of decisions where human judgment is genuinely needed.
· You manage dozens or hundreds of active international purchase orders at the same time.
· Supplier status updates arrive through several channels and are not consistently structured.
· Procurement staff maintain manual Excel trackers because the ERP is not updated quickly enough.
· ETD changes directly affect production schedules, inventory, or customer commitments.
· Suppliers are unlikely to adopt a new portal or rigid form simply for your convenience.
· You already have a reliable PO master in Oracle, SAP, Dynamics, or another system that the agent can use as context.
· You can define clear approval and escalation rules.
· Your suppliers already provide high-quality structured ASN/EDI/API updates, and the ERP remains current automatically.
· There are very few active POs, and manual status tracking is not a meaningful cost or risk.
· The organization has no reliable supplier/PO master data to validate messages against.
· Procurement communication happens mainly through uncontrolled personal accounts that cannot be governed appropriately.
· The business expects the AI to make consequential sourcing or production decisions without human accountability from day one.
The client was a large Saudi manufacturer that imported raw materials and components from more than 60 suppliers worldwide. At any given time, the procurement team was handling over 200 active purchase orders, many of which included partial shipments scheduled across several months.
The client has not been named in this article because the project involved commercially sensitive procurement and supplier information. The operational figures presented below are based on project records and outcomes reported by the client after the workflow was deployed.
The client already had a purchase order system in place. The real challenge was that much of the day-to-day information needed for planning was coming in through channels outside the system.
Supplier ETD updates were coming through several different channels, including email, WhatsApp, WeChat, spreadsheets, and phone calls. Based on the client’s workflow analysis, about 40% of updates came through email, 35% through WhatsApp, 15% through WeChat, and the remaining 10% through phone calls or manually summarized messages. These figures reflect this specific client’s workflow and should not be considered an industry benchmark.
The format of these updates could vary significantly. One supplier might send a clear spreadsheet with the purchase order number, line item, batch size, and departure date. Another might simply write, “PO-4521 first batch 500K around March 15, second batch end of April.” A supplier in China might send a mix of Chinese and English in a WeChat message, while a WhatsApp update could be a photo of a shipment schedule instead of written details. In some cases, suppliers referred to earlier conversations without mentioning the purchase order number at all.
Three procurement coordinators managed a master tracker with more than 2,000 rows. Together, they spent around four to five hours each day reviewing supplier messages, identifying the correct purchase order and batch, interpreting date information, updating the tracker, and making sure the developing Oracle APEX environment stayed up to date.
The biggest issue was not the amount of manual work. It was the delay and uncertainty it created. By the time an ETD was entered into the planning system, a newer supplier message might already have changed the date. Before the new workflow was introduced, the client estimated that around 15% of ETD records were outdated or inaccurate at any given time.
The client had already tried using standard email folders and filtering rules. These helped organize the inbox, but they did not actually understand the information in the messages. A filter could identify a supplier’s name, for example, but it could not reliably tell whether a sentence changed the ETD for the first batch of PO-4521 while leaving the second batch unchanged.
The team also tried encouraging suppliers to use a structured update form, but adoption remained below 20%. The key lesson was straightforward: an internal visibility project cannot assume that every supplier will change the way they communicate. Instead, the system needed to work with the communication channels suppliers were already using.
Traditional RPA was also not a good fit for the unstructured parts of the workflow. The process involved multilingual messages, unclear references, different date formats, attachments, and previous messages that provided important context. These inputs needed to be understood in context before the system could apply rules or make any updates.
· Monitor approved supplier communication channels without forcing suppliers onto a new portal.
· Detect messages that contained ETD or shipment-timing information.
· Match each update to the correct supplier, PO, line item, or batch.
· Interpret English, Chinese, and mixed-language communication.
· Normalize approximate date phrases while preserving the original language and uncertainty.
· Show whether the new ETD was earlier, unchanged, or delayed versus the previous approved ETD.
· Allow procurement coordinators to approve or correct the interpretation quickly.
· Write approved updates into Oracle APEX and synchronize manual corrections back to the agent layer.
· Follow up when expected supplier updates were missing.
· Escalate changes that could affect production or inventory.
· Keep the history of ETD revisions for supplier-performance analysis.
Our company started with the shared procurement mailbox because email was the largest single communication channel and provided the most controlled environment for testing. Using Microsoft Graph, the system monitored incoming supplier emails, grouped related messages into threads, and identified whether each message contained relevant ETD information.
Before interpreting a supplier message, the agent first checked the relevant open-order information. It identified the purchase order, line item, and batch where possible, then extracted the supplier’s timing information and converted vague phrases into practical planning ranges. During the first phase, the agent did not make any direct changes to the ERP. Instead, its outputs were added to a structured review file, allowing the procurement team to compare the AI’s interpretation with the existing manual tracker.
This phase was intentionally designed to build trust. The goal was not to automate the entire process within four weeks, but to demonstrate that the system could accurately interpret a representative sample of real supplier messages, clearly flag uncertain cases, and make corrections simple for the procurement team.
Once the email workflow was stable, the intake layer was extended to approved WhatsApp Business messaging and the client’s WeChat integration. This moved the solution beyond simple email extraction and turned it into a broader agent for managing supplier communications across multiple channels.
Each message was stored with important details such as the supplier, timestamp, communication channel, original message, detected language, attachment status, any required translation, and the purchase order references identified from the message. Attachments were linked to the related message instead of being kept separately, helping preserve the context needed to understand them correctly.
A React-based review dashboard became the procurement team’s main workspace. Coordinators could quickly see the original supplier message, translated text when needed, the extracted or interpreted ETD, the previous ETD, the confidence level, the direction of the change, and the suggested action. The dashboard was designed to make reviews quick and practical, rather than showing users every technical detail generated by the AI model.
The next stage linked verified ETD updates with the client’s Oracle APEX purchase-order tracking application using REST APIs. The agent did not automatically send every AI-generated result to Oracle. Only updates that met the required confidence level and completed the review process were approved for entry into the system.
The integration worked in both directions. If a procurement manager updated an ETD in Oracle after speaking with a supplier or making an internal decision, the revised date was also sent back to the agent’s database. This kept both systems aligned and prevented the AI layer and operational system from developing separate versions of the same information.
Every approved change was also recorded in the system’s history. This gave the team a clear view of the original commitment, each subsequent revision, who approved every change, and how the current date developed over time.
Once the team had confidence in the data flow, the agent began handling follow-ups for overdue status updates. Rules identified when a supplier or purchase-order batch was due for a new confirmation. The system could then prepare or send a polite follow-up through an approved supplier communication channel, using the supplier’s preferred method and the correct order context.
This removed a repetitive part of the procurement workload without attempting to automate supplier relationship management itself. Coordinators continued to handle strategic discussions, negotiations, and unusual supplier situations that required human judgment.
Once enough ETD history was available, the client could compare the dates provided by suppliers with their actual departure patterns. If a supplier consistently shipped later than its stated ETD, the system could suggest adding extra planning time to account for the supplier’s historical delays.
A key design decision was to keep the supplier’s stated ETD separate from the predicted planning date. The forecast did not change the supplier’s original commitment. Instead, it gave planners a second, clearly labeled date based on the supplier’s historical behavior. This added useful planning insight while keeping the process transparent and easy to audit.
The agent could flag a delayed update once it crossed predefined risk thresholds. For high-priority materials, the alert included the latest supplier message, the previous commitment, the proposed new ETD, the relevant purchase order and batch details, and recommended next steps. These could include confirming the date with the supplier, checking alternative stock, expediting a batch, or informing the production planning team.
The agent did not independently switch suppliers or alter the production plan. Those decisions remained with the human team. Instead, its role was to provide the right information and supporting evidence to the right person early enough for them to take action.
The most difficult messages were not always the longest ones. Often, the real challenge came from short messages that relied on context already known to the people involved.
A typical example involved a supplier message that mixed Chinese and English in the same update. It could include a purchase order reference, a date written in Chinese, abbreviated quantity details, and an English note about a later batch. The system needed to recognize that the message referred to multiple shipment events, convert the Chinese date into a standard format, and distinguish between an approximate date and a confirmed supplier commitment.
When a supplier used a holiday-related phrase, the system checked the relevant holiday calendar and the supplier’s location before suggesting a planning range. The original wording was still shown to the procurement team for review. In practice, we found that understanding approximate dates is not only a language challenge. It also depends on business context and the supplier’s historical behavior.
When a supplier message referred to multiple open purchase orders and the correct match was still unclear after checking the previous conversation, the system did not try to guess. Instead, it sent the message to a review queue, showing the possible matches along with the reason why the case needed human review.
The system followed a service-based architecture, allowing incoming messages, language processing, approvals, and ERP updates to run independently and in parallel. This prevented one part of the workflow from delaying or blocking another.
· Python and FastAPI for backend services and APIs.
· Celery and Redis for asynchronous message processing and job queues.
· Microsoft Graph API for approved shared-mailbox monitoring.
· WhatsApp Business API and the client-approved WeChat integration path for messaging intake.
· A large language model for multilingual message understanding, thread-aware extraction, and approximate-date interpretation.
· PostgreSQL for structured supplier, message, ETD, approval, and history records.
· React for the human review and exception dashboard.
· Oracle APEX REST APIs for controlled two-way synchronization with the client’s internal application.
· AWS infrastructure for compute, database, storage, and event-driven processing in the deployed architecture.
The AI model was just one part of the solution. Its value in production came from the controls around it, including context retrieval, rule-based checks, approval workflows, source evidence, identity verification, API integration, error handling, and system monitoring.
The project was evaluated based on its impact on day-to-day operations, not just the accuracy shown in an AI demo. The results below are specific to this client and reflect the measurement periods used during the project.
ETD record accuracy: increased from approximately 85% before the project to 97.5% after the supervised workflow was stabilized. This measures the accuracy of approved ETD records used by the business after extraction, validation, and human review. It does not mean the AI model independently identified the correct date 97.5% of the time without human oversight.
Manual tracking effort: decreased from roughly 4–5 combined team hours per day across three procurement coordinators to about 45 minutes of exception review per day, mainly handled by one coordinator once the workflow had matured.
Supplier-message-to-system latency: decreased from potentially several hours to an average of about 12 minutes in the implemented workflow. This included AI processing and the required human approval step.
Oracle APEX ETD completeness: increased from roughly 60% to approximately 98% as approved supplier updates were added to the system without requiring separate manual data entry.
Supplier response within 24 hours: the client reported that the response rate increased from around 50% with manual follow-ups to approximately 85% after structured, channel-aware follow-up automation was introduced.
Material-shortage production stoppages: before the project, the client recorded roughly three to four shortage-related stoppages per quarter. During the six-month observation period after the integrated Phase 3 workflow was introduced, no such stoppages were reported. This is a significant operational result, but it should not be taken as proof that the ETD agent alone caused the improvement. Procurement decisions, inventory management, supplier performance, and other operating conditions also played a role.
Working capital: the client estimated that greater confidence in ETD data helped reduce some uncertainty-driven safety stock, releasing approximately USD 2 million in working capital. This was an estimate provided by the client’s finance team and should not be treated as a guaranteed financial return from the software alone.
AI accuracy figures can be easy to misread. In this project, the focus was not simply on whether the LLM extracted the correct date. The more important question was whether the ETD used by the planning team was linked to the correct purchase order and shipment details and supported by the latest approved supplier communication.
The 97.5% figure represents the quality of the supervised workflow after retrieving the right context, extracting and validating the dates, checking confidence levels, and applying human corrections when needed. This makes it a more useful business measure than raw extraction accuracy. However, the result reflects the performance of the entire system and review process, not the AI model alone.
The client’s working-capital benefit came from having greater confidence in supplier delivery dates. When planners cannot rely on inbound ETDs, they may keep additional safety stock to protect production from unexpected delays. Better ETD visibility reduced some of that uncertainty, allowing the business to review and adjust safety-stock levels for selected materials.
The client estimated that this helped release roughly USD 2 million in working capital. The agent itself did not directly generate that amount. The financial impact depended on factors such as purchasing volumes, material costs, safety-stock policies, lead times, supplier reliability, and management decisions. A different manufacturer could therefore see a significantly smaller or larger benefit.
· The agent did not autonomously change suppliers.
· It did not cancel or reissue purchase orders because a supplier date changed.
· It did not silently convert every vague date phrase into a hard commitment.
· It did not overwrite a manually confirmed ETD without preserving source and history.
· It did not change production schedules on its own.
· It did not treat a predicted internal planning date as if the supplier had confirmed it.
· It did not guess when the message could reasonably map to more than one PO or batch.
These boundaries were important. They allowed the system to automate useful day-to-day tasks while keeping people accountable for decisions that could have significant operational or commercial consequences.
Many companies see email and chat as separate from their ERP data. In reality, these channels often carry the latest operational updates. If a visibility program ignores them, it may produce a polished dashboard that is still based on outdated information.
We started with email because it had the highest message volume and was easier to govern from a technical standpoint. But during the project, a time-sensitive WhatsApp update showed why volume alone is not enough to determine priority. A channel with fewer messages can still contain some of the most urgent operational updates. Future implementations should therefore assess each channel based on both volume and business impact.
People tend to trust a system more when it recognizes that “mid-April” is a range rather than treating it as an exact date. False precision may make an AI system appear more organized, but it can lead to less reliable planning. The original wording, the system’s interpretation, and its confidence level should be stored separately.
A dashboard can accurately summarize supplier messages, but it still leaves work for planners if they have to manually transfer that information into Oracle, SAP, or another ERP system. Reliable synchronization, a clear history of changes, and a way to recover from errors are essential parts of the solution, not technical details that can be addressed later.
Procurement coordinators adopted the workflow more easily when they could see the original message, the extracted information, the previous Estimated Time of Departure (ETD), the change, and the confidence level all in one place. A well-designed human review process makes AI uncertainty easier to manage by turning it into a clear and actionable work queue.
Historical supplier behavior can help improve planning, but a predicted date should not be treated as a confirmed supplier commitment. Keeping the two values separate preserves the evidence behind the forecast and allows planners to use predictive insights without confusing them with supplier accountability.
A successful implementation starts by defining baseline metrics before the system goes live. Without a clear starting point, the organization may end up measuring how much the AI is being used instead of whether it is actually improving the business.
· ETD record accuracy: percentage of current ETD records that match the latest approved source information.
· ETD completeness: percentage of active PO lines or batches with a usable current ETD.
· Update latency: time from supplier message receipt to approved system-of-record update.
· Manual coordination effort: team-hours spent finding, interpreting, copying, and following up on ETD information.
· Exception rate: percentage of supplier messages that require manual interpretation.
· False mapping rate: updates incorrectly associated with the wrong PO, line, or batch.
· Supplier response time: percentage of follow-up requests answered within the defined SLA or target window.
· Date volatility by supplier: frequency and magnitude of ETD revisions before actual departure.
· Shortage or expedite events: production-impacting shortages or premium-freight decisions linked to late visibility.
· ERP data completeness and freshness: how much of the planning system reflects the latest approved supplier information.
· User correction rate: how frequently coordinators need to correct the agent and what categories cause those corrections.
Always measure accuracy by category. A 98% accuracy rate on simple English emails does not show how well the system handles mixed-language WhatsApp images or messages that refer to multiple Purchase Orders (POs) with unclear references. Measure the more difficult categories separately to get a realistic view of the system’s performance.
List every communication channel suppliers actually use, not just the ones required by procurement policy. Identify the systems that hold Purchase Order (PO) master data, the planning decisions affected by Estimated Time of Departure (ETD), the tools currently used to track dates, the people responsible for approving changes, and the situations where an incorrect date could have a significant business impact.
Start with one controlled communication channel and a representative group of suppliers. Let the AI agent extract and organize updates without making any changes to the ERP. Compare its results with human-reviewed data and classify any errors by type, such as Purchase Order (PO) mapping, date interpretation, language, attachment processing, conversation context, or business-rule validation.
Once the system proves reliable enough to be useful, introduce a review queue that gives coordinators clear evidence and highlights what has changed. Track metrics such as review time and correction rate. The goal is to make the confirmation process much faster than manually going through the entire inbox.
Only make updates that are approved and meet the required confidence and policy standards. Prevent duplicate changes, retry an update when a temporary failure occurs, ensure every change is authorized, keep a clear history of updates, and provide a safe way to restore the previous value if something goes wrong. If users can also edit the same ETD in the ERP, keep both systems synchronized so they do not quietly develop conflicting values.
Add WhatsApp Business, supplier portals, and other channels where they can provide real operational value. Gradually introduce overdue-update alerts and supplier follow-ups, with clear guidelines for messaging and defined limits for escalation.
Once a reliable ETD history has been established, the project can introduce supplier-specific planning adjustments or forecasting signals. Keep forecast dates separate from confirmed commitments, and check whether these signals actually lead to better decisions about shortages, urgent shipments, or inventory.
· Approved channel access only, with explicit ownership of each mailbox, messaging account, and integration.
· Least-privilege permissions for ERP and procurement-system actions.
· Supplier and employee data retention rules appropriate to the jurisdictions involved.
· Encryption in transit and at rest for messages, attachments, and extracted business data.
· Clear separation between model inference, deterministic validation, and authorized system actions.
· Source evidence retained for each approved ETD change.
· Immutable or controlled audit history showing message, interpretation, approver, time, and downstream system update.
· Confidence thresholds and explicit fallback behavior when the model cannot safely resolve context.
· Versioning for prompts, models, date-normalization logic, and business rules.
· Monitoring for integration failures, delayed queues, duplicate messages, and stale credentials.
· A rollback procedure for a bad model or rule release.
· A documented human accountability model for high-impact procurement and planning decisions.
DHL’s logistics research also shows that successful AI adoption depends on more than model capability. It requires reliable data, proper system integration, strong security, effective governance, and clear safeguards for responsible use. These considerations become even more important when AI agents move beyond summarizing information and start updating enterprise records.
An email-summary demo alone is not enough. Ask how the system identifies an implied purchase order, handles messages that refer to multiple purchase orders, and manages a new date that conflicts with an existing commitment.
If the vendor stores all these dates in a single field, the audit trail may not be reliable. You need separate records showing what the supplier stated and what the system calculated or inferred.
A reliable approach should include confidence levels, possible purchase order matches, and human review for uncertain cases. Simply saying that “the model usually figures it out” is not a sufficient control for an enterprise environment.
Supplier messages often leave out details because people assume others remember the earlier conversation. The system should retrieve relevant context in a controlled way rather than relying on each message as a standalone source of information.
Ask for examples that reflect your actual supplier base. Test real-world messages from China, India, Europe, and other key sourcing regions instead of relying only on simplified English examples.
The integration should use approved business interfaces or client-managed connectors. Avoid solutions that depend on collecting data from personal devices or using credentials that are not properly controlled.
Ask how the system handles updates made in both directions, resolves conflicts, retries failed updates, prevents duplicate messages, and responds when the ERP is temporarily unavailable.
A reliable evaluation should measure extraction accuracy, purchase order mapping, date normalization, approved Estimated Time of Departure (ETD) record accuracy, and exception rates separately. A single overall “AI accuracy” percentage can hide the specific errors and failure points that matter in production.
Permissions should be clear and configurable. Reading supplier messages, drafting follow-ups, updating a low-risk status field, and changing a production plan all carry different levels of risk and should have appropriate controls.
Supplier disputes and performance reviews can happen months after the original message. The system should preserve the original communication, show how it was interpreted, record who approved the change, and maintain a complete history of updates.
Ask about version control, testing, limited rollout, and rollback procedures. A supplier communication agent is an operational system, so it needs the same disciplined software lifecycle practices as other production systems.
Your team should understand the integrations, business rules, confidence thresholds, model dependencies, monitoring, and support process. The goal should be sustainable ownership rather than permanent dependence on a single developer for every supplier-related change.
Supplier ETD tracking means keeping the latest expected departure date for incoming purchase orders, batches, or shipments. Manufacturers use this information to plan material availability, inventory, production, and downstream logistics. The challenge is that suppliers often communicate ETD changes outside the ERP through email, chat, spreadsheets, or phone calls.
It is an AI-powered workflow that monitors approved supplier communications, identifies ETD updates, links them to the correct purchase order or batch, interprets the dates, compares them with existing commitments, requests human approval when needed, and updates the relevant enterprise system.
ETD stands for Estimated Time of Departure and refers to the expected departure date from the origin or another relevant shipping point. ETA stands for Estimated Time of Arrival and refers to the expected arrival date at the destination or a later milestone. An AI agent can calculate an ETA from the ETD using transit-time assumptions, but the calculated ETA should be clearly labeled and kept separate from the supplier’s stated ETD.
Yes, if the organization uses an approved WhatsApp Business integration and the workflow has appropriate permissions, retention, and security controls. The system can analyze message text and supported attachments, but unclear updates should still be sent for human review.
It can, provided the client has a lawful and supported integration for the relevant WeChat business workflow. The bigger challenge is usually not translation alone. Mixed-language shorthand, conversation context, and linking the update to the correct purchase order can make the process more difficult.
Yes. In the case study described here, approved ETD updates were synchronized with an Oracle APEX application through REST APIs. Manual changes made in Oracle were also synchronized back to the agent layer to maintain consistency between both systems.
Yes, if the required purchasing and order data is available through supported APIs or integration services. The basic approach is the same: retrieve the relevant purchase order context, interpret the supplier communication, validate the update, and write the approved result using controlled permissions.
Yes, for clearly defined, low-risk follow-up scenarios. The system can identify an overdue confirmation and send or prepare a message through the supplier’s approved communication channel. High-value escalations and sensitive supplier issues should remain under human control.
Historical ETD revisions, supplier responsiveness, and actual departure patterns can help generate predictive risk signals. These predictions should be treated as planning insights, not as supplier commitments. The model should be evaluated against actual outcomes before planners rely on it for inventory or production decisions.
The better use case is to reduce repetitive monitoring, data entry, and follow-ups so procurement staff can spend more time on supplier relationships, negotiations, risk mitigation, and commercial decisions. Human judgment remains important for unclear, high-impact, and supplier-sensitive cases.
A narrow, read-only pilot using one communication channel can often be built within a few weeks if the required purchase order data and channel access are available. A production system covering multiple messaging channels, human review, ERP synchronization, automated follow-ups, security controls, and monitoring usually requires a phased implementation over several months. The timeline depends more on integration complexity and data readiness than on the language model itself.
Start with the communication channel and supplier group that create the most significant coordination workload and have a clear purchase order context. Email is often a practical first pilot, but channel urgency should be considered alongside message volume. If critical updates mainly arrive through WhatsApp, delaying support for that channel can limit the business value of the automation.
Measure time saved on supplier coordination, reduction in outdated records or missing ETDs, faster update processing, fewer expedite or shortage events, improved supplier response performance, better ERP data completeness, and any inventory benefits linked to greater planning confidence. Avoid treating every prevented disruption as if the AI alone caused it.
Supplier ETD tracking may seem like a small administrative task until a manufacturer is managing hundreds of open orders across dozens of suppliers. At that point, every informal message can become a planning issue, and the gap between supplier communication and the ERP can become a significant source of operational risk.
Agentic AI can bridge the gap between supplier communications and the ERP. It can monitor the channels suppliers already use, find the relevant purchase order, understand what has changed, flag uncertain information, involve a person when needed, update the system after approval, and alert procurement and planning teams only when their attention is required.
The Saudi implementation described in this article did not succeed simply because an LLM could identify dates. It succeeded because the complete workflow connected language understanding with purchase order context, validation, human review, Oracle APEX integration, consistent follow-ups, historical tracking, and controlled escalation.
For manufacturers, importers, and distribution businesses facing a similar challenge, the first question should not be, “Which AI model should we use?” Instead, ask, “Which supplier information do our teams repeatedly look up, enter, verify, and follow up on, and which business decisions rely on that information being accurate?” That is the process where automation can deliver the most value first.
Related aTeam capability: AI-driven solutions for logistics and supply chain – including supplier monitoring, procurement agents, document processing, customs workflows, and shipment exception automation.
