Reported operating use · Manufacturing and building materials

How Oldcastle APG reports automating purchase-order entry

Convert customer purchase orders into verified ERP orders

Oldcastle APG (as reported by AWS and Oldcastle APG)

Production use reported by AWS and Oldcastle APG on 13 November 2026; the account says its initial two weeks processed 1,100 orders and more than 10,000 lines.

A published third-party example, not a Nuvepro customer case or evidence of a partnership.

The workflow at a glance

How Oldcastle APG reports moving customer orders into ERP

Illustrative summary of the published account

  1. Customer order intake

    Capture a variable-format purchase order

    Email attachments arrive as PDFs, images, HTML or text and are stored for processing.

  2. Next
    Amazon Bedrock

    Extract order headers and line items

    The system returns customer, PO number, ship-to address, products, quantities and prices as structured fields.

  3. Next
    Enterprise data matching

    Match the customer and ERP products

    Extracted details are matched to customer and product records using fuzzy similarity search.

  4. Next
    Customer service representative

    Check the source and ranked suggestions

    A CSR reviews each order beside its original and confirms customer and line-item matches; corrections are retained for reuse.

  5. Next
    Infor M3

    Create the confirmed order with an audit trail

    The reported system creates the ERP order after CSR confirmation; the source says repeat orders can skip the AI step.

Illustrative summary of AWS and Oldcastle APG’s production account, published 13 November 2026. The initial production results are not independently audited. Receiving and retail-staging workflows are described as future extensions, not current steps. See the source account and limitations below.

What the source reports

Oldcastle APG and AWS describe an event-driven purchase-order workflow for variable emailed PDFs, faxes, scans, HTML and text. The system stores incoming documents, extracts customer, order, address, line, quantity and price fields, matches customers and products against enterprise data, and presents ranked matches beside the source order for CSR verification. After confirmation, it creates the order in Infor M3 and retains an audit trail; CSR corrections are reused and repeat orders can skip the AI step. AWS and Oldcastle report that the first two production weeks processed 1,100 orders and more than 10,000 order lines, with 95% of lines fully automated and the remaining 5% flagged for CSR review. These are source-reported figures, not independently audited results.

Reported technology: Amazon Bedrock with Anthropic Claude Sonnet 4.6, AWS SES, Microsoft Power Automate, Amazon S3, Aurora PostgreSQL and Infor M3. Scope and outcomes are source-reported.

What the enterprise environment would need

The following is a proposed project design for an enterprise doing similar work. Published details are marked; the remaining systems, data and scenarios would be agreed for your project.

Proposed practice environment

Rehearse a customer order with a fictional manufacturer

Illustrative project design to scope with your enterprise.

  1. 1 · Set up the environment

    Synthetic POs and pinned customer/item records

    Use permitted synthetic PDFs and email payloads, a versioned customer master, item catalog, price list and expected ERP records in a disconnected practice system.

  2. 2 · Preparation and agent contributions

    Agent extracts fields and proposes matches

    Link extracted headers and line items to source locations; propose ranked customer and SKU matches while preserving ambiguity and missing values.

  3. 3 · Independent checks

    Independently reconcile the order

    A separate checker compares identity, PO number, ship-to, SKU, quantity, price, totals and duplicate status with a frozen reference and independently prepared expected record.

  4. 4 · People decide acceptance

    CSR owner accepts the exact simulated order

    The accountable owner accepts or rejects the checked practice version. Rejection or changed reference data returns through repair and independent checks.

Only within the accepted practice scope
Practice output

A reviewable evidence package

A versioned simulated order with source-linked fields, match decisions, independent reconciliation and owner disposition.

If a check fails → withhold acceptance.

An ambiguous SKU, stale price or duplicate PO blocks acceptance. Repair the mapping or obtain clarification, rerun independent reconciliation and return the exact revision to the owner.

Failure or owner rejection returns to review

  1. Hold
  2. → 2 · Revise work
  3. → 3 · Rerun independent checks
  4. → 4 · Owner review

Operating boundary. Designed, not executed. Use synthetic or permitted example documents and a simulated ERP only; no live email, customer message, Infor M3 write or payment authority.

Proposed, not an executed Nuvepro project or a description of the named company’s implementation. Owners agree the data, rules, responsibilities and acceptance criteria for their enterprise.
Fictional purchase orders and versioned ERP reference data
Proposed: use permitted synthetic PDFs, scans and email payloads, a fictional customer master, item catalog, price list, ship-to rules and expected ERP records. Pin the version of every reference table and keep production mail and ERP systems disconnected.
Agent extracts and proposes customer and SKU matches
Proposed: extract order headers and line fields, link each value to its source location, and return ranked customer and item matches with uncertainty and unresolved fields. Never let the model invent a missing quantity, SKU, price or delivery date.
Independent reconciliation and CSR decision
Proposed: a separate deterministic checker compares customer identity, PO number, addresses, SKU, quantity, unit price, totals and duplicate status against a frozen reference. A practice CSR accepts or rejects each matched line before a simulated order is created.
Owner acceptance, repair and replay
Proposed: an accountable order-entry owner accepts the exact checked practice order or holds it for repair. Any changed source, reference or match invalidates prior checks; rerun reconciliation and return the revised version for review.

A proposed simulation

An illustrative practice design derived from this work, not an executed assessment.

Rehearse document-to-order extraction and line matching against a fictional manufacturer’s frozen customer, item and pricing records.

Review gate: The practice owner accepts only the exact simulated order whose source fields, customer and SKU matches, quantities, prices, totals and duplicate status passed independent reconciliation. Acceptance grants no production ERP or customer communication authority.

Challenge the result: A source order conflicts with the current price list or produces two plausible SKU matches. Hold the order, repair the mapping or obtain owner clarification, rerun independent reconciliation and return the exact revised practice order for acceptance.

What needs checking

  • Source boundary: AWS published this account with Oldcastle APG co-author Avdhesh Paliwal. The article attributes results to Oldcastle production data; no independent process or result audit was inspected.
  • Proposed: test clear and ambiguous customer/item matches, unreadable source fields, French-language orders, duplicate POs, stale item/price data, arithmetic mismatches and missing ship-to details against independently prepared expected ERP records. Designed, not executed.
  • Proposed: change a price list after CSR review or introduce a conflicting SKU match. Withhold acceptance, invalidate the checked order, repair the affected mapping, repeat independent reconciliation and return the exact revision to the owner. Designed, not executed.

Review environment fidelity, human capability and agent performance separately. The enterprise’s accountable owners decide acceptance of the shared workflow and any operating use.

Shape this around your enterprise

Bring a similar workflow and the people responsible for it. Together, define the records and systems to use, the result your team will build or configure, and how its owners will check it. Nuvepro scopes the simulation, environment and preparation. Agree any existing-agent evaluation, implementation assistance and production responsibilities separately.

Sources and limitations

Curation review: . This is the date this source example was reviewed, separate from source publication dates.

  • The workflow is described in an AWS customer solution post co-written with an Oldcastle APG architect. It is customer/vendor reporting, not independent evaluation.
  • The 95% automation rate is reported for the first two production weeks; the article does not provide an independent audit, denominator definition beyond order lines, or long-run error rate. The remaining 5% are flagged for CSR review.
  • Orders are created only after CSR confirmation in the reported account. Adjacent receiving and retailer-staging flows are described as future work and are not included as live steps here.
  • The proposed practice environment and acceptance checks are fictional and unexecuted. No Nuvepro customer, partnership, delivery or production ERP authority is implied.
  1. How Oldcastle APG Automates 13 Million Order Lines with Amazon BedrockSource publication: 2026-11-13

More customer service examples

All 14 customer service examples

Explore more enterprise work