From ERP to shop floor: one connected thread.
MES integration, production analytics and supplier systems — digitising manufacturing with deployments planned around production.
- Plant survey
- First line, measured
- The vertical thread
- Scale across the plant
- Operate, or hand over
Manufacturing software earns its keep in OEE points and schedule adherence.
Nothing else counts. A plant does not need another dashboard; it needs to know why line three stopped last night, whether the order it is running is still the one the customer wants, and which of the four systems holding a version of that answer is telling the truth. The gap is rarely the machines. It is the layers between them — ERP, MES, historians, spreadsheets, a whiteboard by the packing station — each holding part of the record and none of them speaking to the next without a human retyping something.
We connect the layers that usually don’t talk, and we build the analytics and workflow tools that turn plant data into decisions. Orders come down from ERP with their routings and materials; actuals go back up with the counts, scrap, downtime reasons and labour that finance needs to close the period. Machine data arrives through gateways and OPC UA rather than through a person with a clipboard. Supplier quality, deviations and approvals move through a governed workflow instead of an email chain with attachments. The vertical thread is the product: one fact, entered once, visible at every level that needs it.
One rule governs every deployment: the line comes first. Integration is read-safe against machine networks by default, releases land in changeover windows, and every rollout goes line by line with a rehearsed way back. A plant manager who has been burned by an IT project once will not lend you the second shift, and we plan on that basis.
Four kinds of work, all of them on the plant floor.
These are the engagements we take on in manufacturing.
Production visibility
OEE dashboards, downtime analytics and andon flows fed by machine and MES data — the plant, live. We model availability, performance and quality the way your cost accounting already defines them, so the number on the screen survives its first argument with the controller. Downtime reasons are captured at the line, not reconstructed from memory at the end of the shift, and the same event stream feeds the andon board, the shift report and the analytics layer built with Data Engineering.
ERP–MES–machine integration
The vertical thread: orders down, actuals up, no re-keying between layers. Work orders, routings, bills of material and material movements flow from ERP into execution; confirmations, scrap, consumption and downtime flow back on governed contracts rather than direct table access. Our SAP & ERP practice keeps the core clean while plant systems consume events through an integration layer we can version, test and replay.
Supplier & procurement systems
Vendor portals, quality flows and approvals — often on our Workflow & Procurement product — replacing email-and-attachment chains. Requests for quotation, purchase requisitions, incoming-goods inspection, deviations and supplier corrective actions run as workflows with owners, deadlines and an audit record. Buyers stop chasing status, quality stops rebuilding history from mailboxes, and finance gets a commitment view that matches the ERP.
Predictive quality & maintenance
ML on process data where the evidence supports it — scoped by our discovery practice, not by hype. We start by asking whether your failure and process history contains a signal at all: enough labelled events, enough sensor resolution, enough time. When it does, models ship with the monitoring and retraining discipline of our machine-learning practice. When it does not, we say so and spend the budget on visibility.
What we build, in detail.
OEE that survives the argument with finance.
Most plants already have an OEE figure. Few have one that operations, quality and finance will defend in the same meeting, because each layer counts differently — the machine counts strokes, MES counts good parts, ERP counts confirmed quantity, and the three drift apart within a shift. We build the measurement chain once: a single definition of availability, performance and quality per asset class, calculated from timestamped events rather than end-of-shift summaries, with every number clickable down to the raw event that produced it.
- Downtime captured at the line with reason codes the operators actually recognise, on a gloves-and-glare interface that works offline and syncs when the network returns
- Availability, performance and quality reconciled against ERP confirmations on a schedule agreed with you, with the variance itself reported rather than quietly absorbed
- Freshness from machine to dashboard agreed with you according to your needs, and an explicit alert when a feed goes silent — a stale chart is worse than no chart
- Shift reports, andon boards and the analytics warehouse fed from one event stream, so nobody maintains a second version of the truth in a spreadsheet
One thread from the order to the machine and back.
The re-keying between layers is not a clerical annoyance; it is where schedule adherence goes to die. A planner changes a date in ERP, the change reaches the shop floor as a phone call, the line runs the old sequence, and the shortage shows up days later in the wrong place. We build the vertical thread as a governed integration layer: ERP is the system of record for orders and materials, execution owns the shop-floor state, and every hand-off between them is a contract with a schema, a test suite and a replay path rather than a nightly file somebody watches.
- Order release, routing and material data flow down through versioned contracts; confirmations, scrap and consumption flow back the same way, with reconciliation on both directions
- Machine data enters through OPC UA gateways or retrofit sensors, read-only against the OT network, with the boundary between plant and enterprise networks documented and enforced
- Events ride an event backbone with an outbox at the producer and idempotent consumers, so an outage becomes a backlog rather than a missing confirmation
- Every interface has a named owner, a contract test in CI and a dashboard for lag and error volume — the estate stays reviewable after we leave
Supplier work that stops living in mailboxes.
The back office of a plant runs on attachments. A quotation sits in a buyer’s inbox, an incoming-goods deviation is a photograph forwarded three times, a corrective action is agreed on a call and remembered differently by both sides. Put on Workflow & Procurement, the same work becomes a queue with owners, deadlines, versions and an audit trail — and the approval rules become configuration a category manager can change without a release.
- Requisition, quotation, purchase order, goods receipt and supplier claim as one connected record, integrated with the ERP that remains the financial system of record
- Approval thresholds, delegation and escalation as rules rather than code — with simulation before a change goes live
- Quality deviations and supplier corrective actions tracked to closure, with evidence attached and an exportable history for audits and customer surveillance
- Vendor self-service for documents, certificates and confirmations, so expiry chasing becomes a scheduled job instead of a person’s memory
Three seats, one set of facts.
The shift supervisor’s view.
Operations needs to see the order, its state and the reason it is behind — in one place, on a screen that works with gloves on and keeps working when the wireless drops in the north bay. We design the shop-floor interface for the environment first: large targets, high contrast, offline tolerance, and a confirmation flow short enough that it is done at the machine rather than reconstructed later.
- Work-order state, counts and downtime reasons captured where the work happens
- Andon escalation with a named responder and a clock, not a light nobody owns
- Shift handover generated from the record rather than typed from memory
The boundary, respected.
Plant IT carries the risk of anything that touches the machine network, and rightly refuses projects that treat a PLC as just another data source. Our default posture is read-safe: data leaves the OT side through gateways and brokers, writes back to control systems are exceptional, explicitly designed and jointly signed off. Deployment follows the production calendar, not the sprint calendar.
- Segmented networks with a documented data diode or gateway pattern per site
- Deployments in changeover windows, line by line, with a rehearsed rollback
- Monitoring, logging and access control aligned with our security practice
Suppliers, measured rather than remembered.
Category managers and quality engineers need the same record from opposite directions: one is negotiating with a vendor, the other is deciding whether to accept a batch. A shared supplier record — orders, receipts, deviations, certificates, corrective actions — makes both conversations evidential. Scorecards are computed from that record, so a supplier review starts from the same data on both sides.
- Supplier scorecards computed from delivery, quality and responsiveness history
- Certificate and audit expiry tracked with automatic reminders and escalation
- Deviation-to-closure cycle visible per vendor, per plant and per commodity
The systems a plant actually runs.
SAP S/4HANA / ECC
Orders, BOM, routings, goods movements and confirmations via IDoc, BAPI or OData — clean-core integration through an interface layer, not direct table reads.
MES / MOM platforms
Work-order dispatch, execution state and genealogy through vendor APIs or database views, extended with focused tooling where the package stops short.
PLC / SCADA via OPC UA
Read-safe subscription to tags through an edge gateway, with buffering at the edge so a network break delays data instead of losing it.
Process historians
Time-series extraction for analytics and model training, downsampled and aligned to production events rather than pulled as raw floods.
WMS and warehouse automation
Material availability, consumption and finished-goods movements synchronised with production, so shortages surface before the line stops.
CMMS / maintenance systems
Asset hierarchy, work orders and failure history joined to downtime events — the raw material for any credible maintenance model.
Quality systems and LIMS
Inspection plans, results and non-conformance records linked to batches and orders, retained to the standard your customer audits demand.
e-Invoicing and finance
Turkish e-fatura, e-irsaliye and equivalent regimes handled at the boundary, with reconciliation to the ERP posting rather than a parallel ledger.
How we work in a plant.
- OT/IT boundary respected. Integration with machine networks is read-safe by default, through gateways and brokers rather than direct connections. Any write back to a control system is an explicit, jointly approved exception with its own test and rollback plan.
- Deployments planned around production calendars. Releases land in changeover, maintenance or planned-stop windows agreed with the plant, line by line. The rollout plan is written with the shift supervisors who will live with it, not presented to them.
- Offline-tolerant shop-floor interfaces for gloves-and-glare reality: large targets, high contrast, queued writes that survive a dropped access point, and a visible sync state so an operator can see whether the system has their confirmation.
- Data models that reconcile with ERP cost accounting. Scrap, consumption and labour are modelled the way finance already defines them, and the regular reconciliation reports its variance instead of hiding it.
- Prove it on one line before scaling. The first line is a real deployment with a measured baseline, not a proof of concept — and the decision to scale is taken on its numbers.
- Machine connectivity assessed per asset, honestly. Some machines expose everything, some expose a counter, some need a retrofit sensor. We survey the estate first and price the awkward ones properly rather than discovering them in delivery.
- Multi-plant from the start. Site-level configuration, shared master data and a rollout template, so the second and third plants are largely configuration rather than a fresh project.
- Handover to the people who stay. Runbooks, dashboards and training for plant IT and maintenance, so the system is operable by your team long after the engagement closes.
How a manufacturing programme runs.
Plant survey
We walk the lines with operations and plant IT, inventory the machines and what each can expose, map the systems of record, and agree the definitions of OEE, scrap and downtime that everyone will sign. Output: a connectivity map, a data-model draft and a ranked list of the lines worth doing first.
First line, measured
One line, end to end: gateway or sensors, the event pipeline, the shop-floor interface and the first dashboards, against a baseline captured before anything changed. Everything is deployed in agreed windows, planned so the line keeps running.
The vertical thread
ERP and MES integration in both directions with contract tests, reconciliation and replay; supplier or approval workflows on Workflow & Procurement where they are in scope. Re-keying starts disappearing from the daily routine.
Scale across the plant
Remaining lines and areas on the template built in phase two. Analytics move from descriptive to diagnostic, and any predictive work is scoped only where the failure history supports it.
Operate, or hand over
We run it with a support agreement, or we train your plant IT and maintenance teams and hand over the runbooks and dashboards. Either way the estate is documented, monitored and reviewable — and second and third sites reuse the template.
Frequently asked questions
Neighbouring sectors we work in.
Bring us a problem from your industry.
Describe what is going wrong today. An engineer reads it and replies in writing with a concrete path.