Manufacturing Software Development for Plants That Run on Spreadsheets and Memory

By scope: discovery and process mapping runs $900–$2,250, a single system build $3,000–$7,500, an integrated platform $7,500–$22,500, and a multi-plant programme $22,500–$60,000, with ongoing support from $600/month. USD is the contracting currency; indicative GBP, EUR, INR, and AUD conversions are in the table above. Discovery converts the bands into a fixed written quote.

Indicatively: discovery and process mapping ₹85,000–₹2,15,000, a single system build ₹2,85,000–₹7,15,000, an integrated platform ₹7,15,000–₹21,40,000, and a multi-plant programme ₹21,40,000–₹57,00,000, with ongoing support from ₹57,000/month. These are conversions of the same USD bands, not regional pricing — USD remains the contracting currency.

Buy when a mature vertical product models your process closely — several exist for discrete assembly and packaged goods. Build when the process is the competitive edge, when the product would force the shop floor to work the product's way, or when integration with existing ERP, PLCs, and quality systems is the actual problem. Discovery prices both routes before recommending either.

Discovery and process mapping takes 2–4 weeks. A single system — an MES module, a quality system, a scheduling tool — runs 6–12 weeks. An integrated platform typically takes 3–6 months, and a multi-plant rollout 6–12 months, plant by plant with the pattern proven at the first site before the second begins.

Yes — that is usually the point. Builds integrate with ERPNext, SAP, Dynamics, and Oracle on the business side, and with PLCs, SCADA, and historians on the plant side via OPC UA, MQTT, and vendor APIs. The shop floor keeps running during rollout: new systems run in parallel against real production data before anything is switched over.

You do, from the first commit. Repositories, cloud accounts, and every byte of production data sit in your name. If the engagement ends, the plant keeps running on systems you own — there is no per-seat licence and no exit fee.

MES (Manufacturing Execution System) tracks what happens on the shop floor as it happens — the operator scanning a batch, the machine reporting a run rate, the inspector rejecting a unit. ERP tracks what the business as a whole is doing — the order that batch belongs to, the customer, the invoice. MES lives in seconds and minutes; ERP lives in orders and days. The integration between them is where most manufacturing software projects earn their money: the ERP places a work order, MES tells the shop floor exactly what to do, and shop-floor events flow back up so the ERP always reflects the current state without anyone typing anything twice.

Three patterns cover most estates. Modern machines with an OPC UA server publish their state directly to a broker (typically MQTT via Sparkplug B) and the software subscribes. Older machines with a serial port or Modbus RTU get an edge gateway sitting alongside — a small industrial PC that translates and forwards. Machines that expose nothing get a retrofit: a sensor on the running light, a counter on the cycle button, a probe on the coolant line. The rule is to instrument the machine, not to modify it — the retrofit lives outside the safety-certified envelope and adds no regulatory burden.

ROI shows up in four places. First, throughput: eliminating the spreadsheet handoff between planning and the floor typically recovers 5-15% of shop-floor time. Second, scrap: real-time SPC on machine data catches drift before the batch leaves the cell, cutting rework and scrap by half or more on the affected lines. Third, inventory: MES-driven demand signals let planners hold less safety stock without stocking out, freeing working capital. Fourth, audit: batch genealogy takes minutes instead of days, and recall exposure shrinks to the specific batches affected. Payback windows of 9-18 months are typical when at least two of the four apply.

The backbone is boring on purpose: Node.js or Python services on Kubernetes, PostgreSQL for the transactional data, TimescaleDB or InfluxDB for the machine time-series, MQTT (Sparkplug B) as the plant-floor message bus, and OPC UA for the machines that speak it. The UI is React with TypeScript. On the edge, Node-RED or a lightweight Python service runs on an industrial PC or gateway. Cloud is optional — many plants require the core to run on-premise, and the platform is designed so on-prem is a first-class deployment target, not a fallback.

OT (operational technology, meaning the machines) and IT (the corporate network) sit on separate VLANs with a firewall between them. The software's edge components live on the OT side; the aggregated data and web UI live on the IT side. Communication between them is one-directional where possible (OT publishes, IT subscribes) and uses signed messages with rotating credentials. Machines themselves are not exposed to the internet — updates arrive via the edge gateway, not by opening ports. The pattern satisfies both IEC 62443 for industrial estates and the CIS controls typical IT audits look for.