Meridian Precision Components: Make the ERP earn its place

The board's question was whether digital transformation could recover operating performance without disrupting either factory. It did not begin with approval to buy an ERP. The case therefore compares continuing with current systems, improving processes, buying a cloud ERP core, adding advanced analytics and sequencing a gated combination of process improvement and ERP.

The financial warning is visible before any technology discussion. Revenue rose from RM 39.8m in 2024 to RM 48.6m in 2026, but COGS rose faster, from RM 29.5m to RM 37.1m. Gross margin fell from 25.9% to 23.7%; EBITDA margin fell from 13.6% to 11.0%. Administrative and selling costs broadly followed revenue, while inventory, emergency purchasing, write-offs and overtime grew much faster.

The diagnostic uses 2024 operating intensity, scaled by the 1.221 revenue-growth factor, as its comparison baseline. This separates increased activity from deteriorating control. It is a reasonable case-analysis method, but it does not establish causation: product mix, input prices and operating complexity could also have changed.

The CFO suspected material-price inflation; the Operations Director blamed inventory and procurement inefficiency. At the 2024 gross margin, 2026 gross profit would have been about RM 1.077m higher. Three reported cost pools explain RM 504,387 of that gap: excess emergency procurement, inventory write-offs and overtime. This is 46.8% of the gap, before making any rework assumption.

The report estimates excess rework at RM 207,760, using the 1.4-percentage-point increase in rework, 2026 COGS and an assumed 40% conversion-cost share. Including it lifts the operational attribution to RM 712,147, or 66.1%. The remaining RM 365,240 is unresolved. Commodity prices, supplier purchase-price variance, product mix and customer price pass-through data were not supplied, so the material-cost explanation remains an open question.

This prevents a common investment error: treating the whole margin gap as recoverable ERP value. Even the identified pools are only partly addressable. Supplier disruption can still generate emergency orders, genuine demand can still require overtime, and calibration or operator skills need a quality-engineering response.

Purchasing currently passes through seven manual steps: a planner identifies a need, re-enters it into departmental sheets, procurement consolidates it, a PO is built in Excel and emailed, goods are received into a warehouse sheet, and finance manually matches the invoice. None of these handoffs reliably checks a shared stock position. Duplicate item codes make the same physical material appear to be different products.

The first response is one requisition form, duplicate-order checking, an approval matrix, emergency-order cause codes and weekly supplier-date confirmation. Inventory needs a governed item master, bin and location discipline, cycle counts, adjustment approvals and ABC-XYZ policies that distinguish item value and demand variability. Production needs one shared schedule, a 48-72-hour freeze window and a logged exception process.

Plant B is the most useful first inventory cutover: its 83% stock accuracy and 19-minute picking time are weaker than Plant A's 88% and 14 minutes, while the source assesses its customer mix as lower risk than Plant A's semiconductor work. That is a sequencing decision, not a reason to relax the migration gate. ERP should embed these controls and connect the plants once the rules work consistently.

Reporting is another process problem before it is a dashboard problem. Finance consolidates five sources into about 40 recurring reports, requiring 260 hours a month. The recommendation first rationalises these into an owned pack of about 15 KPIs, preserving the time needed for validation and management commentary.

Average inventory increased by RM 2.9m. Holding the exact 2024 inventory-to-COGS ratio constant, COGS growth explains RM 1.340m, or 46.2% of that increase; the remaining RM 1.560m is associated with higher inventory intensity. The source's sentence saying volume explains roughly 55% reverses these proportions. The article uses the calculation rather than repeating that sentence.

At RM 37.1m of annual COGS, one inventory day is worth RM 101,644. Moving from the model's 79-day baseline to an illustrative 65 days therefore releases RM 1.423m once. The model separately applies a 5% liquidation haircut and a 2% recurring cash holding-cost benefit. Financing savings are excluded from the investment cash flows to avoid adding a second value for the same released cash.

The reported RM 8.1m inventory balance actually implies 79.69 days, so a balance-based calculation gives RM 1.493m released at 65 days. The RM 70,137 difference comes from the rounded 79-day model baseline. Neither number is an achieved result. The option appraisals retain the supplied model's convention so their returns remain reproducible.

Stock reduction must be selective. Slow movers, duplicate stock and inaccurate records need investigation before safety stock is reduced on critical items. Weekly stockout monitoring and protected A-class items guard against exchanging a cash release for production stoppages.

Processing 4,800 purchase orders at 42 minutes each consumes 3,360 hours a year, equivalent to 1.75 full-time roles at 1,920 hours per employee. Even a 50% reduction frees only 0.88 of a role. The eight-person team is retained, so this is capacity rather than a payroll saving. The base case credits a 35% PO-time reduction for cloud ERP and monetises only 35% of released capacity across purchasing, reporting and search.

The model values search and recount effort at 4,830 hours a year, using assumed hours per stockout and adjustment. These hours require a four-week time study; the source does not contain transaction-level observations. Released time should be assigned to supplier management, counting and exception prevention, with cash value recognised only when avoided overtime or an evidenced hiring decision supports it.

Benefits also need an overlap map. Duplicate purchases are reflected through inventory and write-offs; their full order value is not added again. Production interruption is represented through excess overtime, rather than a second downtime line. Schedule adherence and stock accuracy enable benefits but are not separately monetised. Customer retention is an assumed margin benefit linked to delivery performance, with no customer-level churn data to validate it.

The source is internally inconsistent about rework: some narrative says to exclude it from ERP, while the financial model includes a limited 12% capture of the excess pool for cloud ERP, worth RM 24,931 at 2026 volume before ramp and growth. The article reports the model as supplied and makes that credit explicit. Calibration, supplier quality, statistical process control and operator training remain a separate engineering programme.

The technology requirements follow the diagnosis: one governed item, supplier and customer master; multi-plant inventory with locations and scanning; purchasing approvals; connected sales orders and finance; basic planning; and usable audit trails. Advanced planning and forecasting are later options because the required routing and historical data are not yet dependable.

The weighted evaluation gives process improvement 4.05 out of 5, the phased route 3.80, immediate cloud ERP 3.15 and ERP with analytics 1.95. Financial return carries 25% of the score; root-cause fit and implementation risk carry 15% each. These scores are structured analyst judgements, not vendor test results. Process improvement scores highest on cost and risk but only 1 out of 5 on scalability.

Vendor estimates exclude important enabling costs. The cloud core's RM 820,000 estimate becomes RM 1.193m after contingency, RM 120,000 of internal backfill, RM 90,000 of scanners and RM 40,000 of cybersecurity setup. Annual administration and governance also sit alongside the licence. The advanced option's RM 1.927m loaded initial cost breaches the RM 1.5m cap, and month-14 go-live breaches the transformation horizon.

The vendor scorecard requires manufacturing references, migration tools, plant-by-plant rollback, open interfaces, access controls and recovery evidence. Statutory and security fit are acceptance gates. The source mentions Malaysian tax and e-invoicing requirements as items for the eventual procurement team to validate; this historical simulation does not certify present-day regulatory compliance.

The appraisal compares 2027-2031 cash flows with continuing at 2026 inefficiency rates, growing volume at 3% a year. Initial investment is placed at the start of 2027; operating flows are at year-end. Benefits ramp monthly after go-live and are reduced by the scenario's realisation factor. The discount rate is 12%, with no tax, grants, financing structure or terminal value.

The board requires positive NPV, IRR above 12%, five-year ROI of at least 20%, simple payback within 36 months and initial investment no greater than RM 1.5m. Process improvement passes. Cloud ERP's 8.7% IRR and 40.4-month payback fail, even though its undiscounted five-year ROI is positive. The phased route remains negative at today's estimates; sequencing reduces implementation exposure but does not make its economics automatically attractive.

The second NPV view is decisive. Removing net inventory cash leaves process improvement slightly negative at RM 16,265 and cloud ERP negative at RM 1.068m. The return is substantially a working-capital case. By 2031, the cloud option generates RM 512,770 of operating benefit against RM 355,256 of annual recurring cost, leaving RM 157,514 before any additional inventory cash. These forecast benefits must be distinguished from realised EBITDA improvement.

I independently recomputed NPV, ROI and interpolated payback from all four options' cached cash flows, and checked each reported IRR by substituting it back into the NPV equation. The results reconcile with the workbook. This verifies the arithmetic of the supplied outputs; it does not validate the synthetic inputs or prove the assumptions will occur.

The conservative scenario combines a 25% vendor-cost overrun, three-month delay, 80% benefit realisation, slower adoption, 7% licence inflation, 1.5% volume growth and 20% capacity monetisation. Process improvement retains positive NPV of RM 76,365, while cloud ERP falls to negative RM 921,477 and the phased route to negative RM 946,377.

The downside combines a 40% overrun, six-month delay, 60% realisation, slower adoption, 10% licence inflation, no volume growth and no monetised capacity. Process improvement falls to negative RM 174,069; cloud ERP falls to negative RM 1.646m. The phased route's initial investment reaches RM 1.544m and breaches the cap. These are stress cases, not estimated outcome probabilities.

Benefit realisation is the strongest ERP sensitivity: a 25-percentage-point change moves cloud ERP NPV by about RM 0.551m. At 6% annual volume growth, holding the other base levers fixed, cloud ERP NPV becomes positive RM 166,923. The standalone break-even alternatives are approximately 104.3% of credited benefits, a RM 185,038 annual licence, a RM 737,708 implementation estimate, or 66.99 inventory days. They are one-at-a-time tests and cannot be combined without rerunning the model.

Management's claims produce cloud ERP NPV of RM 962,509, a RM 1.057m swing from the credited base case. That reference case assumes stronger KPI gains, full monetisation of time, no cost overrun and no licence inflation. Its attractive result shows how optimistic assumptions can dominate the answer; it is not evidence that the consultant's base case has understated an achieved benefit.

A gated route preserves choice. Authorise the vendor RFP in months one to three while the data and control work proceeds. At month four, assess a fixed-price implementation target of RM 700,000, annual licence and support no greater than RM 180,000, a 15% holdback and a requested annual price-uplift cap of 3%. The procurement decision needs firm terms, rather than the estimates used in the initial appraisal.

At the model's target-terms test, the phased route improves to RM 142,659 NPV, 17.3% IRR and 28.3-month payback. It now passes the standalone tests, but still creates RM 217,311 less NPV than process improvement alone. Reaching 66 inventory days as well increases phased NPV to RM 328,427, leaving an incremental shortfall of RM 31,543. The report therefore adds a stricter incremental test: any residual gap needs owner-signed scalability or customer-mandated control benefits.

The scenario table specifies the RM 700,000/RM 180,000 test but does not separately document a 3% licence-inflation override. Its positive NPV should not be described as the result of that inflation cap. A firm quote must be entered into the live model with its actual escalation terms.

The readiness gate also requires fewer than 2% duplicate item codes, complete cleansing of A/B-class items, at least 90% accuracy in cycle-counted zones, emergency orders trending toward 9% or less, a completed time study and funded delivery capacity. The refreshed base case must meet the report's 15% gate IRR and the conservative NPV must be no worse than negative RM 0.5m. If the evidence fails, Meridian keeps the foundations benefits and postpones the software commitment.

Mobilisation establishes the steering committee, baseline pack, benefit owners, time study and vendor RFP. Foundations then address item, supplier and customer records; inventory policy and counts; procurement controls; weekly sales-and-operations planning; reporting rationalisation; and a cybersecurity baseline. The ERP design and configuration phase is conditional on Gate 1.

Migration covers governed masters, stock by location, open purchase and sales orders, work in progress, ledger balances, open payables and receivables, and three years of transactions, with older history archived read-only. Three trial loads precede cutover. The final reconciliation must reach at least 99.5%, the ledger must tie, and no severity-one or severity-two defects may remain.

Cutover begins with finance, purchasing and sales orders in month ten, followed by Plant B inventory and planning in month eleven and Plant A in month twelve. A four-week parallel run, physical stock baselines, shift super-users, weekend loading and a tested rollback plan protect continuity. A delivery-performance drop greater than three percentage points at Plant A triggers a pause.

Analytics remains an option after the core is stable. The source labels this month fifteen or later but also requires twelve months of clean data. Since the ERP finishes around month twelve, the earliest review must depend on when qualifying data begins, rather than treating month fifteen as a guaranteed analytics release date.

The synthetic survey has 120 responses from 185 employees. Among respondents, 78% say systems create unnecessary work, 72% expect integration to help, 61% worry about learning new software, 44% fear job losses and only 39% feel confident in their digital skills. Because the sampling method is unclear, the findings are directional. Assuming every non-respondent disagreed, the first result still represents about 51% of all staff; that arithmetic bound is not a confidence interval.

The change plan commits to retaining staff, deploying sixteen super-users, backfilling key users and capping training at four hours per employee per week. Role-based practice covers scanner use, reconciliation and realistic planning exceptions. The training workbook totals 1,904 planned hours, rather than the approximately 1,700 quoted in the report; the workbook total is used here, with the HR roster still needed to resolve overlapping role estimates.

The risk register contains twenty open risks: eight high, eleven medium and one low before mitigation. Average likelihood-times-impact score falls from 13.05 to a planned residual 6.55, but these ordinal scores are not probabilities and the reductions are not observed outcomes. Dirty master data, unrealised benefits, cost overruns, cutover disruption, resistance and credential compromise each have an owner and an early-warning trigger.

The CFO owns monthly benefit reporting and reruns the investment model quarterly for the first two years. Each line needs an evidence source: inventory valuation and COGS for cash release, invoice cause codes for emergency premiums, payroll and schedules for overtime, and time samples for released capacity. Any benefit below 70% of plan for two consecutive months triggers a recovery plan. Quality outcomes remain separately owned by production and QA.