14.6 Systems Requirements, Master Data, and Emerging Technologies
Key Takeaways
- Write requirements as outcomes in user-story form — "As a [role], I need [capability] so that [outcome]" — because automating a broken process only produces the broken result faster
- Current-state and ideal-state analysis uses process mapping, flowcharts, benchmarking, user stories, and user interfaces, then identifies gaps, system limitations, costs, process knowledge needed, resources required, and applicable organizational or regulatory policies
- Master-data accuracy compounds: at 95% bill of material, item-master, and inventory record accuracy, one item's plan is right on all three only 0.95^3 = 86% of the time, and a 10-component explosion is fully clean about 21% of the time
- A planning system pays off only when lot sizes, time fences, and safety stock rules are set deliberately — leaving the vendor's default configuration in place is still a policy decision, just one nobody made
- Additive manufacturing changes make-or-buy and spares stocking, IoT converts preventive maintenance into predictive maintenance, RPA removes clerical transactions but not judgment, blockchain proves a record was not altered rather than that it was true when entered, and a digital twin simulates a plan before you commit it
Section 14.3 covered the systems landscape — enterprise resources planning (ERP) as the transactional backbone, advanced planning and scheduling (APS) above it, and manufacturing execution system (MES) and warehouse management system (WMS) interfaces. ECM Section IX.C asks something else: can you specify what a system must do, keep its data honest, and judge a technology against a business need?
Defining Current State and Ideal State
You cannot write a requirement until you can describe both states precisely. The ECM names five tools:
- Process mapping — the end-to-end flow across functions, with handoffs, decision points, wait time, and rework loops. Map what people do, not what the procedure says.
- Flowcharts — the symbol-level view of one process (also a basic seven quality tool, 14.4), exposing branches a narrative hides.
- Benchmarking — internal, competitive, or functional comparison that supplies the ideal state. Without it, "ideal" becomes "current state plus a wish."
- User stories — requirements written from the role that lives with them: "As a [role], I need [capability] so that [outcome]."
- User interfaces — wireframes that test whether the data a planner needs is visible at the moment of decision, before code is written.
The user story is phrased as an outcome, not a screen feature, for one reason: automating a broken process only produces the broken result faster. "Add a button that mass-firms planned orders" hard-codes a process nobody validated. "As a master scheduler, I need to firm planned orders inside the demand time fence so that MRP stops rescheduling committed customer orders" states the outcome and leaves the design open.
Worked example — exception messages nobody triages
| Current state | Ideal state | Gap | Requirement |
|---|---|---|---|
| ~1,900 exception messages per regeneration; planners work the top of the list | ~40 ranked items that change a customer commitment | No prioritization, ownership routing, or noise suppression | As a planner, I need messages ranked by demand value at risk and filtered to my planner code so that I act on what changes a promise date |
| Planner code blank on 22% of active items | Every planned item has one owner | No mandatory-field validation at item creation | As a materials manager, I need item creation blocked without a planner code so that no item enters MRP unowned |
| Reschedule-in messages fire inside the demand time fence | In-fence messages go to exception review | Time-fence policy disconnected from message generation | As a master scheduler, I need in-fence messages routed to a review queue so that the schedule stops churning |
Beyond gaps, IX.C.1 requires you to identify system limitations (what the package cannot do without customization), costs (license, integration, conversion, training, ongoing support), process knowledge needed, resources required (planner subject-matter-expert time is the binding constraint, not IT headcount), and applicable organizational or regulatory policies (segregation of duties, export control, electronic records and signatures, data privacy). A requirement that violates a policy is not a requirement.
Implementing and Maintaining Technology Systems
A phased cutover goes live by site, module, or family — contained risk, but temporary interfaces and two versions of the truth for months. A big-bang cutover switches everything on one date — less interface work, but every defect surfaces the same morning.
- Conference room pilot (CRP) — run a full business cycle in a test environment with your own items, bills, and routings. It proves the configuration handles your business, not the demo company's.
- Data conversion — decide what converts (item master, bills, routings, open orders, on-hand) and cleanse before loading. Converting dirty data permanently destroys user trust.
- User acceptance testing (UAT) — users run scripted end-to-end scenarios and sign off. Not IT unit testing; the test is whether a planner can do the job.
- Training — role-based, on converted data, close to go-live.
- Hypercare — elevated support right after go-live, stepping down once the business is stable.
Then the point exam items key on: the system pays off only when planning policy is set deliberately. Lot-size rules, order minimums and multiples, safety stock rules, time-fence positions, and firm-planned-order discipline are decisions. Inheriting them from the vendor's default configuration is still a decision — just one nobody made.
Managing the Integrity of Master Data
Every MRP result is an arithmetic consequence of four record sets. Master data is not an IT asset; it is planning policy written down.
| Record set | Fields that drive the result | Failure symptom |
|---|---|---|
| Item master | Planning lead time, lot-size rule (lot-for-lot, fixed order quantity, period order quantity), order minimum/multiple/maximum, safety stock, scrap or yield factor, unit of measure, planner code, ABC code | Lead time padded in an old expedite, never reset — orders release early and inventory climbs |
| Bill of material | Components, quantity per, effectivity dates, scrap factor, phantom flags | Quantity-per wrong or a change released without effectivity dating — one component short at every assembly |
| Routing and work center | Operation sequence, setup and run standards, queue and move time, rated capacity, efficiency and utilization | Stale time standards — CRP load looks comfortable while the floor misses every date |
| Inventory records | On-hand by location, allocations, in-transit, lot and serial control | Inventory record accuracy (IRA) decay — MRP nets against stock that does not exist |
The arithmetic of compounding accuracy
Take bill of material accuracy at 95%, item-master accuracy at 95%, and inventory record accuracy at 95%. Independent, the probability that one item's plan is right on all three is:
0.95 x 0.95 x 0.95 = 0.857375 → about 86%
So 14% of single-item plans are wrong before demand moves. Explode a parent with 10 components, each needing all three right:
0.857375^10 = about 0.215, or roughly 21%
Barely one explosion in five is clean; the rest carry a bad element that surfaces as a shortage, an expedite, or dead stock. Move all three record types to 99%: 0.99^3 = 0.9703 per component, and 0.9703^10 = about 74%. Four points per record type more than triples the odds of a clean explosion — which is why targets sit in the high 90s and why "95% is good enough" is a trap answer.
Governance that holds accuracy in place
- Data ownership — a named owner per record type: engineering owns bill structure, manufacturing engineering owns routings and standards, planning owns item-master planning parameters, materials owns inventory records.
- Change control — a change to lead time, lot size, or safety stock carries a reason code, an approver, and an effective date. Silent edits during an expedite are how master data rots.
- Periodic parameter review — planning lead time against actual receipt lead time, lot sizes against current demand, safety stock against achieved service level.
- Audit cadence — cycle counting sustains IRA at ABC-based frequency with discrepancies root-caused; bill audits triggered by unexplained usage variance, routing audits by labor variance.
Data cleanup is not a project. Accuracy decays continuously, so only a running process holds it.
Emerging Technologies and What They Change for a Planner
| Technology | What it does | Planning decision it changes | Realistic limitation |
|---|---|---|---|
| Artificial intelligence (AI) | Finds pattern in large data sets and scores outcomes | Forecast model selection, demand sensing, exception triage, inventory segmentation | Needs clean history; handles structural breaks poorly, and bad master data yields confident wrong answers |
| Internet of things (IoT) | Sensors stream condition and location data from assets and product | Turns calendar-based preventive maintenance into predictive maintenance from actual condition, so downtime enters the capacity plan on purpose | Data volume and a wider security surface; worthless unless the work-order process consumes the signal |
| 3D printing / additive manufacturing | Builds a part from a digital file with no hard tooling | Collapses tooling lead time and enables on-demand spares — changes make-or-buy, spares stocking (hold the file, not the part), NPI bridge production | Unit cost and cycle time lose at volume; material qualification and certification can outlast the tooling |
| Robotic process automation (RPA) | Bots execute rule-based transactions across systems | Removes clerical load — order release, confirmation rekeying, ASN matching — freeing hours for judgment | Automates rules, not judgment; brittle to screen changes; locks in the current process, so fix it first |
| Blockchain | Shared distributed ledger holding an immutable multi-party record | Provenance, chain of custody, country of origin, and certification claims without a trusted intermediary | Proves the record was not altered, not that it was true when entered; needs multi-party adoption |
| Digital twin | Live virtual model of a product, process, or network fed by real data | Simulate a plan before committing it — line rebalance, layout change, network change, capacity what-ifs | Only as good as model fidelity and the master data behind it |
Evaluating Technology the Way the Exam Does
CPIM tests whether you can evaluate a technology against a business need and a data reality, not whether you can name the newest tool. Ask which decision improves, what data must exist and be accurate, whether the process is worth automating or should be fixed first, what integration and maintenance cost against the benefit, and which policy constrains the answer. The most impressive technology in the answer set is rarely the key — the right answer matches the stated need and survives the data-reality check, which loops back to master-data integrity.
A plant measures bill of material accuracy at 98%, item-master planning-parameter accuracy at 97%, and inventory record accuracy at 96%. Treating the three as independent, what is the approximate probability that the plan for a single item is correct on all three record sets?
MRP generates roughly 1,900 exception messages each regeneration and planners work the top of the list until time runs out. Which statement is a properly formed requirement for the replacement system?
Six weeks after go-live, planners report that the new system "creates far more orders than the old one and the schedule never settles." Review shows every item was loaded lot-for-lot with no order minimums, zero safety stock, and no demand time fence, because those were the vendor's default values. What is the correct diagnosis?
A field-service business stocks 4,000 slow-moving legacy spare parts; for many of them the original tooling has been scrapped and the supplier quotes 26 weeks. Which emerging technology most directly changes the stocking decision, and what is its realistic limitation?