Every quality manager has sat through the same demonstration. A polished consultant clicks through a dashboard, a nonconformance gets raised, routed, dispositioned and closed in four minutes, and the audience leaves convinced their document control nightmares are over. I have sat through dozens of these demos, and the one thing the vendor will never show you is your processes. They show you theirs.
The demonstration works because the processes behind it were designed, tested and refined before a single screen was built. When that software lands in your plant, it will faithfully execute whatever you feed it: the approval chain that takes eleven days because nobody defined who owns the decision, the CAPA register where actions pile up unassigned, the supplier deviation process three departments run three different ways. Automating a bad process makes it faster, not better.
The selection project is therefore not a software project. It is a process definition project with a software deliverable at the end. Teams that grasp this early spend their budget well. Teams that treat it as IT procurement buy themselves a very expensive filing cabinet with workflow buttons.
Start With Requirements, Not Feature Lists
Document what your quality system must do today and what it must do differently tomorrow, before any vendor is contacted. The specification must be written by the people who run the processes, not paraphrased by the IT department. A quality engineer who closes forty nonconformances a month knows precisely where the current system bleeds time. Her list of frustrations is worth more than any analyst report.
Structure the requirements in tiers. Critical requirements are the ones without which you cannot run the business: document control with revision history, nonconformance workflow with electronic sign-off, corrective action tracking with due-date escalation, audit management, supplier quality records. Important requirements include analytics, ERP integration for part master data and multi-site roll-out capability. Nice-to-haves are everything the salesperson leads with.
Then test every requirement against a real scenario, not a hypothetical one. Take an actual nonconformance from last quarter — a real part number, a real defect, a real customer — and walk it through the system on a trial licence. Watch where the workflow forces a workaround. If your containment process requires quarantine of stock at two physical locations with separate labels, does the system model that, or does it assume one disposition per record? These details decide success, and none appear on a comparison matrix downloaded from a vendor website.

Process Fit: Where Most Selections Fail
Process fit means the software reflects how decisions actually flow through your organisation: the approval authorities, the escalation paths, the handoffs between quality, engineering, production and the customer. Map your current processes honestly before you look at a single product. Simple swimlane diagrams work: who raises, who reviews, who approves, who verifies closure, and where the record lives at each step. Draw them as they are, not as the procedure manual pretends they are.
The gap between the drawn reality and the procedure manual is your first workload. In my experience, a typical mid-size manufacturer finds that a third of its documented workflows do not match practice — usually because practice evolved and the document didn't, occasionally because someone found a shortcut that quietly removed a verification step. Both cases must be resolved before configuration, because the software will hard-code whichever version you feed it, and changing a configured workflow after go-live costs far more than fixing it beforehand.
Evaluate vendors against the drawn processes, in a workshop, with your own data. Give each shortlisted supplier the swimlanes and two weeks to configure a sandbox showing your flow. A vendor who resists this, or who answers every question with "that's configurable in phase two", is telling you something. Configuration effort is real effort, paid in consultant fees or your own engineers' evenings, and it is the most commonly underestimated line in the project budget.
A selection sequence that protects you from the demo
- 01Write requirementsAuthored by process owners, tiered critical, important, nice-to-have
- 02Map current processesSwimlanes drawn as practised, not as documented
- 03Close the gapsReconcile procedure manual with reality before configuration
- 04Sandbox evaluationVendors configure your flow with your data, on a trial licence
- 05Baseline metricsRecord current cycle times before the contract is signed
- 06Govern after go-liveNamed owner, quarterly reviews, workaround elimination
Never Digitise a Broken Process
Here is the rule I enforce without exception: no process goes into the new system until it has been fixed on paper. Digitising a broken process does not repair it; it preserves it. The waste gets encoded into the workflow engine, and six months later nobody remembers that the extra approval step was a workaround for a supervisor who left years ago. The system now enforces the waste and generates an audit trail proving you do it consistently.
Look for the classic failure modes before configuration begins. Loops where a document bounces between two reviewers indefinitely because rejection criteria are undefined. Parallel processes where the same deviation is tracked in the quality system, a spreadsheet and a shared mailbox, none reconciled. Approval steps with no authority attached — a signature for the sake of a signature. Manual re-entry of data that already exists in another system, which is where transcription errors breed. Each must be redesigned, not replicated.
Ask a blunt question of every step in every workflow: what would break if we deleted this? If the honest answer is nothing, delete it now, while it is cheap. I have watched teams cut entire approval chains during a selection project simply because mapping the process exposed steps nobody could justify. That is the hidden dividend of a well-run selection: you arrive at go-live with leaner processes than you started with, rather than an electronic museum of historical accidents.
What to Measure Before and After
You cannot prove the software helped unless you measured the pain beforehand. Baseline the current state with hard operational metrics: average days from nonconformance detection to closure, average days for a document revision to reach release, count of open corrective actions past due, time to assemble quality records for a customer audit, and the number of duplicate or orphaned records in existing registers. Record these numbers before the vendor contract is signed and store them where nobody can quietly revise them.
After go-live, measure the same metrics plus a few system-specific ones. Workflow cycle time per step will show exactly where electronic approvals stall — and it will usually be the same step where paper approvals stalled, which tells you the bottleneck was never the medium. Data completeness matters too: what proportion of nonconformance records carry a defect code, quantity affected and disposition rationale in structured fields rather than free text? Free text cannot be trended; structured fields can.
Review the measures at ninety days and again at a year. A properly implemented system shows falling closure times and rising first-pass yield on data entry. If closure times have not improved, resist the temptation to blame the software and look at the process again. Somewhere, a human decision point is queueing work, and no amount of licensing will fix a decision that nobody is empowered to make.
Digitising a broken process preserves the waste, enforces it, and generates an audit trail proving you do it consistently.
Integration, Data Migration and the Edge Cases
The quality system does not live alone. It needs part numbers and supplier records from the ERP, inspection results from measurement equipment or the MES, and possibly customer complaint interfaces for major accounts. Establish during selection exactly which integrations exist as standard, which are chargeable custom work, and which are fantasies. Ask to speak to reference customers running the same integration — a vendor's confidence is not evidence.
Data migration deserves more respect than it usually gets. Migrating five years of nonconformance history sounds valuable until you examine the source data: duplicate part numbers, missing defect codes, free-text descriptions in three languages, records closed with the comment "done". Decide deliberately what to migrate, what to archive read-only, and what to leave behind. Migrating garbage into a modern system gives you searchable garbage, which is arguably worse than inaccessible garbage because people now trust it.
Probe the edge cases during evaluation, not after contract signature. What happens when a supplier changes their company name and you need historical records under both identities? How does the system handle a nonconformance spanning two plants with separate quality managers? Can a corrective action stay open across a supplier change on the affected part? Vendors demo the happy path; your business runs on the exceptions, so spend your evaluation time there.
Where evaluation time is spent versus where it should go
What demos cover
- Single nonconformance, single site, clean data
- Standard ERP integration, out of the box
- One disposition per record
- Approval chains that resolve in minutes
What your business runs on
- Multi-site records with split authority
- Custom integrations, priced as change requests
- Quarantine at two locations, separate labels
- Name changes, open CAPAs, edge cases
Governance, Ownership and Life After Go-Live
Someone must own the system after the go-live dust settles, and that someone needs both process authority and system knowledge. A quality person trained as a system administrator beats an IT person trained in quality, in my experience — the configuration decisions that matter are process decisions. Name this owner before selection, involve them throughout, and give them protected time after go-live, because the first six months generate a steady stream of improvement requests that must be triaged, not ignored.
Watch for process drift. Every user who discovers a workaround — tracking something in a personal spreadsheet, emailing records outside the system — is telling you the configured workflow does not fit reality somewhere. Treat each workaround as a defect in the configuration, investigate the root cause, and either fix the workflow or fix the training. Left alone, the workarounds multiply until you are running two quality systems again: one licensed, one in Excel.
Finally, insist on a defined cadence of system review. Quarterly, check workflow performance against the baseline metrics. Annually, review every configured approval chain to confirm the named authorities still exist and are still necessary. Apply a standing rule that no new workflow step is added without answering what would break without it. Software bought without governance is chaos with a subscription fee; software governed with the same discipline you apply to any other process becomes what it should have been all along: a tool that makes good processes visible, enforceable and fast.
