How to create a decision matrix for selecting a laser machine without subjective scores ART-050

Why an ordinary rating misleads

The table “supplier A — 9, supplier B — 8” looks orderly but does not explain where the figures came from. A productivity score of 9 may rely on different materials, modes, parts and scope boundaries. A service score of 8 may mean a promise in a presentation rather than an agreed support channel. Adding such scores creates precision that the data do not contain.

Instead, every criterion needs four fields: requirement, evidence, confidence status and consequence. The requirement describes your production. Evidence may be a specification, official documentation, a test protocol for an agreed part, a site plan or written confirmation of supply boundaries. Status shows whether information is verified or awaiting confirmation. Consequence answers: what happens if the assumption proves false?

| Criterion | Requirement wording | Acceptable evidence | Cannot be treated as evidence | |---|---|---|---| | Product mix | List of materials, thicknesses, geometries and batches | Order data, drawings, agreed test parts | Generalization “we cut everything” | | Productivity | Required flow including preparation | Actual-cycle analysis and scope boundaries | Nameplate cutting speed alone | | Site | Space, power, extraction, logistics | Plan and manufacturer requirements for the specific system | Assumption that “there is enough room” | | Support | Channel, roles, documents, spare parts | Contract terms and service procedure | A manager’s verbal promise | | Economics | Full cost and risk scenarios | Separate list of inclusions, exclusions and assumptions | Equipment price alone |

Build the matrix from parts, not a catalogue

Take a representative work set: regular parts, quality-critical parts, loading-problem parts and prospective orders. You need not include everything, but cannot base a decision on one attractive sample. For every group record material, surface condition, thickness range, geometric elements, batch size, next-operation requirements and repetition frequency. If the data are absent, the honest status is “not defined,” not an excuse to invent a weight.

Then formulate criteria as checkable questions. Instead of “high quality,” ask “what result must be confirmed on an agreed part and by which acceptance method?” Instead of “sufficient automation,” ask “which manual operation is a bottleneck, who performs it, and what does a particular module change?” Instead of “service is good,” ask “what is the claim procedure and who diagnoses?” This does not make an answer guaranteed; it makes the unknown visible.

Weights must not mask risk

A weight is useful only as a description of priority. It cannot be selected so a favored quotation wins. For each criterion, write the reason for its weight: material with a high order share, limited space, or critical transition time between operations. Then check two things. First, do criteria duplicate each other? “Speed,” “productivity” and “throughput” often count the same advantage three times. Second, is there a critical stop factor that scores cannot compensate? Lack of safe infrastructure, incompatibility with required product mix or an unagreed specification is not a low score; it is a reason to stop the decision until clarified.

In practice, three statuses are convenient: confirmed, needs confirmation, does not meet. A number is permitted only inside confirmed alternatives and only when assessment rules are written before comparison. If data are not comparable, do not average them. Add a supplier question or plan an equal test.

Separate basic suitability from advantages

The matrix has two layers. The first is mandatory conditions: conformity with agreed product mix, ability to locate the system lawfully and safely, defined supply boundaries, available documentation and agreed acceptance. An alternative that fails this layer must not advance to the final rating.

The second layer is trade-offs: loading organization, flexibility, programming convenience, scope for future expansion and cost structure. Assessment is possible here, but with a source reference. A shuttle table, for example, may be a priority when loading downtime is confirmed by observation. It is not automatically better when production runs short irregular cycles and the constraint is file preparation or sorting.

A repeatable algorithm

1. Record the decision goal, planning horizon and people responsible for the data. 2. Make a short map of product mix and actual work flow. 3. Separate mandatory conditions from trade-off criteria. 4. Define one acceptable proof for each criterion and the status of its availability. 5. Compare quotations against an identical specification, marking missing rows as unconfirmed. 6. Test scenarios: what changes if the most important part group or volume does not materialize? 7. Record the decision, unresolved risks, owner of every check and review date.

The quality-management approach expressed in ISO 9001 supports process thinking and evidence logic. It does not prescribe which laser to buy and does not replace a technical project. Similarly, NIST Baldrige materials are useful as a reminder to manage decisions from data, not as an industrial-equipment specification.

How to handle price

The lowest price is not a criterion without explaining the scope. Break every quotation into machine, source, head, automation, software, installation, training, infrastructure, service terms and exclusions. Do not invent future costs or a payback period. Create an assumption register instead: what is included, what the customer must provide, what depends on actual utilization, and where a separate quotation is required.

If the financial decision needs a scenario, use several rather than one forecast: current flow, cautious growth and product-mix change. Keep rules equal in each scenario. The difference among scenarios is not manipulation; it reveals which assumptions most affect the conclusion.

Typical mistakes

Scoring an entire quotation. Its individual parts can have different confirmation levels.

Adding desired functions without a production problem. They turn the document into a wish list.

Compensating a critical risk with high scores. Stop factors must be resolved before selection.

Mixing requirements and ways to meet them. Describe the need first, not the name of a catalogue option.

Not versioning the matrix. When product mix, specification or site plan changes, the old decision can lose its basis.

What cannot be determined without data

Without your product mix, work plan, available infrastructure, full specifications and acceptance rules, it is impossible honestly to name the matrix winner or guarantee productivity. The matrix does not confirm compatibility, safety or economic effect; those must be verified through documentation for the specific system, tests and contract conditions.

Checklist

• [ ] The decision goal is written in one checkable sentence.
• [ ] Criteria arise from product mix and flow, not the catalogue.
• [ ] Every criterion has evidence, status and an owner.
• [ ] Stop factors are separated from assessable trade-offs.
• [ ] All quotations are brought to the same supply structure.
• [ ] Assumptions and unknowns have not been turned into scores.
• [ ] Technical, production and contractual sides will review the decision.

Example structure of a working sheet

One matrix row can read: “process the agreed group of parts.” In the requirement column, list concrete parts or their characteristics. In the evidence column, link drawings, production statistics, the specification and agreed-test protocol. In the risk column, state what happens if data are not confirmed: additional tooling, transfer of some work, schedule change or a different configuration. In the decision column, use a status rather than a score: confirmed, awaiting confirmation, does not meet. Only then may alternatives that pass mandatory conditions use the same priority scale.

Keep a separate “questions about the quotation” sheet. Put unclear module names, absent supply boundaries, unclear training scope, infrastructure requirements and productivity claims without a scenario there. When an answer arrives, do not merely tick a box: attach the document or demonstration date. If there is no answer, status remains unconfirmed. This disciplines the process and explains why a decision is not yet ready.

Sensitivity check of the decision

After first completion, ask the team to change only one premise: for example, the share of thicker material, shift volume, or need to operate in limited space. If the winner changes after a small change, that is not a matrix error. It signals that the decision depends on an assumption that must be confirmed or addressed in the contract. If the conclusion does not change, confidence increases, but no productivity guarantee is created.

Do not use sensitivity analysis to fit a result. Record in advance which variables are tested, why they matter and what every outcome means. For procurement, this is often more useful than another decimal place in a rating: the team sees where a reserve is needed and where an assumption is sufficient.

How to conduct a joint review

A matrix works better when it is not filled in by one person. Procurement is responsible for comparable commercial terms, the technologist for the connection to parts, production for actual flow, technical service for site and operation, and finance for cost assumptions. They need not agree everything immediately. The value of the meeting is to name disagreements and assign a verification owner, not agree an “average score.”

During review, forbid three unsupported phrases: “everyone does it,” “it is obvious,” and “the supplier said so.” Every statement must become either a document or an item for clarification. If data are confidential, record their type, owner and verification date without disclosing commercial details in the shared table.

The final protocol must show not only the selected alternative, but also rejected alternatives and the reason. This helps if volume, part mix or available budget changes months later: the team can return to assumptions instead of restarting selection. A review does not mean the earlier decision was wrong; it means it was tied to facts that may change.

Limits of a mathematical model

A weighted sum can be useful only as an auxiliary summary. It cannot turn unknown data into facts, determine system safety or replace professional review of a technical specification. If one criterion is genuinely critical, make it an admission condition rather than allow other scores to override it. A missing or unverified site requirement is not compensated by an attractive price or extra option.

Do not use excessively precise coefficients. A weight of 0.37 rather than 0.4 does not make a choice scientific when input data are approximate. Describe a priority range, the data source and the situation in which priority can change. The matrix must end with next actions: which documents to obtain, what test to agree, what to include in the specification, and which risks to accept by a separate decision. Then it becomes a procurement plan, not a presentation.

Before approval, check whether every criterion actually differentiates alternatives. If both quotations equally meet a basic requirement, do not spend time on artificial ranking. Focus on what changes the decision: different supply scope, actual site readiness, ability to verify critical product mix, service procedure or dependence on future expansion.

The matrix does not cancel a manager’s responsibility for final choice. It shows where the choice rests on confirmed facts and where it rests on accepted risk. That is what permits open discussion of risk before the contract.

The practical value of a matrix is particularly visible for one critical criterion. Suppose an enterprise needs to process a particular group of parts but the commercial quotation has no clear material boundary, test conditions or configuration description. Do not put a low score “by feel” in the row. State the requirement, identify the missing proof and set the status to “unconfirmed.” Then define an action: request official documentation, agree a test part, or exclude the alternative until an answer is received. The decision then does not hide uncertainty behind a figure.

Another useful example is the service criterion. “Support is available” is not a sufficient requirement. Break it down into claim channel, documentation availability, fault-registration procedure, remote-help limits, parts supply and communication owner. Some answers belong to the contract, some to technical documentation, and some may be an internal customer decision. Keep them as separate rows so an advantage in one point does not conceal a gap in another.

Before final decision, conduct a short “what changes the conclusion” review. For every unknown, state which verification outcome changes the selected alternative and which merely clarifies terms. If an answer cannot change selection, it should still be obtained for the contract and safe operation, but its weight need not be inflated. If it can change selection, it is a stop factor or mandatory prepayment action, not a note at the end of a presentation.

To avoid false objectivity, keep source data separately from the assessment. They include drawings, material list, production statistics, official documents, clarification correspondence and test protocols. The matrix itself need only link to a record or give its identifier. If a source file changes, state the new version and date. Another person can then reproduce not only the score but why it was assigned.

For complex criteria, use a short “pass condition.” Instead of “high accuracy,” state which geometry, material, document or test must be checked. Do not invent numerical thresholds unless a drawing, contract, standard or manufacturer document defines them. When a threshold is critical, agree it before comparison; otherwise it can be unconsciously adjusted after quotations arrive.

The last matrix row should answer “what do we do next?” It can be a request for a missing specification, agreement of a control part, site verification or legal review of terms. If the next step depends on another team, state that directly. A matrix that ends only with “winner” does not show whether the enterprise is ready for an advance payment, test or cell preparation.

Need service advice?