Multi-Criteria Decision Making

When dollars are not the only currency

A city operations director chose a fleet vendor by lowest five-year cost. Finance applauded. Six months later, mechanics logged twice the expected downtime, citizen complaints rose, and spare parts sat on ocean freight during a storm season. The spreadsheet was internally consistent. The objective function was wrong for the decision. Cost minimization ignored uptime, emissions, local jobs, and contract flexibility that the council had named in public hearings.

Real managerial choices rarely reduce to a single EMV (expected monetary value, probability-weighted dollars from Lesson 1) line. Multi-criteria decision making (MCDM, structured methods for comparing alternatives on several goals) makes trade-offs explicit when you must weigh profit against risk, speed against quality, or growth against reputation. MCDM does not eliminate judgment. It forces judgment into documented weights and scores so stakeholders can debate what matters, not only what happened in one cell.

The city fleet failure in the opening hook is not rare. Infrastructure, healthcare, and technology purchases all combine cost, reliability, risk, and stakeholder goals. A single NPV line cannot encode council promises on local jobs or emissions without monetization assumptions someone will challenge. MCDM makes those dimensions explicit, debatable, and auditable. It also exposes when a vendor wins on cost while losing on uptime badly enough to negate savings.

From Lesson 2, sensitivity shows which inputs move a single metric. MCDM extends that mindset to vectors of outcomes. From Lesson 3, value of information asks what learning is worth before a fork. MCDM often follows EMV analysis: finance computes NPV (net present value, discounted cash flows minus investment), operations scores service levels, legal scores compliance risk, and leadership integrates. This lesson teaches goal hierarchy, weighted scoring, and cautions about rank reversal and gaming that appear in every scorecard process.

Capital committees routinely face choices where the highest NPV project damages brand, or the cheapest vendor fails security review. EMV and NPV remain essential when cash outcomes dominate and probabilities are credible. MCDM enters when multiple legitimate goals cannot be collapsed into one dollar line without distorting the decision. The art is sequencing: run finance first, apply vetoes, then score strategic factors among survivors. Skipping finance and scoring everything on "strategic fit" invites projects that destroy value with a smile.

Structuring criteria and avoiding laundry lists

Start by listing alternatives (vendors, sites, product variants, hiring candidates) and criteria (measurable dimensions of value). Criteria should be mutually exclusive in definition (no double counting revenue growth twice under different names) and collectively exhaustive enough to capture why you might reject an alternative.

A practical hierarchy:

1. Strategic goals (board level): enter EU profitably, reduce carbon 20%, maintain AA bond profile.
2. Decision criteria (evaluation level): five-year NPV, implementation time, regulatory risk score, customer NPS (Net Promoter Score, survey-based loyalty index).
3. Metrics (measurement level): NPV in dollars, months to launch, red/yellow/green legal rating, NPS from pilot.

Without hierarchy, workshops produce laundry lists of 30 criteria where everything is "important." Prioritize. Ask: "If this criterion improved 10% while others fixed, would we change the choice?" If no, drop or merge.

Use must-have thresholds (hygiene factors) separate from scored trade-offs. A vendor failing SOC 2 (System and Organization Controls 2, a security audit standard*) is disqualified; do not let high cost score compensate for missing compliance.

Document stakeholders and who owns each metric. Finance owns NPV; operations owns uptime; HR owns retention impact. Single-owner metrics reduce political averaging where everyone inflates soft scores.

Criteria definition workshops work best with negative examples. Ask the room to name a vendor or site they would reject despite "good average scores." Reverse-engineer which dimension drove the rejection. That dimension becomes a criterion or veto. Positive-only brainstorming produces vague labels like "culture fit" that resist measurement.

Separate leading from lagging metrics where possible. Uptime next quarter is lagging; spare-parts inventory availability is leading. Weighting both prevents rewarding vendors who look good until failure occurs.

Weighted scoring and simple additive models

The most common MBA approach is weighted sum scoring:

1. Score each alternative j on criterion i (often 1–5 or 0–100).
2. Assign weight w_i with Σw_i = 1.
3. Total score = Σ_i w_i × score_ij.
4. Rank alternatives by total score.

Example weights for a software platform buy: Total cost of ownership (TCO) 0.35, Time to deploy 0.25, Security/compliance 0.20, Vendor viability 0.10, User satisfaction from pilot 0.10.

Scores should use anchors so "4" means the same thing across reviewers. Example anchor for deploy time: 5 = ≤3 months, 3 = 6 months, 1 = ≥12 months.

Normalization matters when criteria use different units. Convert raw values to comparable scales before weighting:

• Linear scale: best raw value maps to 100, worst to 0.
• Step functions: regulatory green = 100, yellow = 50, red = 0.

After computing weighted totals, run sensitivity on weights (Lesson 2 spirit): if Vendor B wins at w_NPV=0.40 but loses at w_NPV=0.30, disclose the crossover. Close weight regions deserve executive choice, not false precision.

Weighted scoring assumes compensatory trade-offs: a bad score on cost can be offset by a great score on speed. Some decisions are non-compensatory (any red legal flag vetoes). Model vetoes explicitly before scoring.

Total cost of ownership (TCO) should include implementation, training, maintenance, and exit cost, not license price alone. A cheap platform with 18-month deploy scores poorly on time even if license fee wins. Document TCO build-up in an appendix table so operations and finance agree on the cost criterion before scoring.

When criteria conflict (low cost, high service), weighted sums hide who bears the trade-off. Publish a stakeholder map: which function championed which weight. Transparency reduces post-decision resentment when operations lives with a finance-weighted choice.

Analytic Hierarchy Process and rank reversal cautions

The Analytic Hierarchy Process (AHP, pairwise comparison method to derive weights and check consistency) asks decision-makers to compare criteria two at a time: "Is cost twice as important as speed?" Pairwise judgments produce a weight vector and a consistency ratio to flag contradictory comparisons (if cost > speed and speed > risk but risk > cost, fix judgments).

AHP helps groups align language before scoring alternatives. It does not remove politics; it surfaces inconsistency.

Rank reversal occurs when adding a clearly inferior alternative changes the ranking of two good ones. Example: A beats B until mediocre C enters and scoring logic makes B look better relative to C. Teach teams to treat rankings as stable only within a fixed set and to re-run when the choice set changes.

Other pitfalls:

• Gaming: vendors optimize demo features that map to high-weight criteria while neglecting hidden costs.
• Illusory precision: weights to four decimals imply false accuracy.
• Double counting: NPV and "five-year savings" as separate criteria.

When EMV or NPV is available, a strong pattern is lexicographic screening: first eliminate alternatives failing NPV hurdle, then MCDM among survivors on strategic factors.

AHP consistency ratio above conventional thresholds (often 0.10) means pairwise judgments contradict. Facilitator sends the group back to compare cost versus risk again rather than averaging conflicting opinions into a false consensus.

Outranking methods (ELECTRE-style, light touch in MBA) flag alternatives that dominate others on enough criteria without a full weighted sum. Useful when vetoes are soft preferences rather than hard rules. Weighted sum remains default for transparency.

Integrating MCDM with EMV, risk, and governance

Use EMV for uncertainties with credible probabilities (Lesson 1 trees). Use MCDM for qualitative or multi-stakeholder dimensions EMV cannot monetize without heroic assumptions.

A hybrid workflow for capital projects:

1. Finance computes NPV distribution or EMV.
2. Operations scores implementation risk and customer impact.
3. Legal assigns compliance tier.
4. Apply veto rules.
5. Weighted score among feasible options.
6. Sensitivity on top weights and tornado on NPV (Lesson 2).

Present three columns to the board: financial metric, strategic score, risk flags. Recommendation states which constraint binds.

Governance habits:

• Pre-register weights before vendor demos.
• Blind score pilots where possible.
• Audit score changes after final presentations (if weights shift, document why).

For public sector and ESG (environmental, social, and governance, non-financial performance themes), MCDM is often mandatory transparency. Private firms use the same mechanics for brand and talent decisions.

MCDM output is a input to judgment, not a substitute. When scores differ by less than measurement noise, escalate to executive choice with explicit trade-off narrative (preview Lesson 5 communication).

Monetization bridge: when ESG or uptime criteria matter but NPV decides, build shadow valuations. If Nord site wins on carbon but loses on NPV, estimate internal carbon fee that makes Nord tie South. The fee becomes a negotiable policy parameter rather than a vague "values" argument.

Decision records: archive weights, scores, vetoes, and dissenting function views in the same folder as the NPV model. Two years later when uptime fails, the record shows whether operations warned before the vote.

Facilitation mechanics that keep MCDM honest

Run silent scoring first, then reveal ranges. Wide spreads on the same vendor mean anchors were unclear or demos biased perception. Discuss outliers before averaging.

Cap criteria count at seven plus or minus two for executive sessions. More criteria signal an unfocused strategy conversation, not richer analysis.

Use reference alternatives: include one obvious weak option (status quo legacy vendor) to calibrate scores. Teams that score everything 4–5 lose discrimination.

Rehearse rank reversal by adding a dummy alternative in the workshop and showing how rankings shift. Executives who see the phenomenon once are less likely to treat a 0.05-point win as sacred.

Pair MCDM output with scenario stress (Lesson 2). Vendor Beta wins weighted score in Base but fails minimum uptime in Downside supply shock. The board needs both tables.

Score normalization deep dive

When raw criteria differ in units, pick a normalization rule and document it. Example: TCO raw $4.2M (Alpha) to $5.1M (Gamma). Linear scale: best Alpha = 100, worst Gamma = 0, Beta = 33.3. Nonlinear scales (log cost) compress outliers when one vendor is absurdly expensive.

Normalization changes rankings. Run normalization sensitivity alongside weight sensitivity when scores are close.

After normalization, combine with weights. Do not let raw dollars enter weighted sum without scaling.

Public sector transparency

Government procurement often publishes weights and scores after award. Citizens challenge decisions when weights shifted post-scoring. Pre-registration of weights in RFP (request for proposal) documents is legal hygiene as well as analytics hygiene.

Hiring and personnel MCDM cautions

Personnel decisions carry legal risk. Criteria must be job-related, documented, and applied consistently. MCDM structure helps defensibility; it does not remove employment law review. Veto unsafe safety record before scoring other factors.

Weighted product and TOPSIS orientation

Weighted product methods multiply normalized scores raised to weights; zero on any criterion zeros total (non-compensatory flavor). Useful when any zero on security is unacceptable without hard veto.

TOPSIS (technique for order preference by similarity to ideal solution) ranks alternatives by distance to ideal and anti-ideal bundles. Popular in supply chain; MBA teams can use it when weighted sum hides closeness to bad outcomes on one dimension.

For most board memos, weighted sum plus veto rules plus weight sensitivity remains the default because stakeholders can trace arithmetic.

Building a one-page MCDM summary

Columns: Alternative | NPV or TCO raw | Normalized scores | Weights | Total | Veto status.

Footnote: crossover weight for top two alternatives. One page forces discipline; appendix holds backup.

Case: software buy with security veto

TechCo evaluates CRM platforms. Must-have: SOC 2 Type II, GDPR (General Data Protection Regulation, EU privacy rules) data residency in EU. Three vendors pass veto. Weights: TCO 0.30, deploy time 0.25, API extensibility 0.20, admin burden 0.15, vendor financial stability 0.10.

Vendor X wins weighted sum 4.1 versus Y 3.9. Weight sensitivity shows Y wins if deploy weight rises from 0.25 to 0.38 because Y ships in 90 days versus X in 150. COO values speed; CFO values TCO. Executive decision records deploy weight 0.32 compromise and documents Y selection with finance dissent noted.

This pattern repeats: MCDM quantifies disagreement; leadership resolves policy, not spreadsheet.

Scoring workshop agenda (90 minutes)

1. Confirm alternatives and vetoes (15 min).
2. Agree criteria definitions and anchors (25 min).
3. Silent individual scoring (15 min).
4. Reveal ranges, discuss outliers (20 min).
5. Lock weights if not pre-registered (10 min).
6. Compute totals and weight sensitivity crossover (5 min).

Agenda prevents demos before weights freeze.

Worked example extension: DataCo site (full Part D)

DataCo site selection (Lesson body) continues:

Part C: After East veto, weighted overlay with carbon weight 0.35, NPV weight 0.65 (normalized): Nord score 0.65(12)+0.35(carbon index) vs South 0.65(14)+0.35(lower carbon index). Internalize $1.5M/year carbon fee on South → effective NPV South $12.5M versus Nord $12M if fee applied.

Part D: Board picks Nord with published carbon policy fee sensitivity: if fee < $1.2M, South wins. Document crossover for investor ESG questions.

Check: veto East removed infeasible sovereign risk ✓

MetroFleet chooses among Alpha, Beta, Gamma for 40 electric buses.

Part A: Criteria and weights

Weights sum 1.00 ✓

Part B: Scores (1–5, higher better; TCO inverted from cost rank)

Check arithmetic: Gamma 3.85 wins; Alpha 3.75 second.

Part C: Weight sensitivity

If TCO weight falls to 0.30 and uptime rises to 0.35, Beta total = 0.30(3)+0.35(5)+0.20(3)+0.15(4)=3.95, overtaking Gamma. Decision fragile on weighting.

Part D: Managerial read

Finance favors Alpha on cost; operations favors Beta on uptime. Council must choose whether TCO or reliability drives policy. Document crossover at TCO weight ≈ 0.33.

Expanded sensitivity table:

Check: weights sum 1.00 at each row ✓. Council should not hide 0.05-point Gamma win when uptime weight move flips to Beta.

DataCo picks a data center region.

Veto: East disqualified on sovereign risk (policy).

Remaining: Nord vs South. EMV/NPV favors South (+14 vs +12). ESG committee weight on carbon makes Nord competitive in MCDM overlay. Board chooses Nord with carbon fee internalized $1.5M/year, flipping effective NPV ranking. Shows monetization link between MCDM and finance.

Two hiring candidates for plant manager:

1. Compute weighted scores.
2. Who wins?
3. If union relations weight rises to 0.25 and cost reduction falls to 0.25, who wins?

Solution

Alex: 0.35(4)+0.30(5)+0.20(3)+0.15(2)=1.4+1.5+0.6+0.3=3.80

Blair: 0.35(5)+0.30(3)+0.20(4)+0.15(4)=1.75+0.9+0.8+0.6=4.05 → Blair wins

3. Alex: 0.35(4)+0.25(5)+0.20(3)+0.25(2)=1.4+1.25+0.6+0.5=3.75

Blair: 0.35(5)+0.25(3)+0.20(4)+0.25(4)=1.75+0.75+0.8+1.0=4.30 → Blair still wins.

Weights sum 1.00 ✓

Explain why a minimum NPV hurdle should be applied before weighted scoring in capital decisions.

Solution

Weighted scoring is compensatory: a high strategic score could offset negative NPV, destroying shareholder value. A hurdle ensures financial feasibility first; MCDM ranks only projects that meet the financial bar (unless public mission explicitly overrides). This mirrors hybrid governance in the lesson.

MCDM without NPV is strategy theater. MCDM with NPV but without vetoes is compliance theater. Strongest pattern: NPV or EMV hurdle, hard vetoes, weighted score among survivors, weight sensitivity disclosed, executive choice when fragile.

Public sector adds transparency: publish weights in RFP, publish scores after award, defend in audit with archived dissent.

Vendor management: pre-register weights, blind pilots, ban score changes after final demo without written rationale.

Teaching rank reversal in every workshop prevents surprise when a decoy alternative enters.

Fleet example crossover at TCO weight 0.33 is template for policy negotiation: council picks weight, not analyst.

Extended practice: site selection with three criteria (solution prose)

Sites L, M, N; weights NPV 0.50, time 0.30, risk 0.20. Scores L (4,3,5), M (5,2,3), N (3,5,4).

L: 0.50(4)+0.30(3)+0.20(5)=2.0+0.9+1.0=3.90

M: 0.50(5)+0.30(2)+0.20(3)=2.5+0.6+0.6=3.70

N: 0.50(3)+0.30(5)+0.20(4)=1.5+1.5+0.8=3.80

Winner L. If risk weight rises to 0.35 and NPV falls to 0.45: L=0.45(4)+0.30(3)+0.35(5)=1.8+0.9+1.75=4.45; M=0.45(5)+0.30(2)+0.35(3)=2.25+0.6+1.05=3.90; L still wins. N overtakes only if time weight dominates.

Weights sum 1.00 ✓

Deep dive: facilitating weight disagreements

When CFO and COO disagree on weights, do not average silently. Run both weight sets, show rankings, escalate crossover to CEO with one-page trade-off. Averaging hides accountable choice.

Deep dive: ESG monetization workshop

Estimate internal carbon fee making Nord and South NPV tie. Fee becomes policy lever; not precise truth but transparent.

Closing integration with Unit 6

Use EMV or NPV first, MCDM second, sensitivity on weights third, communication of crossover fourth. Skipping steps produces either pure spreadsheet or pure politics.

Multi-criteria decision making is how organizations admit they care about more than one goal without abandoning quantitative discipline. The scorecard is not the decision; it is the structured input to an accountable human choice.

Supplemental narrative: procurement scorecard day

MetroFleet procurement day begins with published weights: TCO 0.40, uptime 0.25, charging fit 0.20, local jobs 0.15. Vendors demo after weights are frozen. Silent scoring on anchored rubrics follows. Gamma wins 3.85 to Beta 3.70, but weight sensitivity shows Beta wins if uptime weight rises to 0.35. Council discusses whether winter reliability policy justifies the weight shift rather than asking analytics to "recalculate until Beta wins."

Public sector observers receive the score sheet and dissent notes. Private firms benefit from the same transparency internally: operations dissent enters the record when uptime weight would flip outcome. Six months later when Beta would have prevented downtime, the record shows the trade-off was understood, not ignored.

MCDM without documented dissent becomes politics with extra arithmetic. MCDM with dissent becomes governance.

Supplemental practice walkthrough (hiring, prose solution)

Plant manager hiring from Lesson practice: Blair wins 4.05 versus Alex 3.80. Explain in prose to the board: Blair leads on safety and union relations, critical in a union plant facing contract renewal; Alex leads cost reduction but union relations score 2 versus Blair 4. With union weight 0.15, Blair margin is comfortable. If union weight were 0.30 and cost 0.20, recompute before vote; sensitivity table in appendix shows Blair still wins at 0.25 union weight. Document check: weights sum 1.00 ✓.

Closing standards

Pre-register weights, anchor scores, run weight sensitivity, publish dissent, and escalate fragile wins to executives with crossover table. MCDM maturity is measured by audit trail quality, not by decimal places in totals.

Scorecards without owners, vetoes, and crossover disclosure look rigorous while smuggling politics back through the side door. Treat every weighted score within 0.10 points as a fragile ranking that requires executive sign-off on weights, not only on vendors.

Key takeaways

• MCDM structures choices when multiple goals cannot collapse to one dollar metric.
• Use goal hierarchy, vetoes, anchored scores, and weights that sum to 1.
• Run weight sensitivity to expose fragile rankings.
• Combine EMV/NPV analysis with MCDM for hybrid capital decisions.
• Pre-register weights and watch for gaming and rank reversal.

1. Score two real alternatives at work on four criteria with explicit weights.
2. Which weight change would flip your ranking?
3. Continue to Lesson 5: Communicating Analytical Recommendations.