Most real choices are not between a better option and a worse one. They are between options that are better at different things. One platform integrates beautifully and reports deeply; another is up and running in a fortnight. Pick either and you are accepting a blend of strengths and weaknesses. The only question is whether you blended them deliberately, or let your gut do it unsupervised.
This piece walks one such choice end to end: how to state what matters, how to put honest weights on it, what the arithmetic does and does not settle, and what to do with cost. Every number in it belongs to a single worked example, so you can follow the same case from first score to final ranking.
Three options, three kinds of better
The example that carries the whole piece: a marketing team is choosing an automation platform. Three are on the shortlist, call them A, B and C. Three criteria matter: integration fit (how well it talks to the CRM and the data warehouse), reporting depth, and ease of adoption. A is the premium suite: the strongest integration and the deepest reporting on the table, and the hardest to roll out. C is the budget option: weakest integration, thinnest reporting, pleasant enough to learn. B sits between them on capability and is by far the easiest of the three to adopt.
Asked to choose, the gut does something quite specific. It collapses three judgements into one feeling, "A, probably", and the exchange rates it used to do the collapsing, how much reporting it traded for how much adoption pain, are nowhere on view. Not hidden deliberately; simply never stated, not even to the person whose gut it is.
Watch what that does to the meeting. The integration architect wants A; the team lead who has to land the rollout wants B. They will spend an hour arguing about options, trading anecdotes about A's API and B's onboarding, and get nowhere, because they do not actually disagree about the options. Ask each to score the three platforms on the three criteria and their scores come back nearly identical. What they disagree about is the weights: how much integration matters relative to a painless rollout. Since the weights are never on the table, the disagreement gets argued by proxy, through whichever option each weight-set happens to favour, and the proxy argument cannot be settled because nobody can see what it stands for.
Importance is meaningless without ranges
The obvious fix is to ask the room to rate each criterion "by importance", ten for critical, one for trivial. It feels rigorous and it is close to meaningless, because importance in the abstract says nothing about this choice. Price can be the most important thing in your life and still deserve zero weight here: if the three platforms differ by €24 a year, price cannot tell them apart, however much you care about money in general. A criterion earns weight from the gap between the worst and the best on offer, not from its name.
So the question that works is not "how important is integration?" but: how much is it worth to move from the worst integration on the table to the best? Weight the swing, not the word. The technical name is swing weighting, and it is the difference between a weight you can defend and a number you made up.
Walk it through the example. Score each platform on each criterion, 0 to 100. On integration fit the scores run from 35 (C) up to 90 (A). The judge looks at the three swings, worst-on-offer to best-on-offer, and asks which swing they would most want to capture. Integration's, they decide: pin it at 100. Against that benchmark, capturing the full reporting swing is judged worth 60, and the full ease-of-adoption swing 40. Divide each by the total (200) and the weights are 0.50, 0.30 and 0.20. Half the total weight on integration, not because integration is "important" in general, but because the options differ widely on it and that difference is worth the most.
The arithmetic, and what it doesn't settle
The rest is multiplication. Each option's score on each criterion, times that criterion's weight, summed into a single value.
| Option | Integration · 0.50 | Reporting · 0.30 | Ease · 0.20 | Value |
|---|---|---|---|---|
| A (premium) | 90 | 95 | 40 | 81.5 |
| B (mid) | 70 | 60 | 80 | 69.0 |
| C (budget) | 35 | 30 | 75 | 41.5 |
B: .50×70 + .30×60 + .20×80 = 35 + 18 + 16 = 69.0 · C: 17.5 + 9 + 15 = 41.5
A leads, comfortably it seems. Now the honest part. That ranking is only as strong as its weakest weight, and the weights were judgements, made in minutes, by one person. So the right question is not "is 81.5 bigger than 69.0", the spreadsheet has that covered. It is: which weight would have to change, and by how much, before the leader changes?
In the example, the answer is reassuring in one direction and usefully specific in the other. A's lead survives any weight on integration: slide that weight from zero to one, letting the other two share the remainder in their old proportion, and A stays on top the whole way. Argue about integration as long as you like; it cannot change the decision. Ease of adoption is the live wire: B overtakes A only if its weight climbs above about 0.39, and it currently carries 0.20. The judgement would have to nearly double. If the team lead can argue, with a straight face, that a painless rollout is worth almost twice what the room granted it, the decision flips, and that is now a precise, inspectable claim instead of an hour of anecdotes. Checking one weight at a time like this is called one-way sensitivity analysis, and it is where the method earns its keep.
The output is not the ranking. It is knowing which weight the ranking hangs on.
Two cautions before moving on. First, be clear what kind of model this is: a model of preference, not of uncertainty. Nothing here says how likely anything is; the scores are statements about how good each option is on each criterion, taken at face value. Second, the addition leans on an assumption worth naming: the worth of a point of reporting depth must not depend on the integration score. The label is preferential independence. If it fails, if reporting only matters to you when integration is strong, then the criteria are entangled, and the repair is not cleverer weights but restructured criteria: fold the entangled pair into one criterion that you can score honestly on its own. (The whole apparatus, for the record, is the simplest member of multi-criteria decision analysis: a linear additive value model.)
Cost stays out of the score
One criterion has been conspicuously absent: price. Deliberately. Fold cost in as just another weighted criterion and it quietly buys itself a share of the value score, and the one trade the decision actually turns on, value against money, disappears into the blend. Keep money out of the score and plot the two against each other instead: value up the side, annual cost along the bottom.
Read the plot with one rule: an option is in trouble if another option sits above and to its left, more value for less money. An option beaten that way is called dominated, and removing it is the only free lunch in the whole exercise: you give up nothing by striking it. Whatever survives the cull forms a staircase from cheap-and-modest to expensive-and-strong, the efficiency frontier.
In the example: A delivers 81.5 points of value for €60,000 a year, B delivers 69.0 for €36,000, C delivers 41.5 for €18,000. Nothing sits above-left of anything: pay more, get more, the whole way up. So none of the three is dominated, and the frontier does not decide the choice. What it does is clarify it: the question left standing is what a point of value is worth in money. Stepping up from C to B buys 27.5 points for €18,000 more, roughly €650 a point. Stepping from B to A buys 12.5 points for €24,000 more, roughly €1,900 a point. Steeper, by a factor of three. Whether that is worth paying is not an arithmetic question, and, as with the bar in the first series, the exchange rate you will accept should be stated before you look at the chart, not reverse-engineered from the option you already wanted.
What this piece holds still. The weights here are one judge's; a real shortlist usually has several judges, and their disagreement about weights is information this model does not capture, only exposes. The criteria are treated as preferentially independent; if scoring one honestly requires knowing another, restructure before weighting. And the scores themselves are assumed honest, which is its own discipline, with its own failure modes: see elicitation methods.