Expected value and the weighted scoring matrix at a glance
The expected value framework compares options whose results are uncertain. For each option it lists the outcomes that matter, gives each a probability so that they add up to the whole, gives each a value, and adds the products. The site's clearest application is the guide to valuing a lead by pipeline stage: a lead is worth the margin of the deal it may become multiplied by the share of similar leads that close. The same idea sets a Conversion ValueThe monetary value an advertiser assigns to a conversion, such as the order's revenue or a lead's expected worth. for bidding, ranks a test backlog and builds a forecast. The framework page is plain about the weak point: the result is only as good as the probabilities, and an average hides how widely the real result can swing.
The weighted scoring matrix compares a few known options against criteria of unequal importance. Must-haves are applied as gates first; criteria and weights are agreed and dated before anyone sees the options; each score is backed by recorded evidence against written anchors; and the ranking is re-run under other plausible weights to see whether it holds. The framework page treats the total as a prompt for discussion rather than a verdict, because every number in it rests on judgement. The guide to choosing and evaluating marketing tools builds one for software.
So the difference is not rigour against rough judgement. Both rest on judgement. Expected value puts it into probabilities and money values; the matrix puts it into weights and scores. The question is which kind of judgement the decision actually involves.
Compare table
Expected value vs Weighted scoring matrix
| Row label | Expected value | Weighted scoring matrix |
|---|---|---|
| Question answered | What is each option worth on average, given that its result is uncertain? | Which of a few known options best fits what this decision must achieve? |
| Inputs | The possible outcomes of each option, a probability for each and a value for each | Options, must-have gates, criteria, weights, scoring anchors and evidence-backed scores |
| Unit | One currency, usually margin | No unit; a weighted total used for ranking |
| Handling of uncertainty | Built in: uncertainty is the probabilities, and ranges show how much it matters | Mostly hidden unless forced: a guessed score looks like a measured one unless labelled |
| Handling of unlike criteria | Poor: every effect must be priced, so qualities with no money value tend to be left out | Built in: each quality gets its own criterion and weight |
| Number of options | Any number of repeated, similar choices, such as many leads, offers or tests | A handful of substantial options, fixed before scoring |
| Where judgement enters | The probabilities and, less often, the values | The criteria, the weights and each score |
| Output | A value per option in money, and a range when probabilities are stated as ranges | A ranking with its sensitivity runs and the evidence behind each score |
| How it misleads | A guess written as a probability looks precise; an average hides a ruinous worst case | Weights chosen after a favourite emerges rig the result; overlapping criteria count one quality twice |
| Typical misuse | Valuing a single irreversible bet by its average | Scoring a long backlog, or adding a cost criterion that ignores what each option would earn |
The rows that decide most cases are the unit and the handling of unlike criteria. If every effect that matters can be expressed in money and given a probability, expected value gives a cleaner answer than any score. If the options differ on qualities that have no honest money value, the matrix is the more truthful tool, because it does not pretend those qualities can be priced.
When to choose expected value
Choose expected value when the options are bets whose outcomes can be valued in money and whose probabilities can be read from the business's own history: which lead types deserve more spend, what an enquiry is worth as a bid, which offer to run, which test to run next. The probability is usually a Win RateShare of closed sales opportunities that were won. or a stage-to-close rate, and the value is Contribution MarginRevenue left after every cost that varies with each sale. rather than revenue, as the lead valuation guide sets out. The CPA calculator turns a lead value and a close rate into a maximum cost per lead, a quick check that a target and the value behind it agree.
It suits many small, repeated choices, where results tend towards the average. The guide to prioritising tests lists it as one way to score a backlog, the probability of winning multiplied by the value of a win and set against cost, and calls it the most honest framing of what a test is worth while warning that most teams lack calibrated probabilities. A HypothesisA testable statement of the change to make, the effect expected on a named metric and the evidence-based reason for expecting it. backed by strong research earns a higher probability than a hunch; calibrating that probability is the expected value page's subject, not this one's.
Expected value is the weaker choice for a single large, irreversible commitment, where the business will experience one outcome rather than the average. The framework page points to scenario planning when the spread of outcomes matters more than the average or when probabilities cannot be estimated at all. It is also weak whenever the main differences between the options are qualities nobody can price.
When to choose a weighted scoring matrix
Choose the matrix when a few known options differ on several qualities that matter unequally and do not share a unit: a booking system, a CRM, an agency, the next market. The framework page's rules carry the method: gates before scoring, weights agreed and recorded before anyone sees a demo, anchors for each score, evidence for each rating, the incumbent or doing nothing scored on the same terms, and a sensitivity check. Tie each heavy criterion to the KPIA metric tied directly to an objective. the choice is meant to move, so a heavy weight reflects a requirement rather than a preference.
It suits choices that have to be explained later, because the matrix records what was valued, by how much and on what evidence. It is the wrong tool for a long list of small items, which the comparison of the impact-effort matrix and the weighted scoring matrix sends to a quick two-by-two sort instead.
The matrix is the weaker choice when the options differ mainly in money that depends on uncertain results. Scoring "revenue potential" from one to five throws away the information a probability and a margin would carry, and compresses large differences into a narrow scale.
Decision table
| Situation | Choose | Why |
|---|---|---|
| Deciding how much each lead source or enquiry type is worth | Expected value | Close rates and margin price each source in one currency |
| Ranking a test backlog where past results give a win rate | Expected value, after the traffic check | Probability of winning times value of a win is the most direct measure |
| Choosing a tool, agency or platform from a shortlist | Weighted scoring matrix | Integration, ease of use and support share no unit and matter unequally |
| Choosing a market to enter next | Weighted scoring matrix, with expected value for the financial criterion | Market qualities differ in kind, but the payoff can still be priced |
| One large commitment that could not be undone | Neither alone; weigh the worst case and consider scenarios | An average and a total both hide the outcome the business cannot absorb |
| No history for probabilities and no agreed criteria | Settle the criteria first, and state probabilities as ranges | Without either, both methods produce precise-looking guesses |
| Many similar, repeated bets such as offers or bids | Expected value | Results tend towards the average over many choices |
Using expected value inside a scoring matrix
Most real choices between options have a money side and a quality side. A booking system has a cost and may bring extra margin; it also has an interface, integrations and a vendor. The clean way to handle both is to let expected value supply the matrix's financial criterion. Work out the expected net value of each option over the same period, the probability-weighted benefit less the cost, and score that criterion from it against written anchors. The rest of the matrix carries the qualities that cannot be priced.
That rule has a second half: once the priced effect is inside the financial criterion, do not add it again elsewhere. If a feature's value is already counted as expected margin, a separate criterion for the same feature counts the same guess twice and quietly doubles its weight, which the weighted scoring page lists among the matrix's failures as overlapping criteria. And do not reduce the financial criterion to price alone: a cheaper option that earns less can look better on cost and worse on value.
Read the money side as the business's next best use, not as a comparison with doing nothing. The opportunity cost framework makes that point for spend, time and traffic. And when the open question is how large the whole budget should be rather than which option to fund inside it, the comparison of top-down and bottom-up budgeting covers that decision. The frameworks library lists the other planning frameworks by the decision they serve.
A kitchen fitter compares three lead sources and two booking systems
Dadgyux Kitchens, an invented kitchen design and fitting firm, reads its last 12 months of enquiries by source. Of 900 paid search enquiries, 54 became fitted kitchens (6%); of 240 showroom visits, 36 did (15%); of 60 referrals from builders, 15 did (25%). Each fitted kitchen leaves a contribution margin of £3,200, or £2,800 after the builder's referral fee. A paid search enquiry is therefore worth 0.06 × £3,200 = £192, a showroom visit £480 and a builder referral £700. Paid search enquiries cost £60 each, so the firm sets a ceiling of £96 per enquiry, half the value, and moves time from general advertising to visiting builders. The builder rate rests on only 15 kitchens, so it also checks 20% and 30%, which give £560 and £840: referrals stay the most valuable source either way.
It then chooses between two booking systems. Before any demo it shares 100 points: integration with its design software 30, ease of use for fitters on a phone 25, expected net value over 3 years 25, and support 20. System A costs £4,500 over 3 years and sends automatic survey reminders; the firm judges a 60% chance that they recover 20 missed home surveys a year, each worth £180 in expected margin, so A's expected net value is 0.6 × 20 × 3 × £180 − £4,500 = £1,980. System B costs £2,400 and recovers 8 missed surveys a year on the same terms, so B's expected net value is 0.6 × 8 × 3 × £180 − £2,400 = £192. A scores 4 and B 2 on that criterion; there is no separate reminders criterion, because the reminders are already priced. The totals are A 355 and B 345, and moving 10 points from the financial criterion to ease of use puts B ahead. The firm treats it as a tie on ease of use and runs a two-week trial with two fitters. Both manage A's phone screens, ease becomes a tie, and A is chosen on integration and expected value.
Common mistakes
- Pricing qualities that have no honest money value. Fit, support and migration risk belong in a matrix, not in an expected value.
- Scoring money on a one-to-five scale. Where outcomes can be priced, price them and score the criterion from the result.
- Scoring cost without value. A cheaper option that earns less is not the better financial choice.
- Counting a priced benefit twice. Once it is in the financial criterion, it needs no criterion of its own.
- Using an average for a bet that cannot be repeated. Weigh the worst case of a large, irreversible choice.
- Probabilities and weights written without a source. Label each as measured or judged, with its basis.
- Treating either number as the decision. Read the range of an expected value and the sensitivity runs of a matrix before acting.
The comparison library lists the other pairs that are easy to confuse.
