Jump to Range Model: Baseline / Scenario
Forecast A reasoned estimate of what may happen in the future, based on evidence and assumptions. It is a prediction with uncertainty, not a promise.
Why it matters + example Why it matters: It keeps a decision useful without disguising uncertainty as certainty.
Illustrative example: An illustrative ecommerce forecast estimates the extra contribution a category-template rebuild could create over 12 months.
Do not confuse it with: A target, commitment or guaranteed result.
# Range Model: Baseline
Baseline / no-change baseline The best estimate of what would happen over the forecast period if the proposed SEO work did not happen.
Why it matters + example Why it matters: Incremental value must be measured against what would probably happen anyway, not against zero.
Illustrative example: If eligible demand is already expected to grow by 8%, the project cannot claim all of that growth.
# Range Model: Baseline
Forecast horizon The period the forecast is trying to cover, such as 12 months.
Why it matters + example Why it matters: Evidence and error should be judged over the same period as the real decision where possible.
Illustrative example: A forecast used for an annual budget decision has a 12-month horizon.
# Range Model: Baseline / Backtest
Seasonal-naive baseline A simple baseline that repeats the comparable previous seasonal period.
Why it matters + example Why it matters: It gives a transparent benchmark. A more complex method should earn its place by performing better on unseen history.
Illustrative example: Use last October's weekly demand pattern as the starting estimate for this October.
Do not confuse it with: Blindly applying last year's growth rate to every future month.
# 02 Describe what could change Range Model: Slice
Eligible demand The impressions or search demand genuinely relevant to the intervention, after excluding demand the work cannot reasonably influence.
Why it matters + example Why it matters: A large market is not automatically an addressable SEO opportunity.
Illustrative example: A UK category-template rebuild excludes branded and overseas queries it will not affect.
# Range Model: Scenario
Incremental visibility The share of baseline demand that becomes additional impressions because of the intervention.
Why it matters + example Why it matters: It describes added exposure rather than assuming a particular ranking gain.
Illustrative example: The rebuild is assumed to create visibility across 22% of eligible impressions.
Do not confuse it with: A ranking uplift.
# Range Model: Slice / Learn
CTR Click-through rate: clicks or visits divided by impressions for the defined behavioural group.
Why it matters + example Why it matters: It turns incremental impressions into incremental visits, so the query group and search-results context matter.
Illustrative example: At 5.8% CTR, 100,000 incremental impressions produce 5,800 visits.
# Range Model: Learn
Conversion rate / order CVR The share of visits that create the chosen outcome, such as an order. CVR means conversion rate.
Why it matters + example Why it matters: The denominator and attribution rules materially change the commercial result.
Illustrative example: At a 3.4% order conversion rate, 5,800 visits produce about 197 orders.
# Range Model: Learn
Commercial unit / contribution per order The finance-approved value attached to one incremental outcome.
Why it matters + example Why it matters: Contribution is usually more useful than gross revenue when the decision concerns economic value.
Illustrative example: Each illustrative incremental order contributes £468 after the agreed costs.
Do not confuse it with: Topline sales revenue.
# Range Model: Slice
Slice / behavioural group A group of queries, pages or observations whose response behaviour is similar enough to model together.
Why it matters + example Why it matters: One average can hide material differences, but excessive slicing leaves unstable evidence.
Illustrative example: Model branded, non-branded and product-led queries separately only when their click behaviour differs reliably.
# Range Model: Learn
Response curve The observed relationship between an input such as position or visibility and an outcome such as click-through rate.
Why it matters + example Why it matters: It replaces a generic industry assumption with evidence about how this kind of demand responds.
Illustrative example: The curve shows how CTR changed for comparable ecommerce queries as visibility improved.
# Range Model: Learn
First-party evidence Evidence from the organisation's own historical search, analytics, conversion or delivery data.
Why it matters + example Why it matters: It is usually more relevant than a broad benchmark when definitions and samples are stable.
Illustrative example: Use the site's own query-group CTR history before borrowing an industry curve.
# Range Model: Learn
Prior / labelled prior A wider benchmark or outside assumption used when first-party evidence is sparse.
Why it matters + example Why it matters: Labelling borrowed evidence stops it being mistaken for an observed truth about this site.
Illustrative example: Use an external CTR benchmark as a labelled prior until enough comparable first-party data exists.
# Range Model: Baseline / Backtest
WAPE Weighted Absolute Percentage Error: total absolute forecast error divided by total actual volume.
Why it matters + example Why it matters: It gives a compact view of how wrong a baseline has historically been across the chosen horizon.
Illustrative example: If absolute monthly errors total 120,000 and actual volume totals 1,000,000, WAPE is 12%.
Do not confuse it with: An automatic plus-or-minus interval for the next forecast.
# Range Model: Backtest
Signed bias A check for whether forecasts tend to be systematically too high or too low.
Why it matters + example Why it matters: A model can have an acceptable average error while leaning in one direction repeatedly.
Illustrative example: Six consecutive over-forecasts suggest optimism in the baseline or response assumptions.
# Range Model: Baseline / Backtest
Rolling-origin backtest / rolling forecast-origin evaluation A repeated test that makes historical forecasts using only the information available at each point in time.
Why it matters + example Why it matters: It answers: if we had used this method then, how wrong would we have been without hindsight?
Illustrative example: Forecast each historical 12-month period in turn, then compare each forecast with its outturn.
Do not confuse it with: Fitting a model once using the whole historical dataset.
# Range Model: Scenario
Scenario A coherent conditional story about the future using assumptions that make sense together.
Why it matters + example Why it matters: It shows how the result changes across plausible conditions without pretending each condition is certain.
Illustrative example: A cautious scenario combines weaker demand, response and a delayed launch.
Do not confuse it with: A probability. Conservative, central and upside scenarios do not automatically have likelihoods.
# Range Model: Scenario
Decision range The span of outcomes across named, plausible scenarios.
Why it matters + example Why it matters: It keeps different futures visible without borrowing statistical confidence the model has not earned.
Illustrative example: The illustrative lower, central and higher scenarios produce a £40k-£220k decision range.
Do not confuse it with: A calibrated confidence or prediction interval.
# Range Model: Scenario
Central scenario The scenario best supported by the current baseline, evidence and delivery plan.
Why it matters + example Why it matters: It gives the discussion a useful centre while remaining conditional.
Illustrative example: The central case assumes the agreed launch date and the most defensible response rates.
Do not confuse it with: The promised or most likely result.
# Range Model: Scenario / Backtest
Prediction interval A probabilistic range with a stated coverage claim, produced from a forecast distribution and calibrated against comparable outturns.
Why it matters + example Why it matters: Calling a hand-built range an 80% or 90% interval makes a statistical promise the model may not support.
Illustrative example: An 80% interval should contain roughly 80% of comparable future outturns when properly calibrated.
Do not confuse it with: A lower-central-higher decision range.
# Range Model: Learn / Scenario
Sensitivity analysis A test of how much the result changes when an assumption changes.
Why it matters + example Why it matters: It reveals which uncertainty is most worth reducing before more money or time is committed.
Illustrative example: Vary demand, visibility, CTR and conversion to see which creates the largest value swing.
# Range Model: Learn / Scenario
One-way sensitivity Change one input at a time while holding the others fixed.
Why it matters + example Why it matters: It is useful for screening important inputs, but can miss relationships between assumptions.
Illustrative example: Change CTR from 4.6% to 6.6% while leaving every other central input unchanged.
# Range Model: Scenario
Correlated drivers Inputs that may move together rather than independently.
Why it matters + example Why it matters: Independent one-way changes can create combinations that would not make sense in the real world.
Illustrative example: A search-results redesign may reduce both visibility and CTR at the same time.
# Range Model: Gate
Gate The step that separates the size of the opportunity from whether the organisation delivers the work and the expected response occurs.
Why it matters + example Why it matters: Theoretical value is not forecast-period value if the work is delayed, partial or never released.
Illustrative example: A £200k conditional opportunity is reduced when the rebuild has only a 40% chance of shipping this year.
# Range Model: Gate
Delivery certainty Confidence that the organisation can ship the defined intervention in the planned window.
Why it matters + example Why it matters: Ownership, capacity, dependencies and historical delivery performance affect how much value can be captured.
Illustrative example: A signed-off release slot gives stronger delivery evidence than a ticket in the backlog.
Do not confuse it with: Evidence that the SEO response assumption is correct.
# Range Model: Learn / Gate
Evidence quality How well the response assumptions are supported by relevant, stable and recent evidence.
Why it matters + example Why it matters: A project can be certain to ship while its expected response remains poorly evidenced.
Illustrative example: A tested first-party CTR curve has higher evidence quality than an unlabelled industry benchmark.
Do not confuse it with: Delivery certainty.
# Range Model: Gate
Months live How many months inside the forecast horizon the intervention is expected to be active.
Why it matters + example Why it matters: A delayed launch reduces value captured inside the period without necessarily changing lifetime opportunity.
Illustrative example: A project launched in April is live for nine months of a calendar-year forecast.
# Range Model: Gate
Expected value / expected contribution The probability-weighted average of mutually exclusive outcomes.
Why it matters + example Why it matters: It can support a decision when outcome probabilities are defensible, but it is neither a promise nor necessarily the most likely result.
Illustrative example: Weight the contribution of ship, delay and no-ship outcomes by their assessed probabilities.
# Range Model: Gate
Expected net contribution Expected contribution minus the delivery cost included in the decision.
Why it matters + example Why it matters: It compares the probability-weighted value with the investment required to pursue it.
Illustrative example: £260k expected contribution minus £180k delivery cost gives £80k expected net contribution.
# Range Model: Backtest
Backtest A comparison between a forecast or forecasting method and what later actually happened.
Why it matters + example Why it matters: A miss is useful only if it reveals whether the baseline, delivery or response needs improving.
Illustrative example: Save the original forecast, record the outturn and inspect which assumption failed.
# Range Model: Backtest
Backtest - baseline error The difference between the no-change demand forecast and what demand actually became.
Why it matters + example Why it matters: If this explains the miss, improve the demand baseline rather than blaming the intervention model.
Illustrative example: Eligible demand fell before launch, so the project had less opportunity to influence.
# Range Model: Backtest
Backtest - delivery variance The difference between the planned and actual delivery date.
Why it matters + example Why it matters: A late launch should update delivery assumptions rather than automatically discrediting the SEO response model.
Illustrative example: The rebuild shipped three months late and captured only nine months of year-one value.
# Range Model: Backtest
Backtest - response error The difference between the conditional uplift forecast and the response observed after the work shipped.
Why it matters + example Why it matters: It shows whether visibility, click or conversion assumptions need revising.
Illustrative example: Demand and launch timing were right, but click response was 25% weaker than assumed.
# Range Model: Backtest
Scenario hit-rate How often comparable outturns land inside the decision range.
Why it matters + example Why it matters: It is a useful diagnostic for whether the range is routinely too narrow or too wide.
Illustrative example: Seven of ten comparable projects finished inside their named decision ranges.
Do not confuse it with: Proof that a hand-built range is a calibrated prediction interval.
# Range Model: Learn / Backtest
Reference class A set of genuinely comparable historical interventions or outcomes used to inform future assumptions.
Why it matters + example Why it matters: Comparable work provides a firmer outside view of delivery likelihood and response.
Illustrative example: Use previous category-template rebuilds with similar scope, teams and markets.
# Range Model: Backtest
Shock A material algorithm, measurement, market, product or organisational event that changes the forecasting environment.
Why it matters + example Why it matters: Recording when it became knowable protects the original forecast from hindsight editing.
Illustrative example: A search-results change reduces eligible clicks halfway through the forecast period.
# Range Model: Scenario / Gate
Programme forecast A forecast across multiple SEO initiatives that accounts for overlap, dependencies and shared shocks.
Why it matters + example Why it matters: Adding every task-level opportunity usually overstates the programme result.
Illustrative example: Combine technical, template and content initiatives after removing shared demand and sequencing their delivery.
# Range Model: Slice / Scenario
Deduplicate demand Prevent two initiatives from both claiming the same underlying search opportunity.
Why it matters + example Why it matters: The same impression cannot become incremental twice.
Illustrative example: A template rebuild and an internal-linking project cannot both claim all category-query demand.
# Range Model: Gate
Sequence dependencies Model work in the order it must actually happen.
Why it matters + example Why it matters: A later initiative cannot benefit from an enabling change that never ships.
Illustrative example: Forecast content expansion only after the new templates and publishing workflow are available.
# Range Model: Scenario / Backtest
Shared shocks Factors that can affect several initiatives at once and create correlated outcomes.
Why it matters + example Why it matters: Capacity constraints, brand demand and algorithm changes can move a whole programme together.
Illustrative example: One development freeze delays three initiatives rather than behaving like three independent risks.
# THE POINT
Uncertainty does not disappear when the spreadsheet opens. Good forecasting makes the assumptions visible, tests what matters and leaves something useful to learn from later.
Build a decision range →