Sena measures the forward half directly with real-world signals.
Revenue prediction converts a history of transactions into a statement about a period that has yet to happen. Three properties of that conversion put a ceiling on how far ahead it can see.
A model learns the relationships present in its training window. Those relationships held while the market held, and a shift in what consumers want arrives as an error term before it arrives as a pattern.
Elasticity gets estimated from price points already tried. A pack size, price or format the market has yet to see has zero history, so the one question a plan most wants answered sits beyond the training data.
Model performance improves with more features and better tuning, and a better fit explains more variance while explaining zero of the reason. A prediction that lands wide leaves the corrective action open.
Sena measures stated intent alongside recorded purchase from the same consumers, which gives a model a forward feature over a lagged one.
How many shoppers a specific change would move, measured before the change, per market.
Purchase captured this week, so the input series ends now over a closed quarter.
A pack, price, or format absent from the market takes its size from the shoppers who would buy it.
Revenue predictive analytics applies statistical or machine-learned models to historical data in order to state what revenue will be. The modelling half is well developed.
Two things get called prediction, and they use opposite evidence.
Extrapolative prediction learns from closed periods and projects the relationships it found. It is strong on stable markets and weak precisely when a market turns.
Forward-looking measurement asks consumers what they would do under a stated condition and converts the answer to a volume. It reaches options the market has yet to offer.
A model with both performs differently from a model with one. The stated response enters as a feature carrying information the transaction history omits by construction.
Three failure modes recur, and they share a cause.
Consumer preference shifts, and the model reads the first months as noise, since the pattern is absent from training.
A launch, a pack change, or a price move outside the historical range has zero precedent to learn from.
Units unsold get labelled as weak demand where the item was absent from the shelf, so the model encodes a supply failure as a demand relationship.
| Output | What it settles | Where it goes wrong |
|---|---|---|
| Stated response | How many shoppers would move. | Estimated from past price points. |
| Current base | What is selling this cycle. | Read from a closed period. |
| Availability | Whether stock was present. | Learned as demand. |
| Per-market signal | Which market turns first. | Pooled into one regional series. |
The first is a measurement of the future, and the other three correct the past. A model given all four predicts differently from one given a transaction table.
Catching a turn early depends on four collected inputs, refreshed each cycle across the model's markets.
| Input | What it answers |
|---|---|
| Receipts | What was bought this cycle, and at what price. |
| Store captures | Whether the item was available to buy. |
| Geo-verified photos | The shelf as it stood, dated and placed. |
| Stated preference | The change that shifts a purchase, and the count behind it. |
The modelling stack stays in place. Transaction history, the revenue model, price files, and the planning system connect through 250+ integrations, so a measured forward feature enters the model already in place.
Three inputs sit outside every transaction table.
Real people share what they buy and prefer, under explicit consent. Sena captures it directly at the source, so every figure traces back to where it came from whenever a number comes under question.
Stated response and recorded purchase come from the same consumers, so a forward feature and a historical one describe one population.
The stated response enters the model as a variable, over a judgement applied on top of it.
A pack or price absent from history gets a measured volume from the shoppers who would buy it.
Photographs establish availability, so an empty shelf trains as a supply event over weak want.
A revenue model trained on transactions can only learn relationships its training window contains. Sena builds every figure from real-world signals captured in the current cycle, from the shelf photographed this week through to the shopper stating what would move the next purchase.
The purchase and its price are recorded this cycle, so the input series ends at the present.
Reads what the shelf held, off images captured in real outlets, so availability enters as its own variable.
Signal arrives from the consumer network under explicit consent. What would move a future purchase comes from the person who would make it.
Transaction history, the revenue model, price files, and the planning system connect over 250+ integrations, so a forward feature joins the existing pipeline.
Market, week, and source consumers attach to each figure, so a model input holds when the prediction is challenged.
Prior model cycles beside transaction history and price files, scan by scan.
Four teams work off the same projection, and each one needs a different part of the forward half.
The revenue model. Needs a forward feature over a lagged one.
The projection. Needs untried options sized.
The reforecast. Needs the market that turns first named.
The price decision. Needs response measured outside past range.
The same forward feature, stated against whatever each industry turns on.
Pack switching sized per market, with availability separated.
Occasion shift stated before it appears in sales.
Own-label switching intent, stated before it appears.
Provider switching intent, stated before it lands.
One mechanism, refreshed each cycle in each market inside the model. Each step is documented, which is what lets a measured feature enter a governed model.
Current purchase and stated next choice come from one consumer, so lagged and forward features share a population.
Photographs separate absent stock from weak demand, so the training data holds a supply event as itself.
Stated response becomes a share of category shoppers per market, and the model reads that share as a variable.
Most revenue prediction learns from a closed transaction history, and the question a plan asks is usually about something absent from it. Sena supplies a forward feature.
| Capability area | Typical setup | Sena |
|---|---|---|
| Feature basis | Closed transaction periods. | Stated intent plus current purchase. |
| An untried option | Zero precedent to learn from. | Sized by the shoppers who would buy. |
| A regime change | Arrives as an error term. | Stated before the pattern forms. |
| Availability | Learned as weak demand. | Separated by shelf photographs. |
| Series recency | Ends at the last close. | Ends at the current cycle. |
| Granularity | Pooled regional series. | Held per market. |
| A challenged prediction | A model diagnostic. | Open any input to its consumers. |
Three prediction problems a transaction history holds zero rows on.
Measure the response to a pack, price, or format the market has yet to see, so the model gains a figure over a guess.
Read stated intent this cycle, so a shift reaches the projection ahead of the error term.
Separate absent stock from weak demand, so a supply failure trains as itself over a preference.
A forward feature settles what the model could not learn. The four decisions below are what a finance team runs around it.
Every finance decision Sena answers, in one place
The whole statement forward
Holds the full forward view this prediction feeds.
Explore financial forecasting 02The period that closed
Reads what already happened.
Explore financial analytics 03Where the projection commits
Turns a projection into a number the business is held to.
Explore financial planning and analysis 04Delivery against the projection
Runs the business against the figure once it is accepted.
Explore financial performance managementThe walkthrough takes one revenue projection, reports stated consumer response per market inside it, sizes one option absent from the range, and separates availability from demand while the team watches.
Name the projection and the markets inside it, and we will supply the forward feature.