Will we hit this month's target? A month before the books close
Business risk usually shows up after month-end closing, when the only thing left to do is explain it. This is the predictive-management work for CJ CheilJedang that forecast daily sales with 97.6% accuracy for the Korea Food business unit and turned the forecasts into a traffic-light signal — so teams see target risk about a month ahead, while there is still time to act.
From explaining the month to changing it
Business risks are often identified only after month-end closing. For proactive management, the company needed to know whether monthly targets were likely to be achieved before the month ended — early enough for a sales push, a promotion or a cost decision to make a difference.
What was built
- Scope with the domain experts. Required data and modelling scope were defined together with domain experts. Data on sales, cost of sales, selling and administrative expenses and other business indicators was collected and processed, and the predictable areas were separated from the highly uncertain ones — the latter to be monitored, not forecast.
- Daily sales forecasting. Daily sales models were developed for the business, reaching 97.6% prediction accuracy for the Korea Food business unit's sales.
- A signal people act on. A traffic-light signal system and dashboard translate forecasts into whether the monthly target is at risk, so business users can understand the risk quickly and respond.
Live model, computed in your browser on a generated business unit of six product lines over three years, with a weekday rhythm, an end-of-month push, Seollal and Chuseok weeks, seasonality and persistent monthly swings; amounts are shown as a percentage of target. Left: the month-to-date run-rate and its signal. Right: a daily model fitted on two years, its projection and probability of hitting the target, and its signal from a month before the month starts. The table compares both across the test year. The features of the demo business are invented to show why a calendar-aware forecast beats a run-rate. Open the live model on its own page ↗
Architecture, stack and core formulation
Daily sales forecasts, a projection of each month's total from actuals plus forecasts, a probability of reaching the target, and a traffic light on a dashboard.
Predictable vs uncertain
With domain experts, separate the indicators that can be forecast from those that should only be monitored.
Business indicators
Sales, cost of sales, selling and administrative expenses and other indicators collected and processed.
Daily sales models
Daily sales forecasting models with engineered calendar and business features.
Month total and risk
Actuals to date plus forecasts for the remaining days, with a probability of reaching the target.
Traffic light
Green, amber or red on a dashboard, about a month ahead of closing.
| Layer | Technology | What it does here |
|---|---|---|
| Data | Integrated sales, cost and expense indicators | One daily view of the business unit |
| Forecasting | Daily sales forecasting models, feature engineering | 97.6% accuracy for the Korea Food business unit |
| Risk | Month-end projection with uncertainty | Probability of reaching each monthly target |
| Delivery | Traffic-light signal system and dashboard | Target risk about a month before closing |
month total at day d T̂_m(d) = Σ_(t ≤ d) y_t + Σ_(t > d) ŷ_t
risk P(hit) = P( T_m ≥ target_m | data up to d )
signal green if P(hit) ≥ p_hi; amber if p_lo ≤ P(hit) < p_hi; red otherwise
live model log ŷ_t = calendar(t) + level(d) · φ^(months ahead)
P(hit) = Φ( log(T̂ / target) / σ(d) ), p_hi = 0.65, p_lo = 0.35- Calendar-aware beats run-rate. Weekday rhythm, holidays and month-end order patterns bias a month-to-date run-rate every month; a daily model does not have that bias.
- Uncertainty shrinks as the month fills. The same projection becomes more confident each day as actuals replace forecasts.
- Built for business users. The output is a probability, shown as a traffic light, so the people who act on it can read it at a glance.
| Component | In production | In the live model above |
|---|---|---|
| Models | Daily sales forecasting models (time-series, engineered features) | A calendar regression per product line with a current level |
| Signal | Traffic-light system and dashboard | The same idea with thresholds at 65% and 35% |
| Data | Korea Food business-unit sales and P&L indicators | Six generated product lines, amounts as % of target |
Design notes
Know what can be forecast
Not every line of a P&L is forecastable. Sales with stable calendar structure can be predicted well; one-off cost items cannot. Separating the two up front kept the signal honest: a red light means the model expects a miss, not that something unpredictable might happen.
Why the run-rate misleads
Month-to-date sales scaled to a full month assume every day is alike. They are not — weekends, holiday weeks and the push of orders into the last days of the month make the early weeks look weak every month. A run-rate signal therefore cries wolf until late in the month, and teams learn to ignore it. A daily model that knows the calendar does not have that bias.
A probability, shown simply
The model's output is a probability of reaching the target; the dashboard shows it as green, amber or red. The simplicity is deliberate: the people who act on it are business teams, and the shift the project aimed for was from after-the-fact review to proactive decisions.
Outcome
The signal system enabled business teams to identify monthly target risk about one month in advance, shifting management from reviewing the month after it closed to acting on it while it was still open.
Limitations
- The signal is only as good as the target: targets set without regard to what is achievable produce red lights that are correct and unhelpful.
- Shocks the calendar does not know about — a recall, a sudden price change — show up in the forecast only once they show up in sales.
- The live model's business unit, lines and targets are generated; its accuracy figures describe that generated business, not CJ CheilJedang's.
About the demo and confidentiality
Product lines, sales, holidays and targets in the embedded model are generated, and amounts are shown only as a percentage of target. No sales, cost, target or organizational data from CJ CheilJedang appears here.