A forecast set on Monday is already out of date by Wednesday when a promotion lands, a heatwave hits or a line goes down. Weekly refresh was built for slow data and slow planning, but both have changed. Sub-daily refresh keeps the plan close to what is happening on the floor and in the shops. Teams planning the move can ask iFactory AI how refresh cycles are designed.
From Weekly to Sub-Daily Forecast Refresh for FMCG
Fix the data feeds, layer the models and change the workflow so that several refreshes a day are usable, not noisy. iFactory AI pairs adaptive forecasting with a daily replan.
Why a Weekly Refresh Falls Behind
Demand does not wait for the planning calendar. The week below is illustrative, with one forecast set on Monday and four real events that arrive afterwards. Each day the plan stays unchanged, the gap to reality grows.
Slow signals
Promotion sell-through, weather and retailer stock change by the day. A weekly cycle sees them only after the week is over.
Fast supply shocks
Line faults, late materials and absence change what can be made today, yet the plan still shows the old promise.
Costly workarounds
Planners patch the gap with spreadsheets and phone calls. That effort is a hidden cost of a slow cycle.
Three Cadences, Three Jobs
Moving to sub-daily refresh does not mean replacing the weekly plan. It adds faster layers on top, each with its own owner, horizon and level of change.
| Cadence | Main Inputs | Typical Output | Who Acts |
|---|---|---|---|
| Weekly | Orders, promotions, capacity plan | Volume and stock plan | Planning and supply leaders |
| Daily | Latest sales, stock, line status | Revised production sequence | Production planner |
| Sub-daily | Intraday sales and line events | Alerts and small adjustments | Shift planner and supervisor |
Data Architecture: Budget the Minutes
A refresh is only as fast as its slowest step. Treat each cycle as a latency budget and give every stage a time limit. The example below fits a 60-minute cycle.
Pick feeds that matter
Start with the two or three feeds that move fastest, usually sales, stock and line status. Add others later.
Handle late data
Decide what to do when a feed is late. Use the last good value, flag it, and never block the whole cycle.
Keep one version
Store every forecast with its timestamp. That record is what lets you measure improvement later.
See How Often Your Plan Could Refresh
Book a 30-minute session and iFactory AI will show how feeds, models and replan fit together into a refresh your planners can trust, using realistic FMCG data.
Model Design: Layers, Not One Big Model
Fast refresh works best with layered models. A stable base forecast carries the long pattern, and a short-term layer corrects it with the newest signals. Guardrails keep the result sensible.
| Layer | Purpose | Refresh | Guardrail |
|---|---|---|---|
| Base statistical model | Trend, season and promotion pattern | Weekly or daily | Benchmarked against naive baselines |
| Short-term correction | Learns how today differs from the base | Several times a day | Limit on the size of each change |
| Hierarchy reconciliation | Makes SKU, line and plant numbers add up | Every cycle | Totals must match |
| Exception rules | Handles launches, stock-outs and one-off events | On trigger | Human approval for large moves |
The case for short-term correction has long support. A 2011 Terra Technology benchmark reported that demand sensing, which uses daily point of sale and stock data, cut average weekly forecast error by 40 percent across nine consumer goods makers. The authors were a software vendor and the data is over a decade old, so treat it as evidence of direction, not a promise.
A Worked Day: Four Forecasts, One Promotion
The bars show an illustrative promotion day for one SKU. The weekly plan says 4,000 cases. Demand turns out to be 5,400. Each refresh closes more of the gap.
By early afternoon the plan is nearly right, but only if the plant can still act. That depends on what is frozen, which is the subject of the next section.
Workflow Changes: Freeze, Flex and Approve
A faster forecast is useless if the plant cannot change what it makes. Agree which parts of the plan can move, by how much, and who approves. The zones below are not to scale.
Frozen
Running orders and set-ups stay as they are. Only safety and quality can interrupt.
Flexible
Sequence and quantity can shift within limits, so long as changeover cost stays below the benefit.
Open
The plan can be reshaped freely, including material orders and labour.
Measuring Whether Faster Is Better
More refreshes should improve results, not just activity. Measure forecast value added against simple baselines, and track plan stability alongside accuracy.
| Measure | What It Shows | Good Direction | Watch For |
|---|---|---|---|
| Weighted error | Volume-weighted forecast miss | Falling | Gains only on small SKUs |
| Bias | Consistent over- or under-forecast | Near zero | One-sided misses |
| Value added | Error against a naive baseline | Positive | Steps that make results worse |
| Refresh latency | Time from data to released plan | Within budget | Slow stages creeping up |
| Plan churn | Share of plan changed per cycle | Stable or lower | Constant re-sequencing |
| Service and waste | Fill rate and write-offs | Better service, less waste | Accuracy up, service flat |
Forecast value added compares each step with a simple benchmark, such as next period equals last period, or the same period last year. Use at least two benchmarks. If a step does not beat them, it is adding cost without value.
Common Mistakes When Moving to Sub-Daily
Most failures come from process, not mathematics.
A 90-Day Path From Weekly to Sub-Daily
iFactory AI is built as a turnkey server that racks in, plugs in and goes live in roughly 6 to 12 weeks, so a staged move fits inside a quarter.
What iFactory AI Adds
iFactory AI is smart manufacturing and industrial software. Its Adaptive Forecasting and Daily Replan features support this move.
Which SKUs Deserve a Faster Refresh
Not every product needs hourly attention. Segment the range by volume and by how erratic demand is, then give the fastest cadence only where it pays back.
| Segment | Demand Pattern | Suggested Cadence | Reason |
|---|---|---|---|
| High volume, volatile | Promotions, weather, events | Sub-daily | Large swings and large cost of error |
| High volume, steady | Smooth baseline | Daily | Small moves, so frequent change adds little |
| Low volume, volatile | Sporadic, lumpy orders | Daily with exceptions | Noise dominates, so use ranges and rules |
| Low volume, steady | Rare change | Weekly | Effort is better spent elsewhere |
| New launches | No history | Daily for the first weeks | Learn fast from early sales |
In the 2011 benchmark cited earlier, new product weekly error fell from 65 percent to 44 percent over one year with demand sensing, though early weeks were noisier. That fits the pattern above, since launches need fast learning but also careful limits in the first weeks.
Promotions and Launches: Where Speed Pays Most
Promotions are the clearest case for sub-daily refresh. Uplift is hard to predict, and the first hours of sell-through reveal how far above or below plan the event will run.
Set a range, not a point
Use past events to set a low, expected and high case. Pre-agree the actions tied to each, such as extra shifts or material calls.
Read the first signals
Compare early sales with the expected curve each refresh. A steady gap points to a new case, and the plan moves to match.
Stock cover
Track hours of cover at the depot and key accounts. A fast fall triggers an earlier production call.
Material readiness
Check packaging and ingredients against the high case ahead of time. A missing component stops an uplift plan.
Post-event review
Store the actual curve so the next promotion starts with a better forecast range.
Connecting the Systems Around the Forecast
A forecast has to meet real systems to matter. Plan the links in both directions: data in from the shop and the floor, and plans out to the people who act.
Start with read-only feeds into the forecast, then add write-back to planning screens once the numbers have earned trust. Keeping the first phase read-only lowers the risk and speeds approval from IT.
A Day in the New Rhythm
A sub-daily cycle gives the day a clear beat. Each refresh has a purpose and a short list of decisions, so the team is never overwhelmed.
Does Faster Refresh Pay Back?
Build the case from the cost of forecast error. The receipt below uses illustrative figures for one plant and one product family, so replace each value with your own.
Subtract the cost of running more refreshes, including data feeds, compute and planner time. If the net is small, narrow the scope to the segments where error is most costly, such as promoted and volatile SKUs.
Stock-outs
Lost sales and fines for missed orders fall when supply follows demand more closely.
Waste
Short shelf-life products spoil when overproduced. Better timing cuts write-offs.
Expedite costs
Fewer emergency changeovers, rush freight and overtime when the plan moves early.
Questions to Settle Before You Speed Up
Answer these with planning, production and IT in one room. Each gap you find now is a delay you avoid later.
Plain-Language Glossary
Shared words keep planners and production aligned.
Frequently Asked Questions
How often should an FMCG forecast refresh?
It depends on the SKU and the signal. Fast movers and promotions may justify several refreshes a day, while stable lines can stay daily or weekly.
Does faster refresh always improve accuracy?
No. Gains depend on data quality and how well the plant can act. Measure forecast value added against simple baselines before expanding.
What data do we need first?
Sales or orders, stock and line status, timed to the hour where possible. Ask support which feeds matter most for your product mix.
How do we avoid constant re-planning?
Use frozen windows, change limits and exception-based review. Weigh each move against changeover cost.
Is the weekly plan still needed?
Yes. It anchors volume, capacity and stock policy, and the faster layers adjust around it within agreed limits.
Keep the Plan in Step With Demand
iFactory AI pairs adaptive forecasting with a daily replan, so your plan follows demand instead of trailing it. Book a walkthrough on your own planning data.
