Forecasting
Forecast demand as a range, ready for Ramadan, 11.11 and 12.12.
What this course covers
For data analysts and data scientists who forecast sales, demand or traffic, and for the teams who order stock from those forecasts. A forecast as a range, horizon and granularity, and how a forecast differs from a target. Then weekly and payday patterns, a Ramadan and Lebaran that move every year, double-date campaigns, methods from the naive forecast to exponential smoothing, and how to measure accuracy and bias. The last unit takes it all into Python, with an honest backtest.
New to AI? Start with AI & data fundamentals.
Part of the career paths:Data scientist
What you will be able to do
- Build a forecast as a range, with the horizon and granularity its decision needs
- Separate trend, seasonality and events such as Ramadan and 11.11 in sales history
- Measure forecast accuracy and bias, and compare it with the naive forecast
- Build and backtest a forecast in Python
Units and bites
- Unit 14 bitesAbout 12 min
Forecasting basics
What a forecast is, why it is always wrong, and why that's fine.
- 3 min
A forecast is a range, not a number
Why "we will sell 500 next week" is less useful than "between 420 and 580".
- 3 min
Horizon and granularity
Tomorrow's forecast per SKU is a different job from next quarter's per category, and you need both.
- 3 min
What a forecast is for
Buying, stock, store staffing and promo budgets each need a different forecast, with a different cost of being wrong.
- 3 min
A forecast isn't a target
A forecast guesses what will happen; a target is what you want. Mixing them ruins both.
- Unit 25 bitesAbout 15 min
Seasonality and events
Ramadan, Lebaran, 11.11, 12.12, payday and school holidays: patterns you can plan for.
- 3 min
Weekly and payday patterns
Why weekends and payday often move daily sales.
- 3 min
Ramadan and Lebaran
Indonesia's biggest season moves about 11 days earlier every year, so the forecast has to move with it.
- 3 min
Double-date campaigns
9.9, 11.11 and 12.12 (Harbolnas): big spikes that often borrow sales from the weeks after.
- 3 min
Holidays and the school calendar
Christmas, Lunar New Year, the new school year and long weekends as markers on the forecast calendar.
- 3 min
Building an events calendar
One table of promo and event dates that makes every later forecast better.
- Unit 35 bitesAbout 15 min
Simple forecasting methods
Methods a spreadsheet can do, and when an AI tool is worth it.
- 3 min
The naive forecast: your benchmark
"Same as last week" or "same as last year": the method every other one has to beat.
- 3 min
Moving averages
Smoothing out noise with the average of the last few weeks, and the price you pay for it.
- 3 min
Seasonal forecasts
Combining the recent trend with last year's seasonal pattern.
- 3 min
Exponential smoothing, without the maths
Giving recent weeks more weight: the idea behind many forecasting tools.
- 3 min
When to use AI forecasting
Machine learning helps with many SKUs and many drivers, not as a substitute for clean data.
- Unit 44 bitesAbout 12 min
Measuring forecast accuracy
Error, bias and MAPE: how to tell whether your forecast is getting better.
- 3 min
Forecast error
The gap between forecast and sales, and why you average it without the plus and minus signs.
- 3 min
MAPE and its traps
The popular percentage error, and why it misleads for slow-moving products.
- 3 min
Bias: always too high or too low
An error that leans one way costs more than a random one.
- 3 min
Is your forecast adding value?
Comparing every method and every manual adjustment with the naive forecast.
- Unit 54 bitesAbout 12 min
Forecasting with Python
From spreadsheet to code: preparing a time series, ETS and Prophet-style models, and an honest backtest.
- 3 min
Time series in pandas
Resampling daily to weekly, filling missing dates, and telling a zero from an out-of-stock.
- 3 min
ETS models with statsmodels
Level, trend and season: the same exponential smoothing, now with prediction intervals.
- 3 min
Prophet-style models
Trend, seasonality and an events calendar as regressors, including a Ramadan that moves every year.
- 3 min
Backtesting: testing on the past
Rolling origin, comparing against the naive forecast, and why one test period is not enough.
Questions about this course
Learning
How long does Forecasting take?
About 66 minutes in all: 5 units made of 22 bites of about three minutes each, and every bite ends with a quick check. You can do one bite a day or several in a row.
Do I need to know how to code?
Not to start. The foundations explain every idea in plain words first; where a course uses SQL or Python, it teaches it from the beginning, with a playground in the lesson.
Where should I start?
If you are new to AI, begin with AI & data fundamentals. If you know the job you want, follow its career path; Forecasting has a place in it.
When can I start learning?
Datamew is coming to iOS and Android soon. There is no release date or price announced yet: join the waitlist and we will tell you when the app is out.
Is it available in Bahasa Indonesia?
Yes. Datamew is written in English and Bahasa Indonesia.
More courses
- AI & data fundamentalsHow AI works, how to instruct it, and how to read numbers honestly.
- Data analystHow analysts think and work: from a business question to an answer people use.
- Data engineerBuilding pipelines that deliver data on time and correct: SQL, Python, orchestration and warehouses.
- Data scientistWhat data scientists do, and how to use LLMs in the work.
- Data preparationFrom a raw export to a table you can trust: tidy shape, missing values, duplicates and formats.
- Data visualisationCharts that answer a question honestly, from picking the type to dashboards and stories.
- SQL for analysisAsking the database directly: filtering, summarising, joining tables and window functions.
- StatisticsThe statistics every data role uses: summarising, reading distributions and measuring uncertainty.
- Supervised and unsupervised learningThe two ways machines learn from data: with answers, and without them.
- Product recommendationsHow "you might also like" works, the data behind it, and how to prove it pays.
- A/B testingMeasuring whether a change really worked, with two random groups and an honestly read result.
- Customer segmentationSplitting customers into groups that differ, then treating each group differently.
- Social media analyticsRead Instagram, TikTok and YouTube numbers honestly, and use AI without being fooled.
- AI for finance reconciliationMatching bank statements to invoices: rules first, AI for the rest, people decide.
- AI for SEOHow search works in the age of AI answers, keyword research, Search Console, and writing with AI responsibly.
Read while you wait
Guides from the Datamew team, free on the web today.
How to Become a Data Analyst in Indonesia: Skills, Roadmap and Salary (2026)
To become a data analyst, learn spreadsheets, SQL, one BI tool such as Looker Studio or Power BI, basic statistics, and how to explain a finding in plain words. Most people can learn the basics in about six months part-time, then need two or three portfolio projects to get interviews. In Indonesia the average data analyst salary on Indeed is about Rp6.3 million a month.
8 min readHow to Become a Data Scientist in Indonesia: Skills, Roadmap and Salary (2026)
A data scientist uses Python, statistics and machine learning to predict what will happen and to test what works. Most people get there through data analysis first: learn SQL and statistics, then Python, experiments and machine learning, which takes roughly a year part-time. In Indonesia, data scientist salaries on Indeed average about Rp14.5 million a month.
8 min readBe first on the app
Leave your email and we will tell you the day Datamew lands on your phone. One email at launch, no spam.
- Coming soon onApp Store
- Coming soon onGoogle Play
- Use it now onthe web