Trend forecasting: from signal to decision

A trend report tells you what might happen. The hard part comes after: deciding how much to bet on it, and when.

What is trend forecasting?

Trend forecasting is the practice of estimating which shifts in behaviour, taste or technology will grow. Johnson & Wales University defines the fashion version as predicting upcoming consumer preferences. It uses historical data, consumer insights and other factors. The same logic applies well beyond clothing.

It works at two horizons. JWU describes short-term forecasting as specific details such as colours and fabrics, within the next few seasons. Long-term forecasting covers broader shifts in technology and values, sometimes several years ahead.

A forecast of this kind is rarely a number. It is usually a direction, a reason and a rough timing. That makes it useful for planning a range or a brief, and weaker for setting an exact order quantity.

Older than the internet

Ballmer and Tobias trace forecasting to colour charts, trade catalogues and international communications. After the war it became institutionalised, and it helped consolidate a consumer-oriented supply chain.

They write that forecasting reports are costly and come from a handful of companies. These reports influence fields from womenswear to computer games.

What is the difference between trend forecasting and demand forecasting?

Trend forecasting asks what will change, and demand forecasting asks how much will sell. They feed each other. A trend call can warn a planner that the past may stop describing the future. A sales-based model relies on that assumption.

Point	Trend forecasting	Demand forecasting
Question	What is shifting, and why?	How many units or how much value?
Output	Direction, reason and rough timing	A number per product, place and period
Main inputs	Culture, search, social, retail and experts	Sales history, prices and promotions
Typical user	Design, product, brand and strategy	Supply chain, finance and sales
Main risk	A persuasive story without evidence	A precise number that misses a turn

For the second discipline, read our guide to demand forecasting. For the signals both can draw on, see how to measure market demand.

How does fashion trend forecasting work?

Fashion trend forecasting combines observation, consumer analysis and data, then adds expert judgment. JWU lists four steps: research runways and street style, analyse consumer behaviour, use data tools for sales, search and sentiment, and track global influences.

Online services changed this field from the late 1990s. Before one such service started in 1998, businesses relied on printed trend books issued each season. It supplied catwalk, street style and retail insight much faster.

In the company's own words

One large forecasting service says it works in three layers: macro trends, consumer behaviour and product forecasts. It says the macro layer draws on about 2,800 third-party sources.

Its product forecasts blend five feeds, which it calls social, sentiment, search, shelf and shows. It says data scientists evaluate the algorithms and analysts review the output.

These are the company's own descriptions, not independent findings.

The lesson travels well. Blend many sources, then apply judgment. A pure data feed misses meaning, and a pure opinion misses scale.

Is it a trend or just the season?

Check the calendar first, because a seasonal rise can look like a trend. Forecasting: Principles and Practice draws a clear line. A trend is a long-term increase or decrease. A seasonal pattern has a fixed, known period tied to the calendar.

Compare like with like. Set this period against the same period last year, and the year before.
Read a 12-month average. It removes the yearly pattern, so a lasting rise stays visible.
Decompose the series. STL splits a series into trend-cycle, seasonal and remainder, and lets the seasonal part change over time.
Check the timing. A season that starts earlier or peaks higher is also a finding, though a different one.

Ignoring this has caught large projects. Lazer and colleagues wrote that the first version of Google Flu Trends was "part flu detector, part winter detector". Terms unrelated to flu matched it, because they also rose in winter.

Three patterns, not two

The same textbook adds a third pattern. A cycle rises and falls without a fixed frequency, usually with economic conditions. It typically lasts at least two years.

It warns that many people confuse cycles with seasons. A two-year rise may therefore be a cycle, and not a new long-term trend. Longer history is the only real check.

How do you know if a signal is real?

A signal is worth acting on once it survives the calendar, a second source and a plain explanation. None of these checks proves the future. Together they remove most of the stories that only look like trends.

Persistence. Has it lasted beyond one spike, after the season is removed?
Breadth. Does it appear across countries, retailers or related terms, or in one place only?
Second source. Does a different type of data agree, such as retail listings or consumer research?
Mechanism. Can you explain why people would keep doing this?
Base rate. How often have similar signals in your category lasted?

Locate the signal as well. A shift at Attribute level differs from one at Category level. The five levels below use illustrative examples.

Level	Illustrative example
Market	Small kitchen appliances, Netherlands
Category	Air fryers
Type	Double air fryer
Specification	8 litre capacity
Attribute	Dishwasher safe

The default should be caution. Armstrong, Green and Graefe call this the Golden Rule of Forecasting: be conservative. A conservative forecast stays consistent with cumulative knowledge about the present and the past.

Read the rule as a default

The authors reviewed 105 papers with experimental comparisons, and 102 supported their guidelines. Ignoring a single guideline raised forecast error by more than two-fifths on average.

The paper appeared in the Journal of Business Research in 2015. The rule is the authors' own synthesis, so treat it as a strong default. A real break from the past can still justify a bolder call.

Watch your own judgment as well. Forecasting: Principles and Practice warns that memory overweights recent events and that optimism distorts forecasts. Those are the conditions in which a trend deck gets approved.

What the research supports

Choi and Varian framed their Google Trends work as an attempt to predict the present, because official data arrives late.

They suggest that turning points in search may be useful, but treat a jump to future sales with care. Any longer lead needs its own evidence in your own category.

How do you act on a trend signal?

Match the size of the move to the strength of the evidence, and keep early moves cheap to reverse. Many trend signals need a small test more than a large plan. The table links evidence to action.

Evidence	Reasonable action	What would change your mind
One source, one spike	Add it to a watchlist	It fades, or it was seasonal
Persistent in one source	Run research and small tests of the message	No second source confirms it
Two source types agree	Pilot a product, line or campaign	Trial buyers do not return
Broad, explained and lasting	Commit range, budget and capacity	Your leading signals turn down

Write the call down. Note what you expect, by when, and which signal it rests on.
Name one owner and one decision. A trend without an owner stays a slide.
Set the exit in advance. Agree what would stop the project before enthusiasm builds.
Score the call later. Compare it with what really happened, and keep the record.

That last step is where most teams stop short. The same textbook advises keeping records of forecasts and comparing them with outcomes. It also advises keeping forecasting apart from target-setting, so nobody forecasts what they have been told to hit.

One input on demand

Trendata publishes this guide, and it is not a trendwatcher. It shows how demand and customer needs develop, using aggregated orientation moments.

These are moments when people look into products, needs and brands, explore alternatives and prepare choices. They form in Search, Social and AI environments, and they are not purchases.

Its demand forecasts run up to 12 months ahead, at category, type and specification level. They describe expected demand, not revenue or stock per SKU. Ask any forecaster, including Trendata, how past forecasts were scored.

A forecast you never score is an opinion with a long shelf life.

To apply this to the range, read about product trends.

Frequently asked questions

What does a trend forecasting service deliver?

Usually reports on macro shifts, consumer behaviour and product direction. Treat each as an input to your own decision, and ask how the service reaches its conclusions.

What is the difference between a trend and a season?

A season repeats on a fixed calendar period. A trend is a long-term rise or fall. Compare the same period across several years to tell them apart.

How far ahead can trend forecasting look?

Short-term work covers the next few seasons, with details such as colours and materials. Long-term work looks at broader shifts, sometimes several years ahead, with much less detail.

How do I become a trend forecaster?

No single route exists. The work combines observing culture, reading data and writing a clear case. Build those skills, keep a dated record of your calls, and score them.

Is a rising line in Google Trends a trend?

Not on its own. Check the same period in earlier years, then look for a second source. Trends shows relative interest, not sales.

How do you tell a trend from a fad?

No test is certain in advance. Look for persistence after removing the season, breadth across markets, and a clear reason why the behaviour should last.

Sources

See where demand is heading in your market.

Book a call