September is a dangerous month for a garden-centre buyer.

Not because anything particularly dramatic happens.

Because everything that happened is still easy to remember.

The spring that started badly.

The rose that disappeared in two weeks.

The perennial collection that felt slow.

The one customer who complained three times.

The product that sold out immediately.

The rack that sat too long.

The wet May.

The hot June.

The shipment that arrived late.

All of it is still sitting there, fresh in the mind.

And now comes a very consequential question:

What should we order for next year?

This is where behavioural science should make every retailer slightly nervous.

Because human beings are not particularly good at remembering a year proportionately.

We remember pieces of it.

Usually the vivid pieces.

The recent pieces.

The annoying pieces.

The surprising pieces.

Then we construct a story.

And sometimes we order against the story.

The Last Thing That Happened Feels Important

Researchers studying judgmental forecasting have repeatedly found that people tend to put too much weight on recent observations when trying to predict what happens next.

This is sometimes called recency bias.

In forecasting experiments, people have been shown to overweight the latest data point, especially when trying to extrapolate a time series. Researchers have specifically recommended making longer-term trends explicit to help reduce that tendency.

That should sound familiar to anyone who has ever sat down to write a seasonal plant order.

Suppose a particular program sold:

2023: very well

2024: very well

2025: very well

2026: slower

What does the brain notice?

It is closest.

Most emotionally available.

And perhaps most relevant.

Maybe demand really is declining.

But perhaps 2026 was:

colder

later

poorly positioned

understocked at the beginning

overstocked at the end

delivered at the wrong moment

competing with a promotion

affected by staffing

affected by construction

affected by something peculiar to that season

One year is information.

It is not necessarily a trend.

A Bad Spring Can Become a Bad Forecast

This is particularly dangerous in horticulture because we operate in such a noisy retail environment.

Weather changes.

Delivery dates change.

Mother’s Day moves relative to the weather.

Easter moves.

Competitors change promotions.

Stores move displays.

Plants flower at slightly different moments.

Labour availability changes.

A garden centre is not a laboratory.

It is a living commercial system.

So when a product performs differently from one year to the next, the most tempting explanation is often:

Consumers didn’t want it.

Sometimes that is exactly what happened.

Sometimes it isn’t.

Sales and Demand Are Not the Same Thing

This distinction is crucial.

Imagine a garden centre receives 30 units of a new perennial.

It sells all 30.

How much demand existed?

Thirty?

We don’t know.

It could have been 31.

It could have been 60.

It could have been 100.

Once the last unit sold, demand became invisible.

Operations researchers call this censored demand.

A retailer can observe sales only while inventory exists. Once the product stocks out, additional demand often disappears from the data. Research on retail forecasting has repeatedly shown that using sales without accounting for stockouts can systematically underestimate true demand.

That matters enormously when writing next year’s order.

Because one of the most dangerous conclusions in retail is:

We sold 30 last year, so let’s order 30 again.

Perhaps you sold 30 because you only had 30.

The Best Seller May Have Been Even Better Than It Looked

This happens constantly in seasonal categories.

A rose sells out quickly.

The sales report says:

42 units.

Another rose lasts most of the season and sells:

Which was stronger?

The spreadsheet says 57.

The customer behaviour might say something very different.

The first rose may have generated enough demand to sell 80 units, but inventory disappeared before demand had the chance to express itself.

The second may simply have remained available longer.

Observed sales are partly a measure of consumer demand.

They are also a measure of how much merchandise you made available to be bought.

Retail researchers have spent decades wrestling with this exact problem. When goods stock out, observed sales become a distorted measure of underlying demand, and future forecasts based purely on those sales can create a downward spiral of under-ordering.

Slow Sales Have the Opposite Problem

Now take a product that looked disappointing.

You received 100 units.

Sold 55.

Easy conclusion:

Bad product.

Maybe.

But before cutting next year’s order, ask what actually happened.

When did it arrive?

Where was it displayed?

What condition was it in during peak traffic?

Was it blooming at the right moment?

Did the tag communicate anything useful?

Was it buried behind stronger merchandise?

Was the price correct?

Did staff understand it?

Did shoppers have time to discover it?

Again, I am not horticulturist enough to tell an experienced garden-centre manager whether a particular perennial deserved another year.

But I have spent enough time around retail to know that sales outcomes contain more than product preference.

Execution is mixed into the number.

The Spreadsheet Does Not Explain Itself

This is one of the traps of modern retail data.

Numbers look objective.

$18,742.

63 units.

71 percent sell-through.

14 percent down.

Numbers feel like facts because they are facts.

The interpretation is where things become slippery.

A 71 percent sell-through can mean:

terrific result given late arrival

poor result given promotion

strong result given bad weather

weak result compared with nearby stores

excellent result at full margin

terrible result after markdown

The number describes what happened.

It does not automatically explain why.

Forecasting requires explanation.

Human Judgment Helps

This is where I should be careful not to swing too far in the other direction.

The answer is not:

Ignore experience and let an algorithm order the plants.

Human judgment is extraordinarily valuable.

Experienced garden-centre managers know things no spreadsheet knows.

They remember:

the road construction

the frost

the staffing problem

the wrong rack location

the customer requests

the new subdivision nearby

the competing promotion

the product that suddenly became fashionable

That contextual knowledge can improve a forecast.

Research supports this.

Human adjustments to statistical forecasts can add value when forecasters possess genuine information that is not captured in the underlying model.

But there is a catch.

We Are Also Very Good at Adjusting for the Wrong Things

A major 2025 analysis by Robert Fildes, Paul Goodwin and Shari De Baets looked across roughly 147,000 forecasts from six datasets.

They studied what happened when people manually adjusted system-generated forecasts.

The result was sobering.

Judgmental adjustments improved accuracy and bias for only a little more than half of the SKUs, with substantial variation across the datasets. The researchers found relatively weak evidence that forecasters consistently used genuinely valuable information unavailable to the system. People also appeared to react to cues that were irrelevant or less diagnostically useful.

That is an important retail lesson.

Experience is valuable.

Memory is valuable.

Judgment is valuable.

But none is infallible.

The Product Everyone Talked About

Suppose one strange new plant generates enormous conversation.

Staff talk about it.

Customers ask about it.

Someone complains.

Someone loves it.

Everyone remembers it.

It sells 40 units.

Beside it sits an ordinary perennial.

Nobody talks about it.

Nobody complains.

Nobody posts photographs.

It sells 135 units.

Which one occupies more space in the manager’s memory in September?

Very possibly the first.

The unusual becomes mentally available.

Psychologists have long described this broader phenomenon as the availability heuristic.

Events that come easily to mind can feel more frequent or important than they actually were.

A quiet seller creates revenue.

A dramatic seller creates a story.

The human mind loves stories.

The purchase order should still respect the revenue.

One Customer Can Become Fifty

The same distortion can occur with customer feedback.

One very articulate customer tells you repeatedly:

Nobody wants this.

Meanwhile, 70 other people quietly buy it.

The first customer leaves an impression.

The others leave transactions.

This does not mean ignoring customer feedback.

It means weighting it correctly.

Customer comments are qualitative evidence.

Sales are behavioural evidence.

Returns are behavioural evidence.

Markdowns are behavioural evidence.

Reorders are behavioural evidence.

Stockouts are behavioural evidence.

The strongest decision uses all of them.

Beware of the Year That Was Weird

Every garden centre has one.

The strange year.

The season that started three weeks late.

The spring where every weekend seemed rainy.

The year one product arrived after the selling window.

The year inventory got trapped somewhere in the supply chain.

The temptation is to adjust the entire future around that experience.

Forecasting research warns against exactly this kind of behaviour.

In experiments involving demand projections, forecasters have been shown to treat random fluctuations as though they represented systematic change, especially when the most recent observation looked unusually high or low.

In horticulture, the danger may be even greater because so many genuine external shocks actually do occur.

The trick is distinguishing:

noise

from

change.

Ask Whether the Cause Will Repeat

This may be the single most useful ordering question.

Suppose sales were poor.

Why?

Then ask:

Is that reason likely to exist again next season?

Late shipment?

Probably not.

Temporary staff shortage?

Probably not.

Record-setting cold spring?

Maybe not.

Permanent road relocation reducing store traffic?

Possibly.

Competitor opened nearby?

Yes.

Consumer trend declining for several years?

Possibly.

A product repeatedly underperforming across several stores?

More concerning.

The forecast should follow the persistent cause, not merely the result.

September Is When Stories Become Inventory

This is why ordering is such an interesting act.

A retailer takes:

memory

data

experience

intuition

expectations

risk tolerance

and converts all of it into physical inventory that will arrive months later.

That is quite extraordinary.

The order is a forecast turned into plants.

If the forecast is too optimistic:

markdowns

shrink

labour

crowding

capital trapped in inventory

If too pessimistic:

stockouts

lost sales

thin displays

missed opportunities

customers buying somewhere else

There is no perfect answer.

Ordering is inherently a balancing act.

Canadian Horticulture Is Not Standing Still

It is worth placing individual store decisions against the broader Canadian market.

Statistics Canada reported that greenhouse flower and plant sales and resales reached $2.4 billion in 2025, rising 7 percent from the previous year. Nursery sales and resales reached approximately $852.7 million, up 1.8 percent. Total greenhouse, nursery and sod industry sales rose for the fifth consecutive year.

Potted-plant production also increased.

Canadian growers produced approximately 269.1 million potted plants in 2025, up from about 259 million the year before and 241 million in 2021. Potted-plant sales reached approximately $1.46 billion.

That does not tell any individual Canadian Tire dealer how many clematis to order.

But it does provide useful context.

A weak result in one category or one store should not automatically be interpreted as evidence that Canadian consumers have suddenly stopped buying plants.

The wider market continues moving.

Look at More Than One Year

Suppose you are reviewing a category for your 2027 order.

Do not begin with:

What happened this year?

Begin with:

What has happened over several years?

Three years is better than one.

Five can be better still where comparable data exist.

Look for:

trend

volatility

repeated winners

repeated laggards

stockouts

markdown dependence

changes in retail price

delivery timing

seasonality

A single year can be an anecdote.

Several years begin to form a pattern.

But History Is Not Destiny Either

There is another trap.

A product sold well for five years.

Therefore:

Order it again.

That can become lazy forecasting too.

Markets change.

New genetics arrive.

Customer tastes change.

Housing changes.

Lot sizes change.

Demographics change.

Competing products improve.

A good historical record earns attention.

It does not earn immortality.

Past performance should be the baseline.

Then ask what has genuinely changed.

Separate the Baseline From the Bet

I like thinking about an order in two parts.

The baseline

and

the bet.

The baseline represents merchandise with enough history to provide reasonable expectations.

The bet represents:

new products

expanded quantities

new collections

emerging trends

products you believe deserve greater exposure

Every good garden centre needs both.

A store built only around baseline history gradually becomes stale.

A store built entirely around bets becomes a casino.

The art lies in knowing which part of the order is which.

New Products Have No History

This is where buyer judgment becomes especially important.

A genuinely new plant has no store-level sales history.

Nothing to extrapolate.

So how should it be evaluated?

Perhaps against comparable products.

Similar use.

Similar price.

Similar size.

Similar consumer benefit.

Similar colour.

Similar maturity.

Forecasting researchers call versions of this analogical forecasting.

Instead of pretending we know the future of something entirely new, we ask:

What previously behaved somewhat like this?

That is far more disciplined than:

I think customers will love it.

Though, admittedly, retail would be rather dull if we eliminated that sentence entirely.

Give Newness Enough Oxygen

There is also an uncomfortable truth about innovation.

A retailer can prove a new product doesn’t sell simply by ordering so little that nobody sees it properly.

Six units scattered into an enormous department do not constitute a particularly fair market test.

Neither does one rack buried at the back.

New products often require enough presence to become commercially legible.

This does not mean making reckless orders.

It means recognizing that inventory quantity can influence the quality of the test.

A product needs enough representation for consumers to encounter it.

Your Best Store May Be a Better Forecaster Than Your Average

Multi-store retailers have another advantage.

Stores can learn from one another.

If a product consistently performs strongly in several high-performing garden centres but weakly in one, the question becomes:

What is different about the product?

Or:

What is different about the store?

Those are not the same diagnosis.

Sometimes top-performing stores are revealing demand that other stores have not learned how to capture.

That makes peer comparison extremely valuable.

Instead of asking only:

How did we do?

Ask:

How did stores like ours do?

That begins separating market demand from local execution.

A Sellout Is Not Always a Victory

Here is another behavioural trap.

Managers love sellouts.

Sold every one.

Fantastic.

Sometimes.

If you sold the last perennial on May 25 and customers would have continued buying until June 20, the sellout may represent a forecasting failure.

The store did not maximize sell-through.

It exhausted supply.

There is a difference.

A perfect 100 percent sell-through rate can sometimes mean:

We ordered exactly right.

Or:

We ordered far too little.

Without knowing when the product disappeared and what demand remained afterward, the percentage alone cannot tell you.

Ninety Percent May Be Better Than One Hundred

Imagine two scenarios.

Store A

Orders 50 units.

Sells 50.

Revenue opportunity ends early.

Store B

Orders 80.

Sells 72 at full price.

Eight remain.

Which performed better?

The percentage metric favours Store A.

The absolute economics may favour Store B.

This is why ordering metrics should reflect what the business is actually trying to accomplish.

Not simply perfect sell-through.

But profitable sales.

Margin.

Availability.

Customer service.

Inventory efficiency.

Those objectives can pull in different directions.

The Fear of Leftovers Can Create Empty Benches

Ordering decisions also contain asymmetric psychology.

The pain of staring at unsold inventory in June is extremely visible.

You see it every morning.

Markdowns hurt.

Shrink hurts.

The mistake feels tangible.

The sales you never made because inventory disappeared early are invisible.

Nobody can point at the empty space and say:

There are the 37 hydrangeas we could have sold.

That asymmetry can make under-ordering psychologically comfortable.

The failure disappears.

Over-ordering leaves evidence.

Under-ordering often leaves none.

That is precisely why censored-demand research matters.

Lost demand is easy to underestimate because it literally vanishes from the dataset.

Ordering Should Be a Conversation With Evidence

For each important category, perhaps ask:

What did we sell?

What did we order?

When did we sell out?

When did sales slow?

What was the full-price sell-through?

How did comparable stores perform?

What happened over the last three years?

Were there unusual external factors?

Did the product receive a fair presentation?

What customer feedback was repeated rather than isolated?

What is genuinely different about next year?

Then make the judgment.

That is not eliminating intuition.

It is disciplining it.

Write Down Why You Changed the Number

Here is one practice I think is particularly valuable.

Whenever you materially increase or decrease an order, write down why.

Not:

Order 20 fewer.

But:

Reduce from 80 to 60 because sales have declined three consecutive seasons at stable pricing and full availability.

Or:

Increase from 50 to 70 because stockout occurred before peak selling period in each of the last two seasons.

Or:

Maintain despite weak 2026 sales because shipment arrived three weeks late.

Now something interesting happens next year.

You can audit the reasoning.

Were we right?

The organization begins learning not only from sales outcomes, but from its own forecasting decisions.

Memory Is Easier to Challenge When It Is Written Down

Without a record, next September’s conversation becomes:

“I remember this being weak.”

Someone else:

“I thought it did pretty well.”

A third person:

“Wasn’t that the year it arrived late?”

Now we are forecasting from archaeology.

Written assumptions allow learning.

This is especially important in seasonal businesses because twelve months is a long time.

The people may change.

The circumstances get blurry.

The rationale disappears.

A good order therefore contains two things:

quantity

and reasoning.

Do Not Let One Bad Year Murder a Good Product

This may be the most human version of the whole argument.

A plant has been successful for years.

Then it has a bad season.

We react.

Cut it.

Remove it.

Reduce inventory dramatically.

Sometimes that is correct.

But behavioural science tells us to be suspicious when the newest information suddenly overwhelms everything that came before it.

Ask:

Was this a bad product?

Or merely a bad year?

Those are very different things.

Do Not Let One Great Year Crown a King Either

The same principle runs upward.

A product explodes.

Sales double.

Everyone gets excited.

Increase the order dramatically.

But why did it happen?

Promotion?

Exceptional flowering?

Social media?

Perfect weather?

Competitor out of stock?

A temporary fad?

Again:

Will the cause repeat?

Growth deserves investment when the underlying demand is persistent.

One spectacular year can be just as misleading as one terrible year.

The 2027 Customer Has Not Shopped Yet

This is the strange thing about the order being written in September.

You are buying for a person who does not yet exist in the store.

They will arrive next spring.

Different weather.

Different mood.

Different garden project.

Perhaps different economic circumstances.

Your job is to predict them.

Nobody will do that perfectly.

The objective is not perfect foresight.

It is better judgment.

Perhaps Humility Is a Forecasting Tool

I am still learning the horticultural side of this business, and I would be foolish to pretend that a spreadsheet could replace the instinct of a garden-centre manager who has watched customers shop plants for twenty years.

But perhaps the reverse is equally true.

Twenty years of experience should not make anyone immune to data.

The strongest decisions may come from a productive disagreement between:

experience

and

evidence.

When both point in the same direction, act confidently.

When they disagree, investigate.

That tension is useful.

The Order Is a Hypothesis

Maybe that is the cleanest way to think about it.

Every purchase order is a hypothesis about next spring.

We think customers will want:

this much

of these products

at these prices

during these weeks.

Then the season tests the hypothesis.

Some predictions succeed.

Some fail.

The goal is to learn.

Not simply to remember.

Before You Send the Order

This September, before finalizing an important category, try one last exercise.

Look at the proposed order.

Then ask:

What part of this number comes from evidence?

What part comes from something I remember vividly?

What part reflects a genuine change in consumer demand?

What part reflects something unusual about last season?

Where did we stock out?

Where are sales disguising unmet demand?

Which assumptions are we making about next year?

Then write those assumptions down.

Because next September, the most valuable thing may not be remembering what you ordered.

It may be remembering why.

Last Season Is Evidence, Not Destiny

The garden centre has to learn from the past.

Of course it does.

But learning from the past and repeating the past are not the same thing.

One year can mislead.

One sellout can mislead.

One complaint can mislead.

One spectacular success can mislead.

The buyer’s job is to look underneath.

Find the pattern.

Find the exception.

Find the stockout.

Find the execution problem.

Find the genuine change.

Then place the bet.

Because the 2027 order should not be written by the loudest memory of 2026.

It should be written by the best explanation of what actually happened.

And by the clearest thinking we can bring to what happens next.

Sources

Accuracy of Judgmental Extrapolation of Time Series Data: Characteristics, Causes, and Remediation Strategies for Forecasting

Eric Welch, Stuart Bretschneider and John Rohrbaugh, 1998.

When Providing Optimistic and Pessimistic Scenarios Can Be Detrimental to Judgmental Demand Forecasts and Production Decisions

Paul Goodwin and colleagues, 2019.

Measuring Forecasting Accuracy: The Case of Judgmental Adjustments to SKU-Level Demand Forecasts

Andrey Davydenko and Robert Fildes, 2013.

Use of Contextual and Model-Based Information in Adjusting Promotional Forecasts

Demand forecasting researchers, 2023.

Demand Forecasting Under Lost Sales Stock Policies

Juan R. Trapero, Enrique Holgado de Frutos and Diego J. Pedregal, 2024.

Forecast Value Added in Demand Planning

Robert Fildes, Paul Goodwin and Shari De Baets, 2025.

Greenhouse, Sod and Nursery Industries, 2025

Statistics Canada, 2026.

Production and Sale of Greenhouse Flowers and Plants

Statistics Canada, 2026.

Nursery Stock Sales and Resales

Statistics Canada, 2026.