Transforming the 5 core fashion processes
Fashion’s long agony dates back to way before Covid — to be sure, the list of fashion-related bankruptcies is now getting longer every day thus accelerating its previous trend, but the root causes are not new.
Traditional fashion practice is dead. Time is ripe for scientific overhaul of the industry.
After having collected and processed data on over 1 billion people, 10 million products and $100 billion in baskets spanning across many industries in my job at Evo, I still find fashion to be comparatively inefficient (and therefore interesting!).
I show how the 5 core fashion processes are broken, and suggest how science is finally providing new answers to fix them.
For many fashion companies, this would require a complete transformation of their current management practice; just in the past year, I have met 20+ global CEOs and many were utterly shocked at these ideas! Meanwhile, however, to some readers, this may sound very surprising.
Myth: fashion management is just about being cool and high-desire
This is the first image that comes up on Unsplash for the keyword fashion. It really says it all: what words to add to such a poetic image?
Reality: making money in fashion is hard work, exactly like in any other industry (or even more)
Fashion merchandising is the management discipline tasked with consistently having the product in the right place at the right time.
Easier said than done!
People today do not choose from what they can find, but rather search what they want.
And this, my friend, is a completely different game.
Much of the industry is still stuck in the good old days of PUSH, which was in essence the idea of choosing 1 year in advance what to design for the fashion shows, then push it to the sales channels, and that’s it. Off we go, hope it will work.
And work it did, until there was limited competition, limited information, limited travel: a high-margin industry, dominated by huge egos, envy and all that.
Today, however, competition is fierce, information is everywhere, and people can easily travel or anyways find what they want online, as travelling during Covid is temporarily more difficult.
1. Purchasing is broken. Multi-feature similarity is the scientific answer
One of the leading researchers in the field of fashion forecasting, Sebastien Thomassey of ENSAIT, produced over the years a comprehensive set of scientific publications and detailed reviews to help determine optimal purchasing volumes for fashion managers, among other things.
One of the main ideas? Forget about basic practice to determine puchase volumes like average sales by category or comparable product, which are at the core of the current typical fashion merchandising practice, and instead embrace the true richness of attribute-based forecasting.
The main steps can be summarized as follows: find as many tags as possible for future and past products, then use them to determine multi-dimensional similarity, and buy each new product accordingly.
Essentially, exploding the old concept of buying the same quantity as one past comparable product, by decomposing each new product into many features, like price, style, category, characteristics etc. and then automatically using those to drive the future expectation: just like the comparable method, but many-to-many instead of one-to-one.
2. Planning is broken. Flexibility is the scientific answer
When I presented the Masterclass at London Business School on AI for managers, I shared the example of how a manager planned next year: take the Excel summary of last year’s results, add 5% top down, presto, done.
The senior managers attending the Masterclass started nodding quite vigorously. This is how it is commonly done — and not just in fashion, to be true.
Now, if you are looking at stable trends like the population growth of Istanbul, such an approach may very well be an acceptable approximation of reality. But fashion sales?
The fashion industry deals with volatile, capricious consumers who change their mind from one moment to the next faster than me and you can say hi.
Therefore building in flexibility is crucial to survive, and again science can come to the rescue:
- Exactly how much to produce or buy in advance, more cheaply and in larger volume, versus how much to keep as Open To Buy closer to the target sell date, which is more expensive and harder to do at scale, but safer. A purely mathematical decision!
- What volume to plan from vendor X, with a given lead time and unit cost, versus vendor Y, with its different characteristics; again, a purely mathematical decision based on uncertainty and expected profit
- How to break volumes into 2 or more production batches, early and late, or splitting fabric production (which requires long lead-time) from cutting, sewing and finishing (which tend to be rather fast).
Already in 1993, Sport Obermeyer had figured out and published the case of splitting planning in conditions of uncertainty into two subsequent stages, to drastically reduce the economic loss and wastage of long-term planning. Essentially building in flexibility until fresh information reduces the level of uncertainty.
3. Grading is broken. Cluster of one is the scientific answer
Believe it or not, in 2020 companies are still paying for big consulting projects to pigeonhole locations into grades.
Essentially the idea, which may have made sense back in 1970, is that if you have 2,000 stores, then you bucket them into 5–10 groups mainly driven by size, so you can standardize decisions and simplify your life as a manager. After all, it’s easier to make 10 decisions than 2,000! Makes a lot of sense from the manager point of view.
Smaller stores all get the same, smaller portion of product; while larger stores all get the whole collection.
Current practice gets as sophisticated as to differentiate stores in city centers versus shopping mall, or fashionable versus formal… But still using 5 to 10 grades.
Now, I don’t know about you, but I feel that today with the emergence of hyper-personalization getting even just 2 people to agree about anything can often be very hard. And that’s just 2 people.
I mean, then how on earth can 2 stores, let alone 200 or 2,000, each catering to maybe 10,000 people each, have anything in common as much as to be put in the same grade? Different locations, different real estate, different climate… different everything.
This is so obvious that hardly deserves explaining, and yet, fashion CEOs still allocate budget to projects where the wrong question is asked, what grade should each store belong to.
Ask not what grade a store should belong to, but what you can do to make each store unique.
Is it really necessary to simplify complexity, in 2020, with all the computing power and mathematical methods and machine learning?
Why not embrace complexity and deliver simplicity?
Each location only sees its own product assortment and offering at any given time. Make it unique. Science already allows it today. Business sense suggests it. And survival strategy should dictate it.
To add proof — good practice — an example: the 2019 Harvard Business School case study of Miroglio, one of the largest Italian fashion retailers, which showed how exchanging inventory across stores so to differentiate them while re-balancing the overall inventory levels dynamically, improved revenues by 18% in a test-control experiment.
Differentiation, towards the cluster of one.
4. Replenishment is broken. ‘Expect one send one’ is the scientific answer
Underlying basically all the supposedly advanced logistics systems used by fashion companies, the core concept is
Sell one, send one
I sell one piece, you send me one new piece from the warehouse.
I mean, this is so entirely wrong! Not taking into account basic things like how long did it take to sell this one piece or what is the overall sales potential compared to other stores is really inefficient.
This is, by the way, how situations like the one I describe in this other article can occur: https://resources.evopricing.com/intelligence/expensive-data-science-mistakes-when-your-kpi-lie/
Typically 60% or even up to 90% of the entire available purchased volume is allocated to the stores at once, initially at a set date, and then what’s left in the warehouse gets replenished based on such a simple rule as sell one send one. Meaning that only limited volume is available to top up the initial availability in the stores, before the warehouse runs out; even this small buffer is therefore used inefficiently to address local spikes in sales.
And anyways, keep in mind that already from day 1…
If you have 1,000 stores and a collection of 1,000 products in 10 sizes each: it would take 10 million pieces to just have 1 each everywhere. Impossible!
Size-level complexity is what makes fashion such a challenging industry. So it’s impossible to cover everything everywhere, efficiently. But a simple scientific concept comes to the rescue to save the day:
Expect one, send one
Just shifting from the past (sold one) to the future (expect one) reframes the question much more productively. Aim to cover the bare minimum in the stores, and then gradually discover demand and replenish more.
In this way, for example, Evo helped Boggi Milano, sell +4% more with -12% inventory in just 7 weeks, which is super-fast in the world of enterprise data science. Microsoft published this case study including the CEO interview and my own contribution.
5. Assortment is broken. Dynamic modular is the scientific answer
The result of all the above issues with purchasing, planning, grading and replenishment is flat assortments:
everything everywhere at the same time.
Then it’s no surprise that according to the World Economic Forum:
Fashion has a huge waste problem
This is both physical waste as well as pricing waste: excessive markdowns, forced discounts, and over-stocked outlet stores required to get rid of unsold inventory.
Then why not try a different solution?
Preparing small modular blocks of assortment, and letting them find their own distribution dynamically, based on customer demand.
This new fascinating experiment is worth its own coverage, which will come one day (as the details are still confidential right now).
Dynamic assortment management.
The long story short is that we took away products that were slowly dying in the 4 weeks before the start; by moving them to different locations where they had never been sold before, these products started a new, second life.
Not flat assortment, but highly targeted and dynamic.
The truly optimal markdown is the one you are not forced to offer at all, because the product sells at full price.
The new human-machine alliance in fashion
Similar examples could actually be applied more broadly to other industries with only minor variations.
Fashion provides an excellent example of the inefficiency radicated in the idea of management push.
Much like other industries, at the end the big new idea is about outsourcing the execution side of management decisions to customers!
Pull, not push.
Since all data is about the customer, directly or indirectly, machine-learning methods like the ones described above are tasked with extracting the information and making it usable every day.
Human managers will always have to determine strategy, storytelling, and high level planning. At the end of the day, even the smartest SatNav could only drive your car, but never tell you where to go.
Asking the right questions, and figuring out the right destination, is and will always be our job, as humans. But let customers do the rest.
Now, letting the business drive efficiently towards the management-defined goal, already today, should be left to powerful algorithms that can support the fashion merchandiser much more efficiently than Excel ever could.
And that is because the customer, through day-to-day actions and feedback, drives detailed decision-making much more efficiently than any human manager ever could.
Happy scientific management!
PS more Business Science:
About the author
Fabrizio Fantini is the brain behind Evo. His 2009 PhD in Applied Mathematics, proving how simple algorithms can outperform even the most expensive commercial airline pricing software, is the basis for the core scientific research behind our solutions. He holds an MBA from Harvard Business School and has previously worked for 10 years at McKinsey & Company.
He is thrilled to help clients create value and loves creating powerful but simple to use solutions. His ideal software has no user manual but enables users to stand on the shoulders of giants.