fashionvstech - Tumblr blog

[fashionvstech]

AI in your industry (2/6): recommendation engines

How come, in face of so many things to consumer, so many SKUs to buy, you didn’t succumb yet to choice paralysis? One word: hyperpersonalization. Consumers have been able to survive in a world of infinite, chaotic supply, thanks to the bespoke choices brought to them through recommendation engines...

Netflix does it. Amazon does it. Your e-commerce website does it too. And if it doesn’t, you should reconsider why you’re using your current e-commerce platform at all.

I’m talking about personalization of the experience. Consumer are expecting this, as it has permeated their online lives and made it easier to flow from one product/content to the next without feeling completely overwhelmed.

Clearly, if you’ve been watching shows about bank heist (e.g. Casa de Papel), you’re expecting Netflix to suggest something along the same lines, don’t you? It’d be cumbersome if you had to spend time and energy browsing aimlessly at the infinity of movies and shows that didn’t remotely fit your apparent interest. You would actually leave, even.

Recommendation engine, as they are called (“engine” means “system”), are program that filters through the content of a database based on statistics about previous input, choices and behaviors. It turns data about users’ behavior into actionable customization, relevant for conversion.

Content-based filtering are derived from the current user’s actions and preferences — e.g. the user bought several sunglasses, so she gets recommendation for more sunglasses later on. Another exemple is Facebook: the more you click on a content, the more you’ll get similar content...

Maybe you’re starting to see how this can become an issue. While Facebook makes its bread and butter ensuring you’re clicking more of the same stuff (becoming a dubious echo chamber), it’s more noticeable — and annoying — if you get stuck on sunglasses when you’ve had your share of it.

Another technique is collaborative filtering.

Collaborative filtering draws from a dataset of several (hundreds, thousands...) users. Based on their behavior (purchases), you can predict the behavior of other users. The two variants in this approach are based on either users’ similarities (common behavior between users) or items (how users rated similar items). The latter is obviously a good alternative when you did not, or cannot, accumulate data on each users.

Some engines follow a hybrid approach, mixing the two filtering types.

[fashionvstech]

AI in your industry (1/6): supply chain and demand forecasting

Forecasting is key. Understanding demand prevents overstocking and understocking products. How can AI help supply chain and demand forecasting?

Even if you use existing data, take fashion trends into account and your own experience, no forecasting model will be entirely satisfactory.

Even greater than the economical challenge, there’s the ecological disaster it fuels. About 20% of global water waste is produced by the fashion industry, and still some companies burned unsold stock.

Awareness is growing, but how do you keep up with this challenge and face it, logistically? How do you better deal with short seasons, unpredictable consumer behavior and other unforeseeable variables.

You could, possibly, use the help of demand forecasting AI to mitigate issues in capacity planning, price production, stock keeping...

Deep learning for demand forecasting

Deep learning is a subset of machine learning.

Without getting too specific, you could imagine that machine learning feeds on big amounts of data and tend to be more task specific. While deep learning is more complex, possibly more powerful for some problems, trained on even bigger amounts of data, with results effectively more learning-based, less generalist, than the former. It has shown greater success on “chaotic” context needing to learn from scratch, such as image recognition, self-driving cars, etc.

The complex nature of deep learning is one of its downside. It is fully expressed in one of its main issue — the lack of transparency in how a model obtained results, which is key to make sound business decisions.

Fashion data being as chaotic as can be, demand forecasting tools should remain this: just tools. Your main asset here is your data analyst.

A good data analyst, experienced in the field she works on (i.e. fashion, luxury or beauty), will rely on prediction models at times, and deviate from them on others. Some tools, like Prophet from Facebook’s Engineering Team, help your analyst do her job in a “data-efficient” fashion, while remaining out of her way when needed.

[fashionvstech]

Whats is AI and how does it apply to your industry?

Artificial Intelligence (AI) is a field of computer science. It studies how machines can use “cognitive” functions to do some of the things humans do: learn, solve problem, and maximize chances at achieving goals. But how is it done exactly and how does it affect your industries?

How do machine learn, and why would they?

I’ll answer the latter first. When machine learn, they adapt to different situations. No need to code something specific for every possible situation — and maybe it was even impossible to do so in the first place. Maybe the “possible situations” are infinite and ever changing.

To understand how machines learn, let’s focus on a Machine Learning (ML), a subset of AI that yields the most interesting results and applications for consumers and businesses alike.

What’s machine learning?

It is a way of training machines to identify patterns in data, and ultimately deduce — predict — things based on what they’ve learned. Imagine reading million of rows from an Excel spreadsheet and guessing the future from there, using maths.

Here’s an over-simplified example... What’s the best formula for a bank to deduce if a client is a good candidate for a loan?

The bank has a huge Excel spreadsheet where all past loans applications are listed (one on each row). On each past candidates, they hold the same background information (age, revenue, loan amount). They also know whether they had an incident or defaulted, or if they successfully repayed the loan, and when any of these events happened.

The bank then passes one or more algorithms that they want to try on this old loan applicants list. Algorithm are not witchcraft: “2 + 2″ is an example of a simple algorithm. ML algorithms are much harder obviously.

Each algorithm tries the following: using as input the data from candidates — age, revenue, loan amount, etc. — it tries to infer results, e.g. deduce if the applicant repayed successfully or not. Specifically, results matching the ones the Bank already knows! If an algorithm does it every time, it’s a success. You have your magic formula.

Whenever the algorithm fails predicting results on a row, you make minor adjustments in its calculation to fit your need, and hopefully make it work better. And you do it repeatedly for a long time until you achieve your goal.

This is called training.

The setup — with repeatable inputs and known outputs — is supervised learning (I only mention this because there are other setups, e.g. to see pattern emerging from seemingly chaotic sets of data with no known outputs).

The result of this work, when you have an algorithm tweaked and perfected to meet your needs after a good training, is called a model.

No silver bullet

Although this has proven to be very effective (it’s been used constantly and consistently in finance, on the web and social media, etc...), it would be a terrible mistake to assume this could replace a human being.

Machines are stupid. They really are. They are perfect repeaters of redundant tasks, and great allies when you need to make a decision based on data points. But they don’t “think” like humans do, and although the mathematical complexity of coding AI algorithm is daunting, it’s nowhere near the unfathomable complexity of the human mind and of human societies.

You could argue that driverless car could be safer than human and thus, enamored with seemingly impartial, mathemically “perfect” AI, you would fall for anything branded with “AI”.

First of all — the term “AI” is, more often than not, just used a bait for your money.

Second — without a good flow of massive data, no relevant, bespoke AI for you, would it even be possible in the first place. So unless you receive massive Black Friday like traffic on your ecommerce site, you might want to think twice about the relevance of expensive, “AI-based” solutions for your customers.

Third — garbage in, garbage out. If your data is biased, the results are biased and useless. Maybe even damaging. Basically, using AI today on social subjects already stained by biases, reflect on AI itself. How do you teach AI the link between poverty and criminality? How do you teach it to understand of context of racial History that shaped the very fabric of some societies today? It is part of the current ethical issues surrounding AI, and even you think they don’t apply to your current business issue, they should act as a reminder that in the end, a human is in charge (or should be).

AI and your business

To dive into the practical side of things, I’ve written smaller, in-depth articles detailing how AI plays a role in specific topics of your industries.

— Supply chain and demand forecasting — Recommendation services (soon) — Visual search and image processing (soon) — Trend forecasting on social media and the web (soon) — Chatbots (soon) — Computer vision for augmented reality (soon)

[fashionvstech]

What is an OMS and what does it do?

Before ecommerce, mails or faxes were enough to deal with orders. In a world where sales and supply chain are omnichannel, how do you organize an order fulfillment process?

The order: the holy grail of commerce

A client buys something, and in return for your product being hers, she gives you money. Money that your business needs to exist. That’s the sacrosanct order.

The order has a life on its own. A lifecycle, even, involving many services in your company. Sales, Supply Chain and their warehouses, Customer Service...

If your business has (or grew into) a certain level, the need for an OMS is obvious, because of what it does.

An OMS — Order Management System — is a software that organizes as seamlessly as possible the lifecycle of orders to ensure their fulfillement. It integrates in one central place, information and commands for order processing across all existing channels of your business, for inventory management, for marketing processes, etc. It also connects with third party services (e.g. delivery) to provide related information pertaining to specific orders.

This way, all different facets of your business and their related employees have a single source of truth regarding orders.

A modern, effective OMS enables your organization to have an overview of product fulfillments. It should help you answer questions such as “how and when was this order made for this client?”, or “what is the expected delivery date?”. It also aligns different part of your system with the same exact information — e.g. stock level for your web site and your warehouse.

This is data enabling proper handling of cases impacting user experience.

For instance, your call center has a client complaining about an order — they took the wrong size for a dress. Your customer service agent files a “case” in the OMS for the order. This triggers the return process according to your policy, when the relevant teams are notified of it. To get the information they need, they’ll just refer to the central place — the order data in the OMS — and maybe rely on some of the automations provided by the software.

As implied by the names, automations are automated behavior triggered by an event.

A return process starts (event)? The OMS could send the client an automated, dedicated email to let her know it did (automation).

A client abandons her cart in your ecommerce store (event)? The OMS could send her a email to incentivize the sale (automation)... or file in an internal message somewhere to gather data points about it. It could bring insights for the Marketing team (to conclude that ”if many people abandon carts, as corroborated by Google Analytics, maybe there’s something missing in our shipping options”)

And in an omnichannel commerce world, where the same client shops in your brick and mortar store, on social networks, via affiliates, and in your ecommerce store, just imagine what a good OMS can do for your business strategies...

[fashionvstech]

Why are my developers saying "no" so often?

Being an efficient developer is a balance between getting things done fast and getting things done right. But it’s usually more complex than that...

Software is eating the world, and it turned developers into a coveted work resource. They can make or break a product by the feature they build and the bug they fix. But they are, above all, human beings who want to do their job right and tend to stress out when they fear they can’t. It’s especially true when you realize they are at the very end of the production chain and, as such, have to deal with the consequences of mishaps prior in the process.

What developers do for your business

Developers do many things, but we’ll focus on the two main tasks they do for your business:

1. They take business needs and requirements.

2. They turn it into something tangible and useful, through code.

Task 2 is technical.

Task 1 is all about soft skills and experience. Your developers performs best if they have an understanding of your business, and years of experience, especially experiences at failing. Why is that?

Beyond your expressed need, a developer will have to think about all the consequences, use cases, edge-cases, that come along with it. 

How foreseeable these consequences and edge-cases will be to her, depends on her experience as an engineer, i.e. how many times she has implemented similar or related features in similar or related environments. How many time she has failed doing so develops her ability to recognize patterns and tell-tale signs of failure.

Software development is hard

It definitely is.

Developers are aware of that, more than anyone. The success or failure of a software-related task rest on their shoulders — or so they have been lead to think, at least once in their career. If they have experienced failure due to a lack of planning, of good project or resources management, and no one around took their part of responsabilities in it, they probably took a bullet for the team and it left a scar.

All this, while being nonetheless responsible for actually engineering

So their go-to answer will be “no”. 

New features added during development but same delivery date? — “No”.

Everyone chimes in with ideas for features but no real leader in sight to steer the ship in one direction? — “No”.

Business collaborators at different levels fail to do their homework, and come last minute requests? — “No”.

The list goes on. 

Your developers are in charge of turning your dreams into reality, and will suffer dire consequences if they fail. So they want to make sure they don’t.

We could refer to the Agile Manifesto and methodologies to call out some developers for being quite obtuse — I wouldn’t disagree (I’m always surprised when I hear a developer complain about the lack of “code freeze” in an ever-changing, fast-paced environment like ecommerce). But similarily, we could call out some business managers for using “agile” as a buzzword which real meaning eludes them. “Agile” is not way to hide one’s own lack of preparation and shift responsabilities down onto the next in line.

So these “noes” are a little more subtle than just plain antagonism. More often than not, they are doing you a favor.

Three types of answers

If you ask a developer for something that is impossible, because of technical, financial or time constraints, you’ll get three types of answers.

A bad developer will say “yes”. This is dangerous and prone to lead a project to an excruciating failure. If you’ve ever outsourced any non-trivial work to a shady, extremely cheap developer, you know what I mean.

A responsible developer will say “no”. She will take a stand to prevent disaster by not changing position until you’ve re-examine your strategy.

A business-oriented developer will push it further, and say “no, this is impossible as is — but let’s work together to see how we can make some of it happen”.

Never let go of the last one.

[fashionvstech]

What is SAP?

Ever wondered what is this “SAP” your IT and supply chain colleagues keep talking about?

SAP is the king of Entrerprise Resource Planning softwares (a.k.a. ERP). Its company holds a large chunk of the $41B ERP business worldwide.

What’s an ERP?

ERP softwares are typically a suite of applications, dedicated to gathering, storing and interpreting data from the wide range of activities your business has. It is generally used for handling the core of your business data. Commitments: orders, payroll, etc. Resources: materials, cash, etc.

You could theorically run an entreprise-grade ecommerce business (minus the actual ecommerce website!) solely on SAP’s suite of softwares, called “modules”. There would be one or more modules dedicated to supply chain activities, one module acting as an Order Management System (OMS), one dedicated to HR, another one for Finance, etc.

The reasons your company wouldn’t do it, is because it is hard and expensive to implement. But if it is hard and expensive, why does it even exist?

I’ll provide an answer, but first — a History lesson.

What ERP changed in the game and how they came to dominate

Before the 60s, companies would have squads of white collar workers doing payroll and billing manually. Then computers came in and automated that. “IT” was then called “automated data processing”. There was no IT guy. You took Math(-ish) graduates and taught them programming on the spot. Programming that would seem terribly antiquated to your computer developers today.

In those dark ages of computing, companies would build custom made computer and programs to deal with these redundant tasks. They were basically reinventing the wheel each time.

SAP came in, built a software for one of those companies. Then, they re-used the same sofware for another company! Sounds pretty much what apps are about, aren’t they? You wouldn’t rebuild a new, different Instagram app for every different iPhone unit, would you? Well back in the days, you would. So SAP was a game changer.

They had built a software that was, by design:

— outputting real-time data on computer monitors, while other similar apps were priting data on paper (!) overnight to be read the next morning;

— built from an existing software instead of rewriting one from scratch for each new client;

— extensible from the start: they thought of ways to program this software that would make it possible to create “companion” apps in the future, depending on future needs (think of how Office Suite applications complement one another).

They made a software that would display stuff on screens, could be sold to several clients, and extended to respond to new needs. Sounds pretty basic, but all of this was revolutionary at the time.

It created a paradigm shift that impacted the IT industry as a whole. Companies like Compaq or IBM, building and shipping computer parts, had to change their business model to fit the new enhancements provided by SAP’s model.

As I wrote above, SAP is hard, and expensive to implement. Why is it still relevant nowadays? Why don’t companies get rid of it for something more modern, easier, and maybe less expensive?

Hard, expensive, yet ubiquitous

Each work of engineering engineers its user, proportional to that user's dependency upon it.

In other words — you and your usage of computers are formatted by the software you used first.

Your knowledge of spreadsheets comes from Excel, and this explains why Google Sheets is but a (superb) clone of Excel — they had to make it look and work like Excel for you to adopt it. It doesn’t mean that Excel is the best, smartest way to work with spreadsheets. It’s just the first (and only) way you learned spreadsheets.

In the same fashion, because SAP was the first suite of softwares to permeate entreprise IT solutions, a lot of companies and executives have invested time and money to install and learn how to use it. SAP is so hard to implement, that you can build yourself a lucrative career based upon just that — implementing it.

Once a multinational company has spent literally millions implementing SAP (including paying for license fees, installation from SAP specialists, etc...) there’s no turning back.

(That is, if they even managed to implement it properly!)

Have a few multinationals do it in a specific sector, and you have created the monopoly for SAP. And people working in this field will live by SAP.

That’s how your supply chain collaborators ended up on SAP — they started working on it early on, most likely because the warehouse guys were running on SAP, and now the rest of the company has to find a way to interface their own system (finance, orders...) with SAP. Either by using yet another SAP module in the chain, or by creating there own in-house applications.

Just for the sake of passing data.

Because passing data is hard! For a computer, a text file with the words “hello, world” on a single line is different than a text file with the same words on two lines! So imagine the myriads of ways different softwares format there own data...

If you want to make sense of so many different types of data, store them and pass them around the company in a meaningful way, and hopefully make some sound decisions based on them, you need a core system for gathering, storing and interpreting them. Just like you’d need a brain to make sense of different sensory input from your eyes, skin, ears or nose.

SAP was the first “brain” of its kind. It might not be the most efficient today, but it’s here to stay.