Making sense of online data

Delighted to feature a 2nd guest blog post from Michael Ross, Co-Founder and Director, eCommera and one of the lead authors of the Trading Intelligence Quarterly.

Michael will be speaking at the ecommerce Networking Forum, (#eCloud), London 1st November. This is a by-invitation forum for key executives from the multi-channel retail community to meet, discuss issues of common interest with their peers, exchange ideas, and do business. CXOs and ecommerce directors from House of Fraser,, Glasses Direct, Matthew Williamson, Habitat, Secret Sales, The Hut Group, Chemist Direct, and Groupon are already amongst the confirmed participants. If you are a CIO, CTO, CFO or Director of ecommerce you should give some serious thought to getting there too.

Retail is detail. And data is the manifestation of the detail of retail – the basis of understanding the myriad decisions and choices to be made.

To turn raw data into profitable action, a retailer needs to succeed in:

  • Digesting the data. Understanding what’s happened in your business. Is business good or bad? What does the big picture look like?
  • Diagnosing the problem. Working out where to focus – quickly. What’s driven performance? Where should you place your efforts?
  • Doing something about it. What needs to happen? What actions should take priority? How do you monitor and measure success?

When retail is purely physical, relatively simple data is all you need to make ‘good enough’ decisions. That data is mature, easy to understand and quantised by store. Most physical retailers have enjoyed success without having to employ too many (or any) statisticians.

Online retail has introduced an order of magnitude change in both the number of decisions to be taken, and the complexity of data available. Furthermore, there is still no agreement on eCommerce norms and best practice for how to interpret and respond to what the data shows. The challenge can appear daunting and it is easy to either disengage or to drown in the data, but failing to get to grips with the numbers is a huge gamble. At best you miss opportunities to grow profits; at worst you lose control of your online channel.

Online retail data has to be managed differently to physical retail data, using a fundamentally new approach. This article suggests how to navigate the data minefield of online retail:

  • Why is offline data ‘easy’?
  • What’s the big challenge with online data?
  • What’s the best approach to making sense of your online data?

Why offline data is ‘easy’

There is of course plenty of data in physical retailing. However, the construct of physical retail makes the data clear and quick to interpret:

  • The key ‘inputs’: The things that a retailer can control are mostly decided upfront. Inputs include store location, rent and shop fit (and are thereafter fixed or sunk costs). Each store has obvious constraints on its size and reach so the number of day-to-day decisions and actions are limited. Decisions about product and inventory are the critical controllable inputs on an on-going basis.
  • The ‘outputs’: The things that happen as a consequence – tell the story of the business. Retailers can look at money in the till (and stock on the shelves) at the end of each day and know if business is good or bad. Sales are a good proxy for profit and stock turn is a good proxy for cash flow.

Moreover, the maturity of the market, the historical norms and the focus on manageable entities like stores mean that data is not something that keeps retailers awake at night. A small number have really embraced the use of data (Tesco’s Club Card and Walmart’s Retail Link), but few have made it a core part of their business. So, why is the data so simple to manage?

Digesting the data

A retailer can look at performance by store, by category and if needed by-store-by-category. Given that most retailers have fewer than a thousand stores, the data across the business is relatively easy to digest.

Diagnosing the problems

A retailer can look at store rankings and likefor- like sales – both outputs – and know exactly what is going on in its business. Again, the natural quantisation of retail makes diagnosing problems relatively easy. For example:

  • Store like-for-likes: If all stores are up 5 per cent except one, you can be pretty confident that there’s a problem in that store.
  • Product sell through: A product has either sold, or it hasn’t. The actions are blunt: buy more/ less; buy/cancel similar; pull forward orders; or increase/reduce price.

External benchmarks are well established and widely known and internal benchmarks are available by simply comparing across stores and categories. Also, optimisation is one dimensional. For example, working out how much stock to order of a continuity item – the Economic Order Quantity (EOQ) – is complicated, but solved, and retailers understand on a SKU-by-SKU basis how well they are doing.

Figure 1: Economic Order Quantity
The formula for the economic order quantity looks something like this – not pretty, but now embedded in all retail ERP systems.

Doing something about it

In many ways, the key challenge in physical retail is the doing – getting lots of people across a store network to execute flawlessly is difficult. But the construct of physical retail is an advantage for turning data into action:

  • There are only a limited number of actions that a retailer can make on a day-to-day basis.
  • Actions can be taken independently, so a store manager can re-merchandise a store and a central merchandiser can reallocate stock or plan a markdown.
  • Results can be observed by a manager or CEO walking the store.

The challenge of online data

In comparison, using data generated by online retail is very different, and much more difficult. Why?

  • The ‘inputs’ of online retail are both more numerous and more changeable. Online there are fewer upfront constraints on scope or reach. Combine this with the almost limitless possibilities for adjusting the hundreds of day-to-day activities – keyword bids, navigation, delivery charges, sort orders, prices, marketing, promotions or customer emails. The consequence is that the data is considerably more complex – there is more data, more often, from more interconnected sources (see figure 2).
  • The ‘outputs’ – sales, conversion rate, average order value – are as a result unhelpful, as each can be affected by a wide variety of causes. The outputs online tell you what’s happened but give no clue as to what’s actually going on.

Moreover, there are few accepted norms on eCommerce metrics – no one really knows what ‘good’ looks like. And the variable cost structure of online means that optimisation is multi-dimensional and more complex. These challenges manifest themselves across every aspect of an eCommerce business.

Figure 2: The interconnected activities of online retail

Digesting the data

It is extremely difficult to get a holistic view of online performance.

A typical retailer will create more data in a day from its online operation, than its stores would create in many years. Unfortunately, the data is typically from different systems and often does not reconcile.

The ‘physical retail’ approach treats the website as a ‘store’ and sales-by-category would be the primary dimension for analysis. Unfortunately, this gives no sense of the performance of customers, marketing and the site itself.

Online retailers are either faced with an overwhelming amount of spreadsheets or they choose the ‘easy’ route of focussing on simplistic outcomes like sales, traffic, average order value and conversion which give no insight into what is actually going on. Net sales – which tells physical retailers so much – is affected by the phasing of orders, shipments and returns, and can be easily flattered by promotions and marketing spend which are not readily visible.

Diagnosing the problems

The interconnection of product, site, marketing and operations means that it is rarely obvious what problem you are trying to solve, and mathematically complex to work it out.

So, for example, when conversion rates fall everyone assumes the site is at fault, yet the most likely culprits are the traffic mix, product availability and price competitiveness. Similarly, if a product isn’t selling it may be because no one is looking at it, or people are looking but not buying. Identifying the real cause is not something that should be left to gut instinct! To add to the challenge, in most organisations the varying causes are owned by different departments which can stymie the diagnostic process (see figure 3).

Even once the issue is identified, diagnosing the right action is difficult. Online marketing constructs mean that visitors, and therefore each product view, has a cost. This is completely different to physical retail, where the marginal cost of a store visitor is zero – you pay your rent and traffic is free. As a result online product and traffic are intertwined.

Figure 3: Why is a product not selling?

A retailer can now decide to give a product more or less traffic depending on product stock levels. If a product is overstocked, reducing the price is a blunt (and potentially ineffective) lever if the cause is lack of traffic. The optimisation challenge is whether overall product profit is maximised by:

  • Increasing or reducing stock (to drive turn/sellthrough);
  • Increasing or reducing the traffic to the product (to drive volume); or
  • Increasing or reducing the product’s price (to drive margin/conversion).

Figure 4 illustrates a new type of analysis for online retailers. The natural store decoupling of stock management from store operations no longer works.

Figure 4: Find the optimum views per SKU

Doing something about it

The real challenge in execution is getting visibility on what is being done and to what level of quality.

There are thousands of small actions online that are worth doing, and ironically many are easy to execute from the comfort of a desk and a screen. Conversly this makes it very difficult to track what is being executed effectively.

Having identified the issue, the challenge is to ensure high quality and consistent execution. For example, if a retailer identifies that a thousand products are receiving no traffic from Google, it is often resolved by manipulating bid rules in a keyword management system. It is extraordinarily difficult to give a retail CEO visibility that this task has been executed, and that the problem will not repeat. Now multiply this across the hundreds of daily and weekly activities of a mature online business and the scale of the challenge becomes obvious.

Making online data work

Navigating online retail data requires a new approach. To make the data work for you, to ensure it drives your business growth, requires the following:

  1. A single view of profit (digest): A new data model providing a holistic view of your business is required. This will allow you to quickly understand whether a performance delta is driven by customers, geographies, marketing channels, categories, site or something else.
  2. New algorithms (diagnose): Making good decisions requires a new set of algorithms and diagnostic tools. It is vital to look at the new mathematics of retail – for example, the fully allocated profit per product, customer attrition rate and marketing sensitivity to attribution. Expect new formulas – traditional retail metrics simply don’t make sense in the online world. De-averaging (see ‘Tyranny of Averages’) is critical to understanding what is driving performance.
  3. Manage by input (do): Most importantly, online retailers need to track inputs across their organisation. Managing inputs is the only possible way to understand what has happened in your business and what you can do about it.

There are literally hundreds of inputs to measure. Focussing on the right inputs in the right order is one of the key management challenges online and – critically – the number and specificity of inputs to be tracked evolves with scale. Figure 5 summarises a good starting point: our recommended top inputs to track offering the greatest insights on what is happening to an online retailer’s business. However, retailers doing £20m+ sales online should expect to be monitoring over 100 inputs.

Figure 5: Some good inputs to start

The complexity of online data is challenging for all retailers – both in how to make sense of the vast mountain of information now available, and also in how to use it to make the thousands of potential decisions daily.

Online retail is data. The data required to make almost every decision is available with very high quality, in near real-time. The way forward for those wanting to grasp the opportunities of eCommerce is to re-instrument around inputs.

I once worked with an online retail chairman who used to bang the table and shout: “What are we doing to drive conversion rate?”. The business didn’t thrive, and nor did the table.

Michael will be speaking at the eCommerce Networking Forum, (#eCloud), London 1st November. This is a by-invitation forum for key executives from the multi-channel retail community to meet, discuss issues of common interest with their peers, exchange ideas, and do business. If you are a CIO, CTO, CFO or Director of ecommerce you should give some serious thought to getting there too.