Data-driven E-commerce Marketing Process: The Iterative Engine

In the last article we covered a basic stack of marketing tools for e-commerce. In this post, we examine the second key component of the marketing engine: the process. We define and deeply describe a data-driven company and its components. We propose a clear marketing process you can adopt immediately in your company.

Perks of Data-driven Company

Using data in your e-commerce can provide tangible benefits.

Most companies do activities related to data, but they’re not truly data-driven. Let me give you a few examples.

  1. Buying the hottest BI / visualisation tool to throw fancy dashboards once a year.
  2. Taking a short, occasional look at Google Analytics and discussing it without any context.
  3. Bragging of the reduction in the bounce rate by 10% with 29 a/b tests, all of which lacked statistical significance.

For the purpose of daily work, we created our own definition of a data-driven company, which is aligned with one used by McKinsey Global Institute.

Data-driven company has clear marketing & analytics plan and undertakes an iterative process of improvement towards the goals placed in this plan, based on the statistically sound treatment of data of a proper quality (often with the help of various tools).

The key elements of this definition are: marketing & analytics plan with goals, iterative process of improvement, and statistically sound treatment of data of a proper quality. In the next three sections we’ll analyse each element in-depth.

E-commerce Marketing & Analytics Plan with Goals

Marketing & analytics (further: M&AP) plan tells you how well your marketing is working. It keeps track of the progress towards marketing goals in your e-commerce on the long-term strategic level, medium-term tactical level, and short-term operational level. It is detailed enough to help you gauge if you’re going in the right direction, and general enough to fit on one A4 page.

The M&AP format we’ll walk you through in this post is our authored version. We think it’s a consensus version of approaches postulated by many experts, including our guru Avinash Kaushnik, whom we personally admire, learned a lot from. In his own words:

There is one difference between winners and losers when it comes to web analytics. Winners, well before they think data or tool, have a well structured Digital Marketing & Measurement Model. Losers don’t.

Failure to create M&AP or update it on a regular basis will cause your marketing to turn into mess. There’re just too many moving parts in this system. Winning companies grow according to the M&AP, losing companies disregard it. Supporting the argument with the view of Avinash Kaushnik again:

The root cause of failure in most digital marketing campaigns is not the lack of creativity in the banner ad or TV spot or the sexiness of the website. It is not even (often) the people involved. It is quite simply the lack of structured thinking about what the real purpose of the campaign is and a lack of an objective set of measures with which to identify success or failure.

Marketing & Analytics Plan for E-commerce Company

M&AP for e-commerce business is composed of three categories: stageslevels, and dimensions. 

A sample M&AP for an e-commerce company could look as follows.

There are 5 stages, 3 levels and 4 dimensions in the plan.


Stages reflect the elements of the marketing funnel and comprise of Awareness, Acquisiton, Behavior, Conversion, and Profitability. Awareness measures how many people know the brand of your e-commerce. Although it’s uncommon for other online analytics frameworks to adopt it, we believe this part is valuable and therefore is needed1Acquisition measures how many people know your e-commerce and want to do something about it. It informs about the amount and quality of traffic from paid (advertising), owned (website, content, social media) and earned (sharing) sources. Knowledge of a business always precedes conversion to the acquisition stage, even if this knowledge is fragmented and short-term. For example, a click through the Google results page to the landing page of the e-commerce website by a completely unaware visitor is a conversion from the awareness to the acquisition; the prospect has just been in the Awareness stage for a few seconds. Behavior stage measures the conscious activities of Acquisition graduates. It informs how well the activities of visitors are aligned with the key goals of the website. Conversion is a behavior that carries a monetary exchange. Profitability measures how well the online business is doing. What makes it a valuable addition to the framework is that it bridges online marketing to business variables2.


Strategic, Tactical, and Operational levels lock the marketing stages in the reality of business execution. It puts marketing on a common denominator of a company. Treating marketing as a separate thing is a risk that marketing deliverables will fall through the cracks and nothing will get done.

Strategic level is the long-term of your e-commerce3. Its purpose is to guard your business against irrelevance caused by market trends. Tactical level goals at e-commerce business focus on the medium-term. Their purpose is to reduce the risk that seasonal turbulences will negatively impact your business. Operational level deals with “the now” or “the asap” of your company. Goals at various levels have to be aligned for any impact to happen. You intuitively know how long each of these levels is in your company. If you’re still not sure, triangulate using the similar size competitors.


For each stage-level pair, there should be a clear goal you’re working on to achieve in relation to a particular user segment. For example, at the tactical level, you may want to acquire more millennial visitors, because they had an unusually high conversion rate in the past. Decide on the KPI to measure the overall progress towards the goal and a particular Target you want to win in a given sprint. To resume the example, the KPI would be millennial traffic volume and target – a 15% increase month over month.

Iterative Process of Improvement

The well-orchestrated stack of marketing tools is only a half of the equation. The rest is a bundle of internal company processes that tools are wrapped in. Marketing managers at enterprise companies confirm it4.

Over 50% of the respondents claim that better alignment of internal organizational processes would enable them to make better use of data technology in support of marketing and advertising. In other words, managers noticed that tools often introduce overlap in responsibilities which can be reduced by getting clear on who’s responsible for what.

Fragmented Story

As we’ve noted in the first article, the attention of your users is (increasingly) fragmented over many different channels like Snapchat, Pinterest, Instagram, Facebook, and others. Marketing campaigns feed off users’ attention, so more fragmentation of attention causes the campaigns to be potentially less powerful. However, nothing is lost in the universe. Reduced power of individual channels is balanced out by the power increased carried out by well-orchestrated omnichannel campaigns. These campaigns engage users across various channels in a way that’s native to the channel and increases the propensity of that user to convert.

However, we live in a noisy and turbulent environment, as we’ve showed in the first post in this series. If marketing at your e-commerce company does not evolve and keep up with the changes, the constantly evolving marketing environment will make it powerless, and will expose your company to a significant risk. Noisy environment and a pressing need for the omnichannel campaigns has to impact how companies execute on marketing. It needs to shift from campaign-based execution to iteration-based execution.

Enlightened trial and error outperforms the planning of flawless execution. – David Kelly, founder of IDEO

Your e-commerce company generates data every day. If you know how to analyze data and do it on a regular basis, more data will usually mean more data-driven insights. More insights, then, mean you can better understand the in-and-outs of you business and improve it. If you don’t stop executing on this data-insight-improvement cycle, you’ll see a compounded positive change. The ultimate goal is to create the virtuous cycle of improvement. The value derived from data compounds. If it didn’t, advances in artificial intelligence and deep learning would not be possible.


We’ve tested many types of iterations servicing our clients and we know which one delivers the best results. Each level from the marketing & analytics plan will run on its own iteration.

  • strategic level – 6 months iteration sprint
  • tactical level – 1 month iteration sprint
  • operational level – 2 weeks iteration sprint

At the beginning of each 2-week iteration sprint, hold the sprint planning session. During this session, pick one goal to improve upon. In other words, at the beginning of the two-week period, decide on the operational level goal from a chosen marketing stage (acquisition, conversion, etc.) to improve upon. Repeat this at the beginning of the next month, i.e. pick one goal from the tactical level to improve upon. Repeat for the strategic level goals. Make sure lower-stage marketing goals you’re optimising for are aligned with the higher-stage goals. Don’t try to improve more than one goal per sprint, otherwise you risk receiving confounded results and returning to the exactly the same place where you’ve started, knowing less than you knew before, wasting money and resources along the way.

Operational level is where execution happens. Therefore, put your improvement hypothesis before every next operational stage, as well as prepare all the key contents (e.g. new landing page or e-mail variant). The new contents should air as fast as possible at the beginning of the new operational sprint. Doing so will ensure as much data as possible is gathered, and will further help clarify the improvement hypothesis charted for a given sprint. This brings me to the final point.

As you improve your marketing in an iterative way, don’t forget to review data on a regular basis and create improvement hypotheses for a given level-stage pairDon’t just look at the data (be it in Google Analytics for example) in a purposeless way. Data is trivia presented in numbers, so it’s always engaging to look at it. Always map your exploratory data analysis to a given goal of a particular stage. If you don’t find anything valuable of such a high strategic level, do exploratory data analysis in relation to the tactical level goals. Start each exploration of data with a very specific question.


Let’s sum up. Below you’ll find a 7-step plan to build a growth engine in your company.

Step 1: Fill out your marketing & analytics plan.

Step 2: Decide on a strategic level goal you want to improve for the next 6 months.

Step 3: Find an operational level goal that supports a tactical level goal that supports a chosen strategic level goal.

Step 4: Before the sprint starts, create the improvement hypothesis and improvement content delivery plan. Make sure it gets delivered before the sprint starts.

Step 5: Launch the new contents at the beginning of the next operational cycle. Keep it up for two weeks.

Step 5: After the sprint ends, review the results, assess the lift, continue next experiment towards a given goal or, if the goal was achieved, move on to other goals in support of tactical level goals in line with strategic level goals. Update the marketing & analytics plan.

Step 6: As you’re performing the iterative improvement, do exploratory data analysis in the data you have, but do it with a clear intent. In other words, start with the question and then apply the curiosity, not in reverse.

Step 7: Pick the next tactical goals to improve upon on for the remaining of 6 months. Don’t forget to update the marketing & analytics plan with strategic goals for the next 6 months.

This framework is rigid and demanding, because marketing has a lot of moving parts, and the only way to have everything under control is to have a very clear process around it. Note that this process doesn’t diminish the creativity of your marketing department, or its spontainety. If anything, it can make it even more creative, because creativity likes clear goals and unlimited approach angles. Iterative approach give you just that. You’re optimising for a very particular thing however you like. Creativity without any boundaries or expectations is not useful in any domain other than independent art.

Statistically Sound Treatment of High-quality Data

Data should not be touched unless it’s statistically significant. Misuse of data can have a detrimental effect on your business.

You’ll use marketing in your company in two ways: passive and active. Passive usage happens when data are sent automatically between the tools, or handled automatically by one tool, in order to autonomously fulfil a goal written in the M&AP. In this scenario, you only assess the lift of target towards a given goal, and don’t do anything else. You enjoy machines doing the job for you. Example of a passive data usage is switching from a manual budgeting of Adwords account (stage: acquisition, level: tactical, goal: decreasing the average cost of click) to the automatic one. Active usage, on the other hand, happens when you’re responsible for gathering enough data and deciding if the improvement toward a goal happened, or not.

In the passive usage scenario, you don’t have to worry about statistical significance. If the tool promises to progress your company to a given M&AP goal, it has to employ some sort of learning algorithm behind the scenes (this is the case in Adwords). Such algorithms have statistical significance embedded into them, so you don’t have to worry. Grab the results and run. The only scenario where you actually need to worry about the significance is then the active usage scenario. You’ll most likely use tools in the active usage scenario too. That’s when you have to be the most cautious.

In this section we’ll introduce the concept of statistical significance. In the next sections and article we’ll refer to it frequently as we help you achieve various marketing goals. Although there exist various tools that can make statistical significance faster or easier, the concept itself will shape your thinking in a certain way, and therefore will help you make better marketing decisions. Don’t be alienated by a bit of mathematics here and there. I’ll make it as easy for you to follow as possible.

Defining Statistical Significance

Let’s say you receive the results of the latest a/b testing experiment revealing the response rate difference in two campaigns that differ only by the in-e-mail call to action button color and no other element. You assume that the new button color variant will not impact the response rate. If the post-experiment data shows that the response rate difference between the old and the new variant of the button color is statistically significant, it means that it’d be very, very unlikely for your original assumption to be correct. Statistical significance as a term functions exclusively in the realm of statistics and you shouldn’t mix it with popular meanings. In particular, statistical significance doesn’t equal practical or business significance. For a result at hand to have business significance, it has to be (1) statistically significant and (2) clearly map to the progress towards any of the goals written in the M&AP.

In the statistical terms, your assumption would be called null hypothesis (H0). The phrase very, very unlikely corresponds to the p-value, which is the probability that we would get this result if our null hypothesis was true. If p-value is smaller than a preset level, often 0.05, we reject the null hypothesis. In our example we’d reject the assumption that the new color button variant has no impact the response rate.

This method isn’t perfect and has been widely criticised. The questioning often targets the mentioned preset level of 0.05.

Statistical Significance in Marketing

The difference between theory and practice is larger in practice than the difference between theory and practice in theory. – Jan L.A. van de Snepscheut

Taking advantage of the statistical techniques in day-to-day marketing is a secret weapon, but demands a lot of resources on the level of analysis, trust, and culture. There are two key benefits of doing so:

  • Focus on the things that matter.
  • More cost efficient marketing and greater speed.

Things That Matter

Modern marketing has hundreds of moving parts. It’s very easy to get lost. It’s hard to prioritize. However, only a small section of the entire system matters. Identifying the crucial components of the engine is the key. Getting to the place where noise is small and signal – the conversion results – are huge, can take a lot of experimentation, but it’s the holy grail on online marketing.

The road to this place leads through statistical significance. When you incorporate it into your marketing, you immediately teleport to a new reality where you essentially do not see things that don’t matter. They do not exist in your consciousness and hence they don’t produce noise on your decisions. However, when they finally do become significant, and they’ll if you have a correct process in place and don’t fail to execute on it, you’ll immediately see them.

Cost Efficiency and Speed

Ignorance of statistical significance in marketing leads to huge budgets being thrown out of the window. The proper approach is to treat marketing projects as if they were science projects. In particular, you have to figure out which channels your customers care about, and how much they care about them. Assuming that your target user spends just a few seconds in each channel, you have to figure out how to utilize those microinteractions to convince him to care about you. That implies testing many different configurations of content that may work best in convincing them. This is exactly where statistical significance shines. Conversely to the intuition, it’s is the fastest and safest way to scale your marketing. If only you encounter a sample customer journey composed of a few statistically significant spots, then you can scale the entire funnel as fast as possible, because the risk of this funnel breaking down once it’s micropoints reached statistical significance is very small.


E-commerce companies have to get the most out of every single interaction of their customers. The more channels (e.g. Musically, Instagram, Pinterest) and more platforms (e.g. VR) will only enforce the meaning of this statement. The only way to achieve is to treat your marketing as an iterative science project with a very clear plan on the backend.

  1. Awareness has always had the potential to generate significant value. However, it’s never been so easy to adopt it as a business tactic, because the barriers of entry were too high. Since nowadays everyone can get awareness via social media, it is yet another tactic. Therefore it makes sense to include it in the general-purpose business framework.
  2. Conversion is different from profitability. What if your e-commerce website converts at 50%, but you spend more money acquiring visitors than selling to them. Does it make a viable business model? No.
  3. The length of the long term depends on the business and it can range from 1 year at an e-commerce startup to 5 years at a market leading e-commerce
  4. Source of the numbers and graphs in this section: Winterberry Group White Paper: “Marketing Data Technology: Cutting Through the Complexity.”, 2015. The white paper findings are based on the results of a research that included phone, online and in-person surveys of more than 50 advertisers, marketers, publishers, technology developers and marketing service providers. Click to read more.