11 tips for marketers: how to analyze data properly and win with the help of Artificial Intelligence?

Robots that disguised a man, known from blockbusters like “She”, “A. I. Artificial Intelligence” or “Ex – Machina” are slowly cease to be only cinematic fiction. And although robots pretending to be a human and walking down the streets are still the part of far away future, intelligent learning algorithms, able to analyze data, draw conclusions and recommend us the best solutions – are already part of our reality. We can find them in maps analyzing the fastest way to our homes, virtual translators etc. What’s more – advanced mechanism of data mining – can help us to better understand the customer in any industry and increasing sales. They will set a campaign automatically, prepare a report, they will recommend the changes. But where there is a place for our honest marketer ?

How quickly machines learn?

Internet grows annually by 40%. This is a huge amount of data, which appears online every day. Equally rapidly is growing amount of data about our customers, about what do they click, what do they like, what sites they visit, etc. None marketer is able to analyze this data set itself. But in this case we can be supported by a system that can analyze this data for us, and what’s more, not only provide us with conclusions – but decide what to do next, and put the recommendations into practice.

Simply, Machine Learning is a science dealing with artificial intelligence, and its purpose is to allow the possibility of automatic knowledge enrichment, reasoning and decision-making by modern software. Machine Learning allows us to put together all the data we have about the customer and his behavior and get from this information relevant solutions for business development and sales. It sounds like black magic? Relax, it all happens automatically. For the marketer, it is important to know what kind of data to expect and how to analyze them.

The profession of the future

According to the survey “25 Best Jobs in America” created by the Glassdor Survey’s, most wanted profession this year in the US will be data scientist. What skills should have the ideal analyst? Not only analytical thinking, but also the ability to draw conclusions, creativity, ability to ask the right questions, and combining data and facts from various sources.

It is worth remembering that these skills are also important for current staff in marketing, accounting, CEO and many others. When intelligent systems begin to be responsible for activities related to automation and data processing, the man will be eventually forced to verify the data that system provides him.

How do we get the data for analysis?

There are two ways to gain data for the analysis purposes:

  • behavioral data – related to user behavior on the site – in which places on the website someone clicks, what is watching, what is buying. We receive them with pre-installed code tracking on the website,
  • declarative data – acquired from filled forms, surveys, mailing databases, loyalty programs – data declared by the user. These are mainly information such as e-mail, name, address, age, education, etc .

The system not only collects all the data for us, he also processes it. However, marketer should know what to do with them later.

How to analyze the data properly?

1. From the general to the particular

Do not focus on the analysis of individual cases. Depending on the search based on the analysis of the entire data set.

2. The strength of the correlation

Correlate collection of data about your customers with other collections and search for dependencies. Check if the exchange rate, month, day, world events etc. affect significantly change the buying behaviour of customers.

Synerise is a good example of a system that correlates the various data thus are checking dependencies. An interesting solution, which Synerise proposes, is a comparison of transactional customers’ data with the current weather. In this way we can keep tracking of the relation between the profits from the sale and the current weather.


Source: www.synerise.com

3. Search for links

Pay attention to the fact that the products are linked to each other – a person who is buying a laptop may be interested in a laptop bag, but probably will not buy the keyboard. For more information on related products and analysis of customer shopping cart, see our previous post HERE

4. The devil is in the details

Consider that sometimes the least important data gives us some information about the customer.

  • Operating system – informs us what kind of income can a customer have,
  • Browser – it is worth to pay attention which browser is used by the majority of customers and so adjust pop – ups, dynamic blocks, etc. to properly show up in these systems,
  • Devices – people who use more devices are more dependent on new technologies and tend to spend more time on the Internet – it is easier to hit them with on-line advertising.

5. Learn all about those who buy

Analyze data about customers who have recently purchased your products/services. Pay attention, what devices they used (or perhaps make a purchase in the stationary store?) what time, in which days. Once you know the answers to these questions it will be easier for you to select the appropriate promotion model. For example – if the most people buy your products using the computer, do not invest too much in advertising on mobile phones, and if most of them additionally make a purchase on Saturdays and Sunday, focus on promotion is in these days.

6. Analyze the entire path to purchase 

Check the ways in which customers come to the purchasing process. Check how many of them was led to market basket by a campaign, how many of them had previously been on the site and how much times they previously viewed a particular product. Monitor if they already were at stationary store, how much time they have spent there. All these data may allow you to find the patterns and figure out what influences their purchasing decisions.

7. Build segments

Automatic segmentation allows you to better analyze data and see which results increase or decrease your  sale. For example, if you notice at some time the sharp decline in sales, while a large number of users were on the site. In this situation you can ask yourself a question – why they did not buy anything? You should analyze it on the basis of separate customer segments. If the system creates them automatically – you can quickly locate the segment that contributed to the decline in rates. Maybe the loss is generated by users 50+ (and perhaps we should increase the UX of the page and font size?), or people from specific regions where the product was unavailable. Segments can answer many questions.

Synerise automatically creates segments in terms of demographic and behavioral issues. To the specific segment is added anyone who has agreed to the processing of personal data and somehow confirmed to us for example; his age (eg.selected age category on the survey), location (using GPS, or completed it in the contact form), gender (intelligent recognition of names of male and female) etc. What’s more – segmentation is updated in the real time. If the GPS locate a user in a different city than it was originally labeled – user is automatically assigned to the new segment. The analyst can also create any system segments associated with user’s behavior, eg. if the customer is a man, who bought a minimum of 3 products – and all came from the Casual category – can be put in a segment ManWhoBoughtCasual. You can create as big amount of segments as you want – but remember to create only those that actually give you important information.


Source: www.synerise.com

8. Do not judge by the conversion

Conversion is often used by marketers in their reports. However, it should be noted that the conversion of 20% will achieve when 10 users visit the site and 2 makes a purchase, but also when 10 000 visitors will be on the site and 2 000 users will buy the service. Can you see the difference? Therefore is important to take into account in your reports nominal data (the number of page visits and number of transactions)

9. Pay attention to seasonality

It concerns industries selling seasonal products eg. skis, skates, but also eg. christmas trees, glass balls, bathing suits, etc. Smaller sale of bikini in winter does not mean that the store is unprofitable. Likewise – if you notice an increase in access to the site in March, which does not goes into sales – it is not mean that we have a weak deal. Probably users are just beginning to look for swimsuits before the summer season, but they need more time to decide. You should then take care of the page optimization, launch a new promotions, analyze how many of those visits ended up buying from store etc. Real conversions should be count not only by the average value from whole year, but also for a particular season of the year.

10. Search for errors

Propoper analysis of data allows you not only to find the accuracy and better personalize promotional activities, but also to quickly detect anomalies. This was the case of Walmart, which has long been monitoring all the data from sales of their products. On Halloween, they monitored the sale of special Christmas cookies. According to statistics, in many shops total sale of these products were high, but in some of them it was zero. Analysts immediately contacted these stores to ask whether cookies are present in the right way – it turned out that in some shops they have not been laid on the shelves.

An example can be also statistics about the duration of page visit. If Internet Explorer users are staying on the website much shorter than the others, the system determines that they might not be interested in the offer. We as analysts can check the speed of page loading – this may be due to the fact that older browsers will be slower – adjusting the page to every browser, we can gain (if such users were plenty) new customers. Right analysis allows us to quickly find anomalies, and even faster to counteract them. It’s important to know these hot spots to identify and check why some results – from specific stores, communication channels, days – are so markedly different from the others.

11. Watch real

Much data correlated with each other, gives the illusion that both variables actually have an impact on each other. And here we see the advantage of a specialist who can look at these figures realistically and say whether that influence is not just a coincidence (system currently can not tell us that). A few years ago, Forbes reported that according to studies there is a correlation between the murders in the United States and the market shares of Internet Explorer. But even if you can see a certain mathematical relationship between the abstract data, it does not mean that in fact there is a causal relationship between them. As you can see – the attendance of a human in the process of data analys is still necessary.


Source: http://www.everydaysociologyblog.com/2013/05/thinking-critically-about-statistics-and-their-sources.html

Worth analyst weight in gold

World managed by the machines, once and for all change the responsibilities and requirements for the different jobs. Many of them will be fully or partially automated and replaced by intelligent systems. But that does not mean that people will no longer be needed. Creation, campaigns management and promotion through developing artificial intelligence will never again be the same. We can observe that the amount of data and information about customers is still growing. The question is not whether do so – but how to analyze the data. Complex analytical system would be useful in this process, but also the man who will be able to draw valuable conclusions from these data which can help in creating more effective strategy and provides bigger profits and sale.

Małgorzata Wojtasik