Prevention is more effective than cure, just as making decisions based on data and predictive modeling is more effective than making decisions by adapting aging reports to the modern reality.
The Gartner Analytic Ascendancy Model describes types of analytics based on the questions they help answer:
The modern market for analytics is stuck somewhere between descriptive and diagnostic analytics. Why? Because everyone has learned how to harvest data on customer behavior but few have learned how to activate it. Currently, only a small percentage of data collected by businesses is used for analysis. We need tools to extract insights from big data.
There are five main mistakes companies make when they start building predictive analytics and forecasts.
Mistake #1: Ignoring the complete data set and analyzing only part
We want to understand not only why things happened in the past but what the future will bring. Most modern analytics is retrospective; it’s based on historical data to explain past events, reactions, successes, and failures. Though companies collect fragmentary data, they still try to build reports manually and ignore necessary metrics, leading to unreliable results and incorrect conclusions based on faulty analysis. Until this endless cycle of fragmented data is broken, it’s way too soon to talk about fishing true insights out of data and getting forecasts.
How can you deal with this? Start by building a business intelligence system where each stage of data collection, storage, and processing is automated, preventing human errors.
Mistake #2: Using primitive models
Modern AI algorithms can forecast the payoff from clients but can’t explain why it will happen: why client X will bring more revenue than client Y, what factors matter for increasing revenue the most, and when you should pay attention to them. The problem is that marketers can forecast the near future based on current data without using the latest technologies, but the number of parameters they can process are too few and the volume of data they can process by hand is too low for use in huge prediction models that provide deeper insights. The true value of AI in marketing can be realized only in more complicated scenarios: What will happen to the company’s revenue if we change our geographic and age targeting but keep the advertising budget the same? What will happen if we leave the same geographic parameters but decrease the advertising budget?
Nowadays, the best tools based on AI, machine learning, and game theory can answer questions like What if… or Why will it happen in this way? What’s most important is that the answer you get must be broad and understandable.
Mistake #3: Using limited data
Despite the rampant development of marketing models, they’re still raw and badly trained, requiring a whole world of data to build forecasts. This is the case because companies train their models on internal data alone; thus, corporate models are isolated from the outer world. Forecasts created by such models can’t be precise because nobody is alone in the market. We depend on competitors, demand trends, etc. All of these factors must be included when training a model.
Data you can use independently:
- Raw behavioral data on your own customers
- Logs from Yandex.Metrica and Google Analytics: Adjust, Segment, Mixpanel, Heap data
- Data on conversions and orders
- Data from different sources like your CRM and Yandex.Metrica
Can you build a model yourself and give it all the necessary training data? It’s possible. But if building models isn’t your main business area, you’ll likely expend a lot of resources on it and may not be satisfied with the result. Besides, the model you create must be supported and trained constantly. If you’re not ready for this, try to use a commercial solution.
Mistake #4: Underestimating the true value of analytics
As we mentioned, only pioneering companies can synthesize insights from their data. For most companies, the true value of forecasts is still foggy – like science fiction, and with a similar probability of being applicable in real life. Nevertheless, you should implement data-driven marketing starting today, even if you don’t know how to do it perfectly.
Market leaders all made mistakes years ago, which is why they’re leaders today. If you start collecting data, in a year you’ll have a solid foundation to start building a model or to try another approach to predictive analytics. You’ll get used to your data and see that making decisions on your advertising budgets based on forecasts is much more efficient than doing so based on historical data. You can still change the future, while the past is, as they say, history.
What you should try to forecast:
- Quantity of future conversions: orders, payments, subscriptions
- Revenue based on your customer’s LTV
- LTV alone
- What will happen if you concentrate on one set of channels instead of another?
- What will happen if you increase the budget for SMM or decrease the budget for PPC?
- What goods will be popular in California for an audience of working moms?
- Any metrics based on historical data
It’s considered best practice to apply this data for:
- RFM analysis. Create segments based on customers’ purchase activity and predict which segment will cool down first and what offer they’re waiting for today so you can increase your revenue tomorrow.
- Dynamic pricing. By collecting historical data on customers’ actions, you can form the most appropriate price for your customers’ favorite goods that they’re most likely to buy. Thus, you don’t have to offer lower prices than your competitors, as your customers are already interested in your offer.
- Managing bids for LTV segmented clients. You can forecast LTV for different cohorts, and if the forecasted revenue from cohort A is higher than from cohort B, you can increase the advertising bid for cohort A.
Mistake #5. Assuming that analytics is a race, not a marathon
This is the main mistake for everyone who doesn’t prepare for implementing predictive analytics. Implementing analytics costs a lot of time, takes a lot of effort, and brings changes to all processes in the company. So before you start, get prepared with this checklist:
- Ask the right questions
If you want to get the right answers, learn how to ask the right questions. Come up with your hypotheses and check them with A/B tests based on certain KPIs to evaluate success or failure. Remember the main questions you want to answer and be ready to go forward one step at a time.
- Use the right data
Before jumping to conclusions, make sure your input data wasn’t sampled or filtered. After you’ve defined your questions, define the data you need and check that you have enough of this data to reliably answer your questions.
- Use the right technologies
The market for data-powered technologies is growing intensively, as is the market for tools and services that help to collect, store, and process data. Most analytical services are starting to offer tools for predictive analytics that are based on different methods and mechanisms. You have to choose the most efficient for your company. Be sure you can use the tool you choose for a long time, as changing horses on the way will cost you lots of resources and is always an additional risk.
- Choose the right team
You’ll never build predictive analytics without a professional team who supports you and understands how to implement analytics in everyday processes. Without smart people around, you can’t ask the right questions and set ambitious goals. Also, people make all data processes work together, starting with preparing data for analysis and ending with synthesizing insights from processed data. The team is at the heart of introducing changes, creating the strategy, and building up the data culture. So even in our technological era, you should focus on people first.