Artificial Intelligence for Marketing
Updated: Feb 21, 2021
"Half the money I spend on advertising is wasted. The trouble is, I don't know which half."
- John Wanamaker, American marketing visionary
John Wanamaker's lament, though tongue-in-cheek, is as true now as it was when he said it at the turn of the 20th century. Today, marketers estimate that they waste 26% of their budget through a failure to draw meaningful insights from historical data. This failure has led to a loss of valuable customers for 30% of firms. And the stakes have never been higher.
The marketing industry is a global behemoth and shows no signs of slowing down. In 2020, the industry was estimated to be worth USD 322.5 billion. It is projected to grow at 10.3% until 2027, when it will be worth USD 640.2 billion. Big money is being spent and every cent counts.
How do you know your marketing strategy is optimised? What does an optimised marketing strategy even look like? Deep Blue AI's suite of solutions for retail businesses can help answer these questions. The following case studies explain how.
Case Study: Bakery Marketing
A hypothetical small bakery Baked Goods Inc. has the contact details of 10,000 customers - all customers that have registered on their website. Baked Goods Inc. has a new doughnut that they want to market to its customers. Based on historical data, they estimate there is a 2% response rate, ie. 2 customers out of every 100 contacted will come to the bakery and buy their doughnut. Each doughnut is sold for a $4 profit. Contacting each customer via email costs 2c.
Baked Goods Inc. might do the math and decide that their best strategy is to contact every single customer. Then their total spend on this advertising campaign is $200. Their profit from selling doughnuts to the 2% of respondents is $800, for a net profit of $600. Their net profit per marketing dollar spent is $3.
But is there a smarter way to market their doughnuts?
What if we could find out which customers will buy our doughnuts and send marketing emails to only those customers?
Assume Baked Goods Inc. had some way of predicting exactly which 2% of customers will respond and only contacted those customers. Then their total spend on advertising is $4, for a net profit of $796 - an increase of 33%. Their net profit per marketing dollar spent skyrockets to $199, with a 98% reduction in the marketing budget. This marketing campaign is clearly better than the contact-all-customers campaign.
33% increase in net profit and 98% reduction in marketing budget
Outcomes of the bakery's ideal marketing campaign
In this simplistic example, the key is knowing which group to pursue, which is often difficult in real-world applications. In the next case study, we use real data and make real predictions using Deep Blue AI's suite of customer response models to determine which group to pursue.
Case Study: Bank Marketing
This case study uses an open-source dataset based on marketing efforts by a financial institution in Portugal. The bank made several direct marketing efforts to 11,162 customers via phone calls to promote a term deposit. Customer demographics as well as previous marketing efforts were recorded, along with the final outcome of whether or not the customer signed up for a term deposit.
There are two key differences between this case study and the Baked Goods Inc. case study:
The data is real-world, with all the unpredictable behaviour of human beings
The cost of marketing is high - phone calls involve the time and labour of salespeople for each individual customer
The outcomes of this case study are below. Read further for details of how these outcomes can be achieved using Deep Blue AI's customer response models.
Outcomes of the bank's carefully planned marketing campaign
For this case study, 80% of the dataset is considered historical data - it is assumed that these 8,929 records exist from previous term deposit marketing campaigns. Of these customers, 47.3% signed up for the term deposit.
The remaining 2,233 phone calls to customers are presumed to be in the future - the outcomes of those marketing calls are unknown. The question the bank should be asking is,
"Based on our historical data, which of our remaining 20% of customers should we spent the time and effort marketing this product to?"
For these calculations, we may assume each phone call costs $5 to make and each term deposit makes the bank a profit of $40.
Following the first approach taken by Baked Goods Inc., the bank can choose to contact all 2,233 customers. The marketing cost is $11,165 and profit on term deposits is $42,248, making for a net profit of $31,083. Each marketing dollar spent on this campaign brings in $2.78 in profit, which is a tidy return on investment.
Alternatively, suppose the bank applied machine learning to its 8,929 historical records and developed a model that could predict to some degree of accuracy which of its customers would take up the offer and which would reject it. Applying one of Deep Blue AI's customer response models to the historical data produces the top 65% of customers most likely to respond positively to the offer - what if the bank only made phone calls to these top 65% of customers?
Comparing these names to the real known outcomes of those phone calls, we can establish that 1009 of the 1453 customers in the top 65% actually responded positively to the offer. 90% of all positive respondents are contained within the top 65% predicted by the model, so a full 35% of the customer list can be eliminated from the marketing campaign with an expected loss of only 10% of positive responses. This reduces the marketing cost to $7,265 and increases net profit to $33,095, a 6.5% improvement on the contact-all-customers approach.
Using a Deep Blue AI customer response model produces a 6.5% improvement in net profit, a 35% decrease in marketing expenditure and a 64% increase in profit per marketing dollar spent
The gain curve for this model. The large area between the blue curve and the dotted black line shows where machine learning can make your marketing dollar stretch further than random guessing.
Perhaps more importantly, the net profit per marketing dollar spent jumps from $2.78 to $4.56, a 64% increase. This means more marketing money is available to launch other campaigns and bring profit into other areas of the business.
Deep Blue AI has built a suite of customer response models that can give your marketing team the edge it needs to compete in this competitive, rapidly growing industry. Using our models, you can determine which groups of customers to target and what to target them with. Armed with this knowledge, you can make your marketing budget stretch further and get incredible outcomes from investing in a data-driven approach to marketing.