Constraints stimulate creativity. During an economic crisis, your budget is a typical constraint. Not enough customers, not enough cash, not enough financial support from banks and investors. Actually, when your budget is tight, you have two options: to come up with a new creative solution through a business process redesign, or to carefully redistribute your resources, prioritize and optimize your budget allocation, and find an optimum balance. When your business is rich and fat, you can spend overspend. Life is good, so why bother to hold back? But during a crisis, every penny counts. And that’s when budget optimization happens. People only do it when something forces them to. In your case, this something is the economic crisis.
If your advertising budget is low, then your sales are low. You want to see your sales grow, but you don’t want to spend a lot on promotion. Optimization seeks to balance the tradeoffs you need to make. Speaking scientifically, in the promotion budget allocation, you want to maximize the effectiveness of your promotional efforts for a given budget. These are your objectives and constraints. Direct-mail marketing; ads in magazines, newspapers, and web; public relations activities and social network pages; radio and TV advertising; search engine marketing – these are your promotion vehicles. But how do you allocate your promotion budget across these different vehicles to maximize the number of ad impressions? In the world of analytics, we refer to situations of this type as ‘optimization problems’. Optimization seeks to find maximum and minimum values for an expression called the objective function, subject to a set of constraints.
Ironically, lack of revenue may force you to seek ways to increase your profit with the same goods in same market conditions only by optimizing the price. Why you have not done this before? How do you set the price for your products? There is a whole sophisticated methodology around pricing. In general, consumers tend to buy more when the price is low, and vice versa. We are not talking about luxury goods – we’re talking about frozen vegetables and pizza. They feel they can “stock up” at the lower prices in the short term. In the long term, they don’t continue to buy because there’s a limit to the amount of frozen vegetables they want to own, even when they’re offered at very low prices. The bottom line is that you can sell more at lower prices, and eventually gain more, through sales volume in certain market environments. The fine balance you have to find is called the ‘optimal price’. To find it, you have to collect data about your customer demand, conduct surveys, analyze past sales data, even conduct a market experiments and construct a demand curve (the number of goods consumers are ready to buy at a given price). Once you know a product or service’s demand curve, you can find its optimal price. For each point on the demand curve, you can calculate the resulting revenue and cost. With known revenue and cost, you can calculate profits – just subtract the cost from the revenue. The optimal price is the price that gives you maximum profit at lowest cost. It’s that simple.
Sorger, in ‘Marketing Analytics’, suggests conducting optimization of promotion and pricing activities using Excel Solver, a popular tool among marketers. Indeed, the tool will work, but you’ll never know whether the solution it delivers is the best possible one. If your solution space looks like a simple parabola, with just one maximum or minimum, depending on how you look at it, then it’s probably fine and you can safely do a basic optimization. However, what if your parameter space looks like the surface of an ocean and you have to find the highest wave? When you optimize the parameters of your redesigned process, you may have to adjust dozens, even hundreds of factors to improve your key performance indicators (KPI). To do that, you need two things: a process model, or a function that links your factors with your KPIs and the optimization algorithm.
There are plenty of optimization algorithms. You’ll probably never encounter most of them, and none if your software is smart enough and has only one button (‘Solve’ or ‘Optimize’). You don’t have to be a control freak, but it’s good to understand what happens behind the scenes during an optimization. There are local optimums and global optimums. Again, think of waves on an ocean surface. Some waves are high enough to suit your needs. If you’re an average surfer, these are the local optimums. But if you want the biggest wave – not because you’re desperate but because, like Mercedes, you prefer ‘the best or nothing’ – you want the global optimum. This represents the best possible combination of your process parameters, the best possible price for your product, and the best possible allocation scheme for your promotional budget.
The optimization algorithms also operate in terms of exploitation and exploration, much like the algorithms of predictive analytics. Some algorithms are good for combinatorial optimization – for example, genetic algorithms, where the variants are coded in ones and zeros and cross over and mutate like genes in biology to exploit and explore the solution space. Others are good for continuous landscapes, like the search for hypothetical waves. For aesthetic reasons only, I want to describe an algorithm called a ‘swarm’, or more precisely, a ‘particle swarm optimization’. Engelbrecht nicely covered it in his book ‘Fundamentals of Computational Swarm Intelligence’. To sum up: one fine day, a psychologist and a mathematician were watching a flock of birds fly over a city square in search for food. The birds had a social and individual component in their behavior. The social component made them fly the same direction as their neighbors and maintain the integrity of the flock. The individual component permitted some birds to deviate from the general flock direction and explore adjacent areas. If one bird saw that its neighbor had found some food, it followed and eventually changed direction of the whole flock. Swarms and shoals also demonstrate similar behavior. The mathematician managed to create an equation so general that today it is successfully used to solve complex optimization problems. It can find global optimums faster than other techniques.
To conclude this chapter, I will repeat the well-known statement: there’s always room for improvement (i.e., for optimization). So treat the constrains as an opportunity, be smart, and use analytics! Rise and shine!