Big data, predictive analytics, artificial intelligence (AI) and machine learning are all the rage these days. They are sure to be a popular topic of hallway conversations at this October's Professional Pricing Society Fall Conference in San Diego (Oct. 24 to 27, 2017). The theme for the conference is 'Data, Change Management, and the Profit Landscape' and there are a number of interesting sessions on these themes on the agenda.
Ibbaka is conducting its own research and development on applications of AI to pricing topics. As we believe that good pricing is built on a value-based market segmentation, we are beginning with finding ways to use AI to work more effectively and efficiently and to move towards dynamic market segmentation. More on this later.
The term 'AI' is used in many different ways these days, so it may help to have a simple taxonomy of what we mean when we use the term. Basically, AI is a family of loosely related techniques to build computers (hardware and software) that can reason about information, identify new patterns that were not programmed in, and recommend actions. The larger vision for AI (strong AI), around creating machines that reason and behave in ways that imitate and then surpass humans is perhaps too large a vision for the day-to-day work of understanding how to create value, set prices and to get those prices in a competitive marketplace. People engaged in pricing work are more concerned with weak AI, which has more modest ambitions.
There are a number of different techniques being used in AI: symbolic AI (concerned with the manipulation of symbols), evolutionary computing (where the codebase evolves), probabilistic (including the use of Bayesian models of causality), and machine learning (neural networks applying deep learning). The latter, deep learning, is the flavor of the week (or more accurately the flavor of the decade) and many of the applications getting press apply some version of deep learning. Popular open source platforms like TensorFlow are implementations of deep learning and are a very good place to start one's own learning of these techniques.
In the pricing world, we deal with complex, fast-evolving problems. AI is going to be an important part of how we address future challenges.
The big metal pricing platforms like PROS and Vendavo have been implementing AI for many years. See for example, 'PROS Espouses Its Artificial Intelligence (AI) Strategy to Power Modern Commerce.' or 'How Business Benefits From Artificial Intelligence.' Newer companies like Price f(x) have names resonant of AI though it is not clear how advanced their AI systems actually are.
Of course, the big tech companies are all very active in this space, through their own offerings, investments, and partnerships. A good snapshot of where things are as of October 2017 can be found in this CB Insights article on Big Tech in AI.
Ibbaka is focussed on the pricing of Industrial Internet of Things solutions and B2B SaaS. This colors our thinking.
The first question - What applications?
The first question one may have about the application of AI to pricing is "Where shall we apply these new techniques first?" "What are the low hanging fruit?" The steps you need to take to find the places you should be exploring AI are straightforward.
- Map your current pricing methodology.
- Look for places where you use or could use data.
- See where do you need to discover patterns.
- Ask "What do you want to predict?"
Ibbaka is often dealing with early-stage innovation, where there is not a huge amount of transactional data to leverage. Our current process is as follows (it is not quite as linear as this suggest, but you get the idea):
- Hypothesize value drivers (emotional, economic, social, clinical)
- Conduct surveys and interviews to test these value drivers, discover new ones, discard those that do not resonate
- Model the buying process and decision making units
- Define a set of value drivers and buying processes
- Build a test set of potential customers (this is not the current customers, one wants to consider a much larger potential set)
- Generate multiple segmentation axes, built alternative segmentation frames, test these against the test set of potential customers
- Estimate LTV and CAC for each segment (these are two of the key SaaS metrics, LTV is the Lifetime Value of a Customer and CAC are the Customer Acquisition Costs)
- Select and sequence target segments
- Build buyer persona for target segments
- Generate pricing metrics that map to one or more of the value metrics (your pricing metric is how you price your offer; your value metric is the unit of consumption by which the user gets value)
- Use the pricing metrics and value metrics to build the pricing architecture
- Define the role of each part of your pricing architecture
- Design the data models that will track usage, value and pricing
- Set the initial prices
- Track market response and ask (i) Is each part of the pricing architecture playing its intended role, (ii) how is the market responding (track the funnel), (iii) what actions are competitors taking
- Develop predictive analytics based on the interactions of the usage, value and pricing models
All of this is done in the context of the company's strategy and financing options. Pricing models need to support general go-to-market strategies and be able to deliver the target SaaS metrics.
At Ibbaka, our initial explorations of AI are focussed on market segmentation. Market segmentation is the foundation of all pricing strategy, it is data dependent, and the underlying patterns can be hard for humans to discern.
Other companies, focussed more on transactional data, will concentrate on price optimization or demand forecasting.
Eventually, these come together as the connections between value, use and price become clear. This will be a huge opportunity for AI and will be the foundation of a predictive science.
The second question - What data?
All AI solutions, and especially modern deep learning solutions, depend on data. Preferably lots of data. What data is needed to explore AI? The honest answer is 'nobody knows yet.' A wide range of different internal and external data could be relevant and there is no general rule that will answer this question for you.
Cast a wide net.
Internally, gather as much data as possible about use, from the pipeline and from your customer support applications. All of this data may be relevant and help you find critical patterns or make predictions.
Externally, gather as much data as you can about how your potential customers (all your potential customers, not just your current customers or even prospects), how they create value for their customers, and the input factors that determine their decisions.
Yes, this is a lot of data, and it will be hard to gather and organize. But at least it will be cheap to store in today's world.
The more data you gather the more likely you are to find the patterns that will give you insights.
The third question - What data rights are needed?
Any time we start talking about data, we need to think about data rights. This is a murky subject today. Most companies have preferred to ignore it or to bury it in obscure language deep in subscription or license agreements. This will not fly in the future. Start building a data rights model that makes it clear what rights you have to collect and use data and how you will inform users about the data you are using.
Companies that fail to do this are inviting legal problems and will find that they are limiting their opportunities to develop new data-based monetization strategies. Pricing needs a seat on your data and privacy council. In many cases, pricing people will lead need to lead the discussion as we have the most insight into the value that the data can create.
Data rights even apply to data you collect on usage of your own applications. This data is critical to your future pricing technologies. Smart companies will lead and not be pushed by regulators or customers.
The fourth question - What skills will be needed?
The new world of AI, deep learning, big data and predictive analytics is going to require new skill sets for pricing experts. All of us will need to develop a working understanding of these concepts. It is a lot to learn.
We will also need to become experts on data rights. This is too important a topic to leave to lawyers and data privacy officers. One can begin by familiarizing oneself with frameworks like Creative Commons. Something like this will emerge to cover data rights over the next decade.
Pessimists may think this will take the joy and creativity out of pricing work and reduce it to a subsector of data science. I doubt this. It will rather open new avenues for pricing innovation. Pricing experts will need to embrace innovation methodologies like design thinking (see "Don't Set Prices. Design Pricing.").
Ibbaka's affiliate TeamFit did some research into the skills of the members of the Professional Pricing Society Group on LinkedIn (if you are not a member of this group we recommend joining). See the blog post "Skills in the LinkedIn Professional Pricing Society Group."
There is cause for concern. Based on a random sample of 400 members of this 15,000 person group, there is a real lack of the critical skills needed to develop AI-based solutions for pricing. Foundational skills needed here include data modeling, statistics, linear algebra, probability theory and so on. Languages used include R and Python. Common platforms are TensorFlow, Theano, Infer.net and so on. This area is evolving rapidly and now is the time to invest in learning.
An invitation from Ibbaka
Ibbaka is looking for a few companies who want to work with us on dynamic market segmentation. You will need to provide us with datasets that we can work from. We will provide you with proposed market segmentations and recommendations on how to price for each segment. What kind of data? We can begin with your current customers and prospects and see what new insights we can provide from this. There is a participation fee of US$19,700 which we will waive for the first three participants. If you are an early-stage company contact us. We want to support early-stage innovation and we will find an arrangement that is a win-win for both parties. Contact karen@ibbaka at (415) 799-8326 or email@example.com.