Salesforce isn’t even waiting for Dreamforce to begin the drumbeat over its AI offering which is or will be called Einstein. There is so much to discuss over this turn of events that it’s hard to begin so rather than starting at a conventional jumping off point let’s think for a moment about the name.
You couldn’t have lived at any point in the 20th century and not have some idea of who Albert Einstein was. For most of that time he was regarded as special, a savant, one who could see things that no one else could. But I prefer to think of him as one of the first to build models of reality that he could test his ideas against. History is full of these people most of whom did a one off. But Einstein systematized the approach—he had to because he was a theoretical physicist.
Einstein may have invented the word “thought-experiment” because his work centered on the very small and the very large, things that happened in nanoseconds and things that have taken up the 13 or so billion years of the universe’s life and the billions more to come. Things you couldn’t easily see or handle.
In theoretical physics you don’t set up an experiment in a lab. You look for evidence in nature that you or others discover because your alternative is to build an impossibly expensive collider to smash nuclei together to simulate the Big Bang. You also do a lot of thinking because your imagination is the best modeling tool.
Perhaps Einstein’s greatest modeling effort was to imagine what reality would be like to ride a beam of light. That simple idea led to understanding that the speed of light is the ultimate speed limit in the universe and that if nothing can go faster then all communication is ultimately limited by the speed of light. This includes time or the perception of time. Traveling at or very close to the speed of light causes time to slow down. Einstein did all of this with a few equations and his mind. He was quite a modeler.
In this context I think Einstein makes perfect sense for an AI product name because it enables a business to model the salient points of its reality in dealing with customers and helps us all to make logical deductions that we might have missed with prior brute force methods. Salesforce’s Einstein brings together several types of AI/BI/machine learning/deep learning/machine intelligence and whatever from prior acquisitions including RelateIQ, Implisit, PredictionIO, Tempo, and others.
Here is why I think this is significant. AI of this caliber significantly signals the end of transaction oriented business and conventional CRM that supported it. The better the model of your business the more you can see into the processes you are involved in and importantly the less you’ll focus on any transaction because transactions will be a foregone conclusion. You’ll focus on the process that leads to a transaction, which will help to assure a transaction takes place.
As a practical matter this doesn’t mean you’ll suddenly have superstar sales people who nail every deal. But having good AI might tell you when you’re wasting your time, when a deal isn’t going to be yours. Knowing this you’ll redeploy resources with the confidence that you’ve made the right decision.
Don’t worry this isn’t only about selling, it’s equally important for every aspect of your business. AI in service can tell you not only the next best action or offer, but it can also tell you the customer’s likelihood of being satisfied with the engagement and therefore what else you need to do. AI in marketing has many opportunities but let’s look at one that seems to always get neglected, the installed customer base.
Too many businesses assume that customers will come back for more and many do, but others keep looking for better opportunities. Worse, too often businesses that do big deals with multiple tranches of purchases spelled out in agreements forget where they are in the life-cycle and leave money on the table. There’s no easier way to make your numbers than to sell to existing customers but for some reason, we neglect the low-hanging fruit. AI could be an effective agent for changing that situation.
I don’t know all of Salesforce’s plans for Einstein, although they are a client and they brief me from time to time, it’s a big kimono. But with this early signal, I expect that Dreamforce 16 will have a distinctly AI flavor. That’s not very surprising and the company has been signaling in that direction for a while without making announcements like this. What will be surprising will likely be the many different applications of AI they’ll find and the timeline for releasing it all.
There’s lots of talk pro and con these days about bots, AI, and intelligent assistants. A lot of this talk is not necessarily new; it’s been percolating around the industry for decades. Vinnie Mirchandani, a friend and truly gifted analyst, wrote a book recently, Silicon Collar that accepts that this automation might be eliminating jobs but optimistically holds out for the silver lining. Mirchandani firmly believes and documents how businesses and individuals are taking advantage of an opportunity to build new human mediated processes (and jobs) that leverage intelligent systems.
Another friend, Esteban Kolsky who is also a gifted analyst, says not so fast. Like many of us Kolsky has seen this movie before. He points out that adoption has been painfully slow—so slow in fact that AI fails what I call the Gates Test. You might recall that Bill Gates once said that we over-estimate what we can do in two years and underestimate what we can do in ten. Indeed, the gestation period for an overnight success seems to be ten years these days.
But as Kolsky points out in a recent post, “The latest survey, to be shared at Dreamforce 2016 and published soon thereafter, says that from the single digits in adoption they enjoyed for the entire 2002-2012 decade we are seeing adoption nearing 15% now for automated bots and intelligent assistants.” Slow indeed. What’s been holding things back has been a lack of two things: 1) not enough computing power and 2) a clear need.
We can all think up scenarios where a little help from something with AI embedded might be good but on closer inspection we realize there are other ways to get the job done. AI is a heavy lift, or at least it was once. Back when the working models of AI were set down, computing power was not up to the job but really fast processors, multiple cores, flash memory, and the cloud have made it possible to concentrate the power needed to drive AI. But this still leaves us with finding a clear need.
I offer the following analogy: we live in a spreadsheet-dominated world with a linear mindset but we are moving to a world where the lines are anything but straight. To make sense of curved lines you need calculus. It’s calculus, especially the integral variety that tells us what’s going on in a process that has plenty of funky ups and downs. In the spreadsheet era, which I firmly believe is ending or at least transitioning, we searched for averages and made straight-line derivatives from them.
This led to some dumb ideas like calculating what an average deal is and trying to fit all deals into it as if it were a straight-jacket. It also harkens back to the statistical awakening in the 19th century when the term “average man” first came into use. The average man is a fiction but a highly usable one that gives us a basis for modeling.
But when you go for an average you have to ignore some profitable outliers or other things that don’t fit your model. In the age of business by transaction, the straight-line model was good enough. Nonetheless over most of this century so far, we’ve seen that model become less effective as the vendor-customer relationship moved toward the micro-transactions of subscriptions. A straight-line model doesn’t work very well in subscriptions because at a micro level all transactions look the same. It’s only when you expand your view that you can see the micro-transactions that show trends that might be good or bad. As a result we’ve been left without a model.
A model for the vendor-customer relationship that works involves calculus, at least at the metaphorical level. Calculus gives us the flexibility to model many variables involving customer demographics, purchase history, life-cycle stage, and of course the transaction before us.
I think many people in business have a working appreciation of all this, though they are certainly still in the minority and this is where AI comes in because I see its algorithms as calculus in a box. AI gives the average businessperson who has no interest in calculus, or who might have studied it decades ago, the ability to apply more sophisticated modeling to increasingly complex business.
So this is a long-winded attempt to say that at last we have a clear need for AI as well as the horsepower to run it. The need is all around us and if you’ve ever caught yourself wondering at how sophisticated business and our supporting systems have become, you’ll likely be grateful that there’s a new weapon in the arms race.
Modeling is a big idea and one that I started noodling on many years ago. In the last couple of weeks I’ve looked at the cloud computing model, or what it ought to be. Today I am looking at a broader paradigm. One of the great things about new paradigms is that there is no model per se, in a way a new paradigm is an opportunity to create the model; it’s an incipient model perhaps.
The most interesting models or modeling I can think of are those that take place in four dimensions. You will recall without much prompting that we live in four dimensions, time plus the three that define space and from this we get Einstein’s space-time. People who study the big bang speak of matter, space and time as “condensing” out of the event. That choice of word has always fascinated me.
There might even be more dimensions that we don’t know about or comprehend. How can that be? I don’t know the only analogy I can make is that my dog lives in the same four (or more) dimensions as I do but he’s only apparently aware of the ever present now.
Four-dimensional modeling is not hard to comprehend, but like a dog, we routinely fall back a dimension to deal with reality, especially in business. Let’s use the analogy of riding a bike. The bicycle stands against a wall in four dimensions like everything else though it can be comprehended in three very easily. But riding the thing is definitely a four dimensional experience. You can’t ride a bike unless you make a conscious effort to go through space-time balanced on those two wheels.
Learning to ride a bike requires modeling — either training wheels or the expert hand of an adult or perhaps an older sibling who models for you the feeling of keeping your balance. No amount of discussion before hand is very instructive for a first time rider because words fail to communicate the feeling in your stomach as you take your first ride.
In business our models are typically three-dimensional. The two that make the most interesting contrast for me are the accounts receivable (AR) report and the sales forecast. The AR report tells you in a two dimensional grid about what was done in the one dimensional past. It gives you an accurate representation of what is owed and what will come in barring some future four-dimensional miscue.
The sales forecast is much different. We treat is like the AR report but with far different expectations. The forecast is purely four-dimensional but we insist on treating it like the AR report which, though not perfect, is physics compared to the sociology of the forecast.
My point, and perhaps it is not a big one, is that good forecasting requires a four-dimensional model and that can’t be done with a report. You might disagree and the evidence is on your side, mostly. For decades we’ve used a standard forecast report to predict future revenues but honestly, it’s been far less than satisfying. We get our forecasts wrong quite a bit. My data invariably shows that sales forecasts rarely have a ninety percent confidence level. It’s a narrow range — fifty percent is as good as a dartboard so there isn’t a lot of room to work in. We complain about forecast accuracy with the same frequency we complain about the weather, but as Mark Twain wryly observed about the weather, we never do anything about it.
That standard sales forecasting via reports and spreadsheets has worked so well for so long is not a tribute to the method but an artifact of times when we were able to sell standardized products into huge markets. Demand was usually sufficient to backfill one opportunity with another when necessary.
But today’s markets are a bit different. We are doing far more cross- and up-selling than ever. Product lines are expanding but categories are relatively stagnant. Getting the forecast right has never been more important because margins are smaller and there are fewer deals with which to backfill.
The solution for the sociology of the forecast might be the same as the solution for the weather — a model moderated by computer processing power. Rather than focusing on a single report that amounts to a three-D snapshot in time, we need two things. First, a model that captures past information and integrates it into the present but also the model has to offer enough predictive value from prior experience to enable us to self-correct and avoid a crash.
A child (or a trained circus animal for that matter) on a bike can do this and it should not be terribly difficult for us big-brained adult humans to do so with a forecast. I have been impressed with the advances made by analytics companies in this area in the last few years though they often discuss everything in terms of analytics. I would rather they approach this in terms of a model or riding a bike though. Perhaps that would make the idea of using analytics less daunting.