How to Tell What’s Real
Have you tried the new “AI Latteccino” at Starbucks? Absurd—and not real, of course. But don’t be surprised if food and beverage companies become the latest vendors to co-opt the Artificial Intelligence mantle. Nearly every technology company claims that AI is “at the core” of their offerings. Actually, few use anything resembling true AI. By repackaging classical statistics and data analytics, many companies add useful functionality but fail to produce the revolutionary value that real AI can deliver.
In his July, 2018 article in Fortune, Arif Janmohamed of Lightspeed Venture Partners cautions, “A number of SaaS and automation companies out there are positioning themselves under the AI banner, even though all they really do is use data analytics to orchestrate applications and workflows. The technology doesn’t get more intelligent over time, and it never reaches the level of autonomy of bona fide AI.”
AI and Machine Learning
Artificial Intelligence defines computer systems patterned after human intelligence in their ability to learn and recognize—so that previously unseen information can be acted on in ways that produce useful results. This is distinctly different from systems that merely organize, report, and help visualize what has already happened. As interesting as insights may be, reporting-based analytics cannot make the leap to project those observations onto new, never-before-seen data.
Machine Learning is the subset of AI that solves complicated analytical problems without needing to be explicitly programmed for each task. By learning patterns in data through sophisticated AI algorithms, Machine Learning can build accurate models and apply them to brand new data. This is the breakthrough that allows formerly impossible analyses to be made practically and quickly.
“Explicit programming” is slightly misleading in the definition of machine learning, because it does not necessarily mean “no programming.” It refers only to the general usefulness of analytical algorithms without regard to the data being targeted. Until recently, managing setup, loading, processing, and output of analytical results with machine learning has required plenty of programming—even learning new languages expressly created for those purposes. Fortunately, Squark has removed this final obstacle to rapid, productive use of Machine Learning. Squark is completely codeless Automated Machine Learning (AutoML).
Why Is AI Better?
Artificial Intelligence in the form of AutoML solves analytical problems faster and with more concrete results. AI has the ability to build models from known outcomes and to project these patterns onto new, future data. It really does work. Predictive value of AI is proven to be faster and far more useful than conventional, statistical analysis.
AutoML gives businesses their first, real tool for augmenting intuition with data science. Finally, we there are ways to apply imagination with confidence. AutoML frees organizations to do what they do best; what they really want to do in place of system wrangling. Marketing, for example, is the hottest application for AutoML right now because results are easy to monetize. Marketers become free to market. Armed with predictive knowledge, actions are easier to implement.
- Which visitors are likely to abandon carts?
- Which ads work best in which media?
- What leads will produce the highest customer lifetime value?
- Are there cross-sell/up-sell offers that will boost revenue?
- What digital marketing actually accelerates customer journeys?
It isn’t magic. AutoML is merely automation of classical AI computational methods within an online application made for business users. Just like hand-made AI, AutoML can find the patterns and associations that have always been present in clients’ data.
How AutoML Works
Squark uses an AI process called Supervised Learning, where algorithms learn from sets of known outcomes and build models that can be applied to new, never before seen data. Like a bloodhound getting the scent, it can pick out data that foretells similar outcomes—clues that cannot be deciphered with conventional programming logic.
How it accomplishes this is the stuff of arcane algorithm builders. That is does it is provable. By trying, testing, refining, and testing again—perhaps thousands of times—Squark builds remarkably accurate models that can be measured against hold-out data. If a model can predict accurately on data that it has never seen, but which we know to be true, then it is proven to be generally useful for future predictions. That is exactly what happens in the Squark AutoML process.
Squark takes a cold, unbiased look at all the data and builds accurate models no matter how many elements there are. There are no formulas to build and maintain. The data tells its own story. Acting on the latest data, AutoML automatically takes the newest information into account and disregards factors that no longer predict success. It isn’t magic or mystery.
To train itself, Squark ingests sets of data that represent known outcomes, such as lead tables for the previous period. Lead tables typically show the basics—whether or not a lead converted to a different stage, became an opportunity, or closed won or lost. They may also contain many more values related variously to industry, company size, competitive status, time and date, inbound and outbound activity, or hundreds of other traits. Clicking a column in the table tells the AutoML which variable you want to predict. Will the lead convert from marketing qualified to sales qualified? Is likely customer lifetime value forecast higher or lower than target? Which one of five possible messages will elicit a response?
The next step is to link the new data, the leads on which you want to make predictions. Squark learns the patterns and builds scores of models, testing them against one another to find the best. Running the new leads through the winning the model then happens quickly—minutes, typically. Output is a table with predictions appended, including probabilities for each record. Now you have a prioritized list of prospects that is richer than any scoring formula could provide.
If you have delved into the world of AI data science already, you know that they really is complicated. Here is a brief summary of some of the complexity that is abstracted by Squark. Those who understand technical details can see that advanced AI data science is being employed to prepare data and tune performance. Those unfamiliar with the terminology will be pleased to know they do not have to learn it.
Squark does much of what an experienced data scientist would do to make sure data is structured for predictions and that the system is adjusted for the peculiarities of unique data sets, including:
- Data Preparation – Cleaning and normalizing data columns (features).
- Feature Extraction – e.g. Turning a single date/time feature into year, month, day, day of week, and time to make infirmed predictions.
- Feature Selection – Identifying features that are most predictive of outcomes.
- Algorithm Selection – Applying particular AI algorithms that are suited to making a prediction on particular data.
- Hyperparameter Optimization – Tuning the algorithm execution process for efficiency and preventing runaway calculation.
- Cross-validation – testing algorithms’ predictions against a hold-out set of know outcomes to validate accuracy.
- Leakage Detection – ensuring that data sets do not contain the values to be predicted, which would invalidate results.
It is less important to understand the details of how these data prep capabilities work than to be certain they are working on your behalf. If your data is imbalanced—meaning you don’t have a lot of outcomes from which to learn—AutoML can compensate for that. If you think your data is dirty or too sparse, or has missing values, AutoML has ways to detail and prepare your data for use. Remember that the goal is to produce reasonable predictions that are more accurate than guessing and timely enough to be useful.