Why didn’t I think of that? Sometimes we get caught up in our day-to-day lives and don’t stop to see if we are solving the right, bigger picture problems. That’s how I felt when I took a deeper look at Aible – Gartner 2019 Cool Vendor and Forrester New Wave Automation-Focused Machine Learning Strong Performer. If you’re headed to Tableau Conference, check out Aible at booth #424. If not, please join me on November 21st for an overview.
AI that speaks business and data science language
In addition to automating manual machine learning, Aible applied optimization and economics techniques, flipped the data science process, and uniquely designed a business-guided experience for building AI. Most vendors offer solutions that require you to learn the science of AI and how to interpret results rather than build AI that speaks the language of business. Instead of only showing true positives or false negatives, Log Loss, ROC curves, and other data science accuracy metrics, Aible results are also shown in terms the business uses ie. missed sales, the value of a churned customer.
With Aible, citizen data scientists can build AI that respects costs, benefits, and constraints to optimize business impact.
Build AI to Optimize Business Impact
Starting with a library of AI use case blueprints to one-click app integrations, Aible simplifies the AI model building process while still enforcing governance, compliance, transparency, and auditability.
After a blueprint gets populated, Aible automatically prepares your data, runs quality checks, performs feature engineering, and evaluates many machine learning models using popular open-source algorithms such as TensorFlow, GBM, H2O.ai, Spark, Scikit-learn and others. Not only do you get to see the business-ready results, Aible also generates Python code in collections of Jupyter Notebooks. Thus, data scientists can see exactly what was done by Aible in their language.
The Model Selection area in Aible includes the following information to help you decide which model to select for your business requirements:
- Current State: Shows business impact improvement or deterioration over the Current State. This is how much better your business can be with AI.
- Impact: Shows total business impact delivered by the selected AI and other evaluated models. Shows business impact improvement or deterioration comparisons between models.
- ROI Chart: Shows the performance of the models Aible evaluated by predicted business impact and data science metrics.
- Operating Resources: Shows what percentage of operating resources can be reallocated based on the selected AI.
- Prediction Drivers: Shows what features and values of your data most influence an outcome.
- Results: Shows business impact value differences between top models.
- What-If: Allows you to run simulations with different feature values to get predictions.
Solving the right problems
Over the past two decades, I’ve been trying to democratize predictive analytics using traditional data science methodology. In all that time, I never questioned the approach.
I spent the past eight years integrating predictions from data science platforms into apps and reporting platforms to improve AI adoption. I used accepted data science best practices, with and without automated machine learning (AutoML), to speed up the process. Some organizations succeeded. Most failed. Universally, the science of data science confused the business.
Many groups don’t recognize the disconnect between data science and business realities.
All businesses operate with resource constraints. Be sure you are solving the right problem. Machine learning models trained for accuracy alone may hurt your business. A highly accurate model that confidently recommends targeting fewer prospects than you should pursue or selling items below cost to maximize sales would reduce profits. There are countless other horror stories of AI gone wrong.
Expert data scientists know there are different ways to find the best models to improve business outcomes. Without expert help, the business cannot create or get value from AI successfully. AI project failure rates cited by top industry analysts between 50% to 85% in 2019 confirm there is a big problem that had not yet been solved.
Times are changing. New approaches and innovation are lowering the barriers to rapidly get value from AI successfully.
Aible solved many common citizen data science challenges. With Aible, citizen data scientists can finally give AI a try without the steep learning curve. That’s exciting. Much like the traditional to self-service BI evolution, I expect more collaboration with experts, more governance, and more innovation to come. Let’s see what happens next.
For More Information
I barely had time tonight to walk through Aible in any depth. There is so much more to explore. If you’d like to learn about Aible or get a free trial to test it, please check out the following recommended resources. If you are at Tableau Conference this week, please stop by booth #424 and say hi.