Aible Advanced launches as startup aims to scale AI with business sense

AI models are trained for accuracy not business results. Aible is recruiting data scientists and developers to change that situation.
Written by Larry Dignan, Contributor

Aible, a startup focused on real-world artificial intelligence that takes into account business variables and constraints, launched Aible Advanced, an automated machine learning platform for data scientists and developers.

With the move, Aible, which has a version for business users, is aiming to automate the machine learning process (AutoML) and enable data scientists and developers to connect to enterprise systems that house critical data.

Arijit Sengupta, CEO of Aible, said the idea behind Aible Advanced is to give data scientists and developers the ability to think through models and how to link back to enterprise systems. "The missing piece is the link between enterprise applications, AI and how often predictions actually work," he said.

Sengupta, one of the lead executives behind Salesforce's Einstein, started Aible based on the idea that AI had become too theoretical to be of business use.

Primers: What is AI? | What is machine learning? |What is deep learning? | What is artificial general intelligence?    

"I've been in the space of AI for more than 20 years and what I finally realized is that AI is trained on the wrong thing. It's trained for accuracy, but became too theoretical. What are the business implications?" he asked. "There's no way to teach business users how AI works, but you can teach AI to understand business, cost benefits, business problems and constrains in a connected way."

That business-centric approach to AI led to Aible's launch in March. The launch of Aible meant business users could enter business objectives and constraints and then create a prediction model designed to maximize business impact. Traditional AI would focus on predicting three people out of 10 will buy a product. The model could be accurate, but leave money on the table if a business really needed 50 leads to convert, explained Sengupta.

Also: From low code and cloud, to AI and encryption: What you do with data needs to be about more than buzzwords | The data scientist job interview: Questions to expect and questions to ask (free PDF)

While Aible's business model revolves around software as a service, Sengupta said the company also will take a cut of the money attributed to the models, say 10% to 15% of revenue, in some cases. That results as a service approach would be interesting to watch develop. Sengupta said the model is possible because Aible has built-in monitoring of performance.

Sengupta added that it has some large enterprises on the revenue share model and it tends to appeal to business executives, but procurement pushes back.


In theory, Aible can create models for individual salespeople based on their styles and aggressiveness. Aible's approach has garnered attention from research firms such as Forrester and Gartner. Gartner recently published a case study on Merrow, a manufacturing and distribution company, that used Aible to maximize its market opportunities. 

Sengupta said that Aible Advanced will use 10 steps to generate models, double other AutoML tools. The steps include:

  • Requirement Gathering where AutoML learns the business objectives, realities, and use cases.
  • Blueprints where good predictors are surfaced.
  • Data Recipes to gather the training data from existing enterprise applications and data repositories.
  • Data Enhancement to clean and prepare data for machine learning and creates derived variables.
  • Model Customization to ensure the model is trained to maximize the actual business objectives and respect the business constraints.
  • Hyperparameter Tuning to train various types of models and try out settings.
  • Model Selection to recommend the model that best optimizes the business objectives and conducts evaluations like sensitivity, what-if and scenario analysis.
  • Model Deployment where it starts running the model in the customer's environment.
  • Prediction Writeback where it writes predictions back to the enterprise applications and business users.
  • Monitoring where it observes the actual business outcomes and compares it to the predicted outcomes for retraining.

The big takeaway is that the model training revolves around those first two items. Business reality usually isn't considered.

To get the word out about Aible Advanced, Aible is launching a 30-minute AutoML challenge. In less than 30 minutes, Aible will create a predictive model based on a user's proprietary data in their own AWS account. If Aible fails, users keep the model free of charge. 

Related stories:

Editorial standards