What is Project Bonsai?

Microsoft Project Bonsai is an AI toolchain which can be used to control and optimize production systems. It uses state-of-the-art techniques by combining deep reinforcement learning (DRL) and machine teaching to develop AI agents (or Bonsai “brains”). These AI agents can provide operators or control systems with the best decisions to make to achieve an optimization goal.

These solutions are best suited to the operations of complex industrial systems where existing controllers or human interventions are insufficient, inefficient, or unsustainable. Bonsai can help manufacturers overcome everyday operational challenges and complexities such as variation in inputs, machine performance degradation, and differences in operator expertise, to name a few.

To learn more about Project Bonsai and Autonomous Systems, check out our 101 guide here.

How do businesses get started with Project Bonsai?

Working with a Microsoft Partner that has deep expertise with the Bonsai toolchain is a great place to start. Neal Analytics works with manufacturing clients to train, develop, and deploy Bonsai solutions.

We take an end-to-end lifecycle approach to these projects, which creates three distinct phases for Bonsai engagements:

1. Pre-Engagement
2. Delivery and Real-World Trials
3. Scaling

Phase 1: Bonsai Pre-Engagement

The Bonsai Pre-Engagement process, or Phase 1, is a rigorous multi step process that can take between 6 to 8 weeks to complete. Multiple checkpoints are included in the process to assist all parties in determining the applicability, viability, and feasibility of a Bonsai autonomous system offering.

In this phase of the engagement, Neal Analytics will work with the customer to…

• Identify and prioritize value-creating use cases
• Work with key stakeholders and SMEs to map current processes
• Design an initial Bonsai brain based on current strategies and environment
• Peer-review the Brain design with Neal Practitioners and Microsoft SMEs
• Draft a proposal that includes scope, estimated timelines, and required resources

Step 1: Identify value

In the early stages of the process, Neal Analytics introduces the Bonsai offering and partners with clients to identify and prioritize potential use cases. A winning use case will generally be at the intersection of the following 3 criteria:

• High ROI – a single digit improvement will be worth more than a $1M
• Further optimization is required – optimization techniques have been exhausted
• Time to value – looking to gain value in the near term

Step 2: Discovery

The subsequent step, Discovery, focuses on working with key stakeholders and SMEs (process engineers, operation managers, operators) to refine our team’s understanding of the issue and map the current process(es).

Step 3: Design

In the Design step, we then do a deeper dive of the current process to surface up current control strategies and common environment states. These will serve as key inputs to our initial Brain design.

Step 4: Peer Review

In the Peer Review step, the Brain design undergoes a rigorous review where concepts and their underlying assumptions are validated against expert Bonsai community (Neal Practitioners and Microsoft SMEs).

Step 5: Proposal and presentation

During the Proposal step, we refine the scope and estimate timelines and resource needs. These are supported by a roadmap that highlights several proof points along the way. These artifacts are then packaged and presented to the clients

Transitioning to Phase 2

The final step will kick start once all approvals have been attained. This is where we transition into Phase 2: Delivery and Real-World Trials.

To learn more about real-world Bonsai solutions, check out our work with PepsiCo where the Neal team helped optimize the extrusion process for Cheetos production.