Getting started with Bonsai: Phase 3 – Scaling
In my last article, I showed you what happens in the second phase of a Bonsai project: Delivery and real-world trials. This included building a simulated environment, training, and validating the AI agent (or Bonsai brain).
Now you may be wondering: What comes next?
As the first Bonsai solution stabilizes, organizations will logically expect greater benefits from scaling their solution. In fact, there hasn’t been a single engagement where we weren’t asked the following two questions:
- How do we quantify the benefits of scaling?
- Can we scale this solution?
To scale or not to scale
To address the first question, we put together a simple infographic that looks at this in a slightly different way: What’s at stake if you do nothing? This could apply to a single plant or deploying the same solution in multiple locations.
While this is a simplistic formula (and by no means exhaustive), we have found it to be extremely effective in helping our clients understand the value they can capture by scaling their Bonsai solution.
Scaling and the multiplier effect
Here, scaling multiplies the value of reductions in operational and opportunity costs. Operational cost can include indirect (training, recruitment, etc.) and direct (wages, fuel, power) costs, as well as costs borne by poor performing production lines (machine or human generated). Opportunity cost is measured as unmet demand due to varying factors such as skilled labor shortage or rapidly changing conditions requiring just-in-time (JIT) strategic/market adjustments.
The multiplier is self-explanatory. It reflects the number of lines that would make great candidates for the piloted Bonsai brain.
Which brings us to the next question: Can we deploy it to other lines?
Deploying Bonsai to other lines
We have developed a framework to help clients navigate through this question. It boils down to identifying and addressing the sources of variation that exist within candidate lines.
These sources of variation can be grouped into three main buckets:
- Critical To Quality –> Variations in product specification such as product diameter, density, etc.
- System Dynamics –> Variations in equipment such as sensor types, equipment, etc.
- Control Strategies –> Variations in control strategies (if this happens, then do this) required for either specific products or equipment
Over the past few years, we have generated a number of practical steps (for example, leveraging dynamics randomization to train more generalizable agents) to help achieve economies of scale with this solution. We encourage you to contact us to discuss how your organization can benefit from Microsoft’s Bonsai platform.