Practical AI @ Flatiron
Ensuring your AI investments deliver high returns
Flatiron takes an extreme approach to making sure that any money you invest in AI pays off.
Good AI Projects
Good AI projects have a clear large benefit, a safe mechanism of deployment, and (usually) multiple possible solution approaches that all allow for incremental iteration. What we want to avoid is AI for AI’s sake, and also money pit exploratory projects that never get anywhere.
.png)
Our Approach
We’ll sometimes be the loudest voice pushing back on AI inquiries. “Right now, you’ll see higher payoff from doing <x> base work with your money instead of this AI work.”
AI Project Benefits
With that said, a well-structured and highly feasible AI project can deliver huge
returns, particularly at scale. Typical AI project structures that we like are:
returns, particularly at scale. Typical AI project structures that we like are:
01
Deliver better, faster, or cheaper by partially automating a human employee’s work, decisions, information collection, or text composition. The human employee usually still checks automated output, allowing for safely iterating quickly.
02
More accurate, more consistent, and quick decision making
03
Customer delight through high-upside features that have
little downside if mistakes are made
little downside if mistakes are made
04
Experience customization
.svg)
Explore Practical AI
AI Techniques
The word “AI” can mean so many things, and we simplify for our clients by putting
more techniques into this category than is strictly correct.
more techniques into this category than is strictly correct.
Computer Vision
3D/2D NNs/GANs;
3D/2D traditional
3D/2D traditional
Language Processing (LLMs, NLP)
LLMs (like ChatGPT),
Natural Language processing (traditional)
Natural Language processing (traditional)
Machine Learning (ML)
Supervised, Unsupervised,
Reinforcement
Reinforcement
Advanced Methods,
Data Science
Optimization, Dynamic
Programming, Advanced Experimentation, etc
Programming, Advanced Experimentation, etc
Process Notes
01
Structuring the problem is worth taking extra time to do correctly
01
For more exploratory problems with less established solutions we like to outline at least two approaches to solving the problem, one traditional. We set quick deadlines to get first crude signals for each.
01
For most problems there is usually a potential approach that involves cleverly/artistically stitching together highly developed tools/models, albeit in an unusual manner. We always look for these.
01
Oftentimes business knowledge is absolutely crucial to success. The right “debugging team” can make or break the success of a project, and must include: practiced technical expertise, business domain experts with analytical leanings, a project lead that can bridge between the two groups but also shrewdly set cutoffs.
