1. Accurate
An accurate recommendation yields high-scoring solutions while following agreed-upon rules and constraints.
Challenges
- Quantifying items that are difficult to quantify.
- Slaying sacred cows and threatening the status quo.
- Mixing short-term and long-term objectives.
Our Strategies
- Calibrate against existing decisions—using pairs-trading to find sensitivities.
- Find the right executive sponsorship.
- Provide ranked alternatives along an efficient frontier.
2. Clear
A clear recommendation is unambiguous, precise, and concise.
Challenges
- Users want to understand why, not just what.
- The best choice may be very complex.
- The entire solution may be too large to easily inspect and validate.
Our Strategies
- Build and install the model incrementally, solving the way employees do at first.
- Focus on user-interface design on solution validation.
- Allow the user to easily inspect and edit input data and parameters.
3. Fast
A fast recommendation is: low latency – it uses up-to-date data and gives answers quickly; and rapid throughout – it solves quickly; and scalable – it has the capacity for more.
Challenges
- As real-world, non-linear factors are added, the solving time explodes.
- Data and transaction volumes and constraints keep increasing.
- The solver must be fast enough to reflect “what if” usage.
Our Strategies
- Avoid one-size-fits-all software tools.
- Combine state-of-the-art approaches.
- Use parallel processing and spare cycles for continuous improvement of the solution.
4. Robust
A robust recommendation is adaptive as conditions change, self-detects faults and failures, and has a low bug count.
Challenges
- Market conditions may change rapidly.
- Simple heuristics often fail when conditions change.
- Data can be late, missing, or erroneous.
Our Strategies
- Make real-time data and solution validation part of the core model.
- Use self-tuning parameter techniques.
- Mix experienced optimizers in a team setting and avoid solo efforts.