Why not simulate?

Why is my business not already using simulation? This post is a quick guide to some commonplace historical hang-ups and where they’ve left us…

Simulations are exceptionally powerful tools for optimizing systems, guarding against risks, and ultimately for making better-decisions. In spite of this, adoption remains a fraction of what it could be, and many organizations fail to utilize simulation at all.

The lowest-hanging fruit within data science and decision-support has, for a long time, been simple regression analysis and anomaly detection. Neither of these require any form of ‘simulation’, and both may yield relatively high-payoffs (or cost savings) when applied properly.

As companies look to increase the sophistication of their data-driven operations and ask where future value may lie, simulation offers the next-best bet.

Why, though, hasn’t simulation taken off to date?

Historically expensive to build, as well as time-consuming and costly to maintain, “digital twins” of organizations, their people, and their processes have been out of reach for all but the largest of players — and even then they have rarely found their way out of siloed applications.

The most common forms of simulation modeling today are:

  1. partial-implementations of things like Monte Carlo methods within software such as Excel; and
  2. highly specialist, domain-specific applications (such as Finite Element Analysis software) which carries with it a high barrier to entry, and typically lacks interoperability with other analytical frameworks.

Tooling around simulation has additionally been expensive, and hard-to-use, all but ruling out simulation modeling for everyday, general-purpose and cross-organizational use.

Modern simulation

Our mission at HASH is to eliminate information failure, and as part of that we’re working to democratize access to powerful simulation tooling through a free-to-access, vertically-integrated simulation platform.

Simulation finds itself broadly-speaking where Machine Learning was approximately ~15 years ago. Proven out in academia, utilized by defense departments of advanced nations, and adopted by quant-hedge funds — but beyond the reach of everybody else.

HASH aims to address this by being:

  • Easy to get started with — no development environment setup is required (thanks to hCore), and no DevOps experience is necessary, even when running large models in a distributed fashion (see hCloud). This means that data scientists with minimal knowledge of Python, and those within organizations who have even only rudimentary knowledge of JavaScript, can create hyper-realistic multi-million agent simulations.
  • Fast to build meaningful models with — identifying data in, and transforming data to, an “agent-mapped format” has historically been time-consuming. Through hIndex, and Flows in HASH respectively, we make this easy. hIndex additional contains community-published simulations and behaviors which can be forked and cloned for use in one’s own business, covering a wide range of supply chain, logistics, cloud computing, social distancing, and other key business concerns.
  • Free to use in most cases, and low-cost at enterprise-scale (pricing);
  • Built atop an open-source core (hEngine);
  • Reproducible and verifiable at heart, earning the trust and confidence of users who may not be familiar with, or otherwise open to, simulation-backed decision-making.
  • Production-ready: through scheduled runs, background recompute, and programmatic outputs, insights from HASH simulations can be continuously incorporated into real data-science workflows, and algorithms.

This latter point is perhaps most important. Productionizing simulation outputs has long been a manual, time-consuming process. Outputs of simulations have, at best, sometimes been seen as mere “cool visualizations”. With HASH, simulations of all kinds (including system dynamic and agent-based models) can become first-class citizens within data scientists’ toolkits at last.