Case Studies

To retain repeat customers, retail locations must monitor their happiness levels and intervene prior to the end of any unsatisfactory visit.  

Upscale / Entertainment / Casinos / Hardware Integration / Behavior anticipation

Problem statement : How do you maximize spend in a shifting casino marketplace

The casino model has evolved over time. Historically, there were big spending “whales” who generated the most revenue and casinos focused on providing the best experience to that small group. Today, the revenue is predominantly generated from average consumers, which requires flipping the model to scale and maximize the spend from this group. The ability to predict when people are having a good time as well as the duration of their visit enables humans to intervene to optimize the experience (“Hey don’t leave yet, grab some food at the buffet”). The AI is able to anticipate when a guest is going to leave with 95-98% accuracy in real time.

Screenplay development takes too long and is costly. Rapid review and feedback loops can significantly reduce cost and time.

Upskill / Startup company / Entertainment industry / LLM

Create LLM-based categorization modeling of screenplay scripts. An automated process to quickly categorize scripts on variables such as genre of film (comedy / action / western / etc.), patterns of the primary characters, and ultimately predict estimated profit from the production. The purpose is to create a rapid and quantifiable feedback loop mechanism to provide screenwriters with input to tweak screen plays and make a profitable film.

People frequently see products which interest them in images but do not know how to look them up for more information.

Upscale / Advertising / Marketing / Computer Vision

The primary purpose is to monetize images using computer vision by implementing automated product tagging within images. These tags appeared in the image when the image was reused in other news articles, blogs, etc. The software could recognise that it's the same image and those are the same objects so they could annotate it to enable a mouse over on products within the image which would bring up product information including a “buy now” option.

Extending an elderly person’s ability to live independently requires a monitoring solution that recognizes when to call for help even when that person can not.

Upskill / Startup / Medical Device / Hardware Integration / Edge Computing

“Help! I've fallen and I can’t get up.” (It’s an old ad from a company called LifeCall that first ran in 1987...

The technology has advanced since then. That’s the idea but there are many complicated nuances to properly identify this scenario (including variables such as active lifestyle) and appropriately notify caretakers. The goal is to provide better patient care and reduce health care costs. The wearable device is proprietary to the company with longer battery life options for constant monitoring. Implemented edge computing for thousands of devices in the field with satellite / cellular / GPS integration.

Wiring diagram updates in the aerospace industry need to be manually diff checked for each aircraft tail number. This is error prone and slow. Every diagram for every plane needs to be checked for updates or costly manufacturing errors can occur.

Upskill / Manufacturing / Aerospace / Hardware Integration / Change detection

The wiring diagrams are still stored in physical libraries. The diagrams were digitized which enabled change detection in the version history through automation.

Increasingly large, complex systems require self healing capabilities. Humans alone can’t but AI can enable the systems to learn how to heal. Automated rule-building for anomaly detection.

Upskill / Telecommunications / Hardware Integration / Automated rule building

Problem statement : How do you take a human centered, manual rule creation process and scale it through exponential complexity and reduce person hours.

Cellular services continue to evolve over time, and each new iteration of 3G / 4G / 5G capabilities introduces several orders of magnitude of complexity in how communications can fail. Historically, this was managed through humans capturing learned lessons in rule form. Any breakdown could cause cascading failure events and the human teams were not scalable to effectively train the network to self heal. Modern telecom networks have far too much complexity for humans to deal with alone. Generating resilience logic by hand would have taken decades. An automated rule building process was required to enable anomaly detection, without human input, for predictive and proactive problem solving.

The most common reason given for clothing returns is “it didn’t fit”.  Garment return rates at both online and retail stores very directly impact profit and online return rates are unsurprisingly higher than retail.

Upskill / Retail / Image and data analysis

The problem is, brick and mortar return rates are ~30%. Online is more difficult to quantify but known to be higher. Reduce return rates and increase profit. It’s a simple yet a complex problem. Image recognition can size a person perfectly. However, that would require a person bares it all and stands in front of a camera to size one perfectly. Not everyone’s “cup of tea” so to speak.  The preferred option is to ask a small number of questions that can be readily answered and use them as inputs to infer body shape and size and ultimately provide a more targeted fit customized to different garments from different manufacturers.

The extremely wealthy are a complicated persona and have entertainment requirements that bend the conventional norms. Create a marketplace for luxury services where money is not a variable, and specifically curated, extravagant services are. 

Upscale / Entertainment / Luxury / Highly specialized industry / Graph network mining / Complex financial transaction processing

https://www.myria.us/

The problem is, how do you google for the ungoogleable? This is beyond the capabilities of Open Table or Ticketmaster or Airbnb. There was no marketplace that brought together the extremely wealthy and the small group of highly skilled experts who can curate and produce a custom experience in real time. Add to that the processing of complex financial transactions typically involving large sums of money. There are not enough experts to do this kind of work manually. If one could connect the wealthy personas to the experts via a marketplace, then you have a business. 

Reduce Accounting/Operations Costs for hundreds of nursing homes (by 30% in 6 months)

Upscale / Accounting / Nursing Home

A holding company that manages accounting and operations for hundreds of nursing homes Increased accountant productivity by 30% in 6 months with improved tools for Accounts Payable and Billing. Goal: Productivity enhancement for the accounting department, allowing lower error rates, decreased AR aging, and ability for existing personnel to handle more physical locations per accountant. Implementation; Did deep dive into over 90 uses including user interviews, and chose 5 initial cases for implementation based on time to implement and user productivity saving.  

Need to acquire more customers by opening new channels

Weather company / ChatGPT LLM / Business expansion

Large weather sensor company wants to become better known outside industry, looking for additional uses for existing data sources. Implementation: Built prototype “format weather data the way you want it” application, released chatGPT plugin that became the most-downloaded weather data plugin

Large company had significant legacy costs, reduced costs by $10M/year (of $150M server budget)

Upscale / Business Operations / Optimize Cleanup Efforts

Goal: Reduce cost of operation and improve system reliability. Implementation: Analyzed existing tooling for outdated code, and created systems to ensure that code ownership was clear and complete, process bottlenecks were addressed, and non-performant code was refactored, resulting in a 6.7% cost savings per annum.

Reduce legal costs for standard NDA reviews

Upskill / Legal / LLM
A large software consulting company wanted to reduce lawyer billing for initial NDA reviews, especially in discovery phase (pre-contracted). Implementation: LLM based contract modification without custom tooling reduced average lawyer review time from 2 hours to 15 minutes billable per NDA (in initial trial).

Project management is error prone with late surprises and costly delays.

Upskill / Decision Support / Project management 

Project management beyond a small scale (single team and single plan) becomes very difficult when cross-team dependencies, variable rates of progress, and complex probabilities make knowing what’s going on extremely difficult. An AI algorithm can assess the combined probabilities and provide people with objective green / yellow / red indicators for milestones and deliverables while exposing the sources of risk. Project management becomes data-driven and actionable.

Second Opinions and HealthChecks, for computer vision, scheduling, regulatory filing generation, and more

Second Opinion / AI HealthChecks

We have been the tech due diligence and review team for many AI and algorithmic systems. We take an honest and constructive approach to reviews to look for value and opportunities for improvement.