Our verdict
Mosaic Data Science is worth hiring for an operations-heavy company, especially in aviation, logistics or energy, that wants experienced US and Canadian data scientists to build a forecasting, optimization or document search tool. It publishes no rates, has no verified client reviews and keeps most clients anonymous in its case studies, so ask for references before signing.
The right buyer
- Airlines, airports and logistics operators. It built taxi-out time and departure runway predictions that reach a leading airline’s pilots through the Aviobook app at Newark, Chicago O’Hare and San Francisco (case study).
- Government agencies that want language model search kept inside their own cloud. For the FAA it built chat tools over internal documents with an open-source model hosted in the agency’s AWS environment (case study).
- Companies that want AI search to run on their own servers. A financial recovery audit firm got an email search tool that runs a local language model on an in-house server and returns results in about five seconds (case study).
- Teams that need more data scientists for a while. Its Rent a Data Scientist program places full-time onshore staff with a client’s team (The Silicon Review).
Reasons to look elsewhere
- Pricing is hard to compare. It publishes no rates or minimum project size and has no Clutch profile.
- Client identities are mostly hidden. Its sales and marketing vice president told The Silicon Review that non-disclosure agreements keep most client identities private, so its case studies rarely say who the client was.
- Results are thinly reported. Its recent language model projects are written up as prototypes handed over with training, and the case studies report search speed rather than money saved or hours freed (case study).
- Its deepest proof is in aviation and logistics, much of it from its parent company’s work for the FAA and NASA. Buyers in other industries have less to go on (About).
Mosaic Data Science, At A High Level
Mosaic Data Science is an AI development, implementation and consulting firm in Leesburg, Virginia, that builds custom machine learning, optimization and language model software for operations decisions.
- Set up in 2014 by Chris Brinton and Steve Atkins as the commercial data science arm of Mosaic ATM, the aviation analytics company Brinton founded in 2004 (About).
- About 20 people on its team page, most of them data scientists with operations research, physics or engineering backgrounds. It says it employs full-time North American data scientists and has served clients remotely for close to a decade; Director of Data Science Sean Osis leads its operations in Canada (About).
- Its parent company leads machine learning and AI integration work for the FAA under a 2020 task order (Mosaic ATM).
- Works by project, by placing data scientists with a client team, by training and road mapping for in-house teams, or by deploying and running models in production (The Silicon Review).
- Says it has delivered more than 200 AI and machine learning solutions and keeps 89% of its customers (homepage).
Its documented client work includes:
- Taxi-out time and departure runway predictions for a leading airline, delivered to pilots through the Aviobook app at Newark, Chicago O’Hare and San Francisco, with 27 more airports planned. The airline expected more than $3 million a year in fuel savings (case study).
- Chat tools for the FAA’s enterprise information management office that answer questions from internal documents, built with LangChain, OpenSearch and an open-source Vicuna model inside the FAA’s AWS environment (case study).
- A decision support system for a UPS air hub that recommends when to start aircraft engines. In a CIO Review profile, Mosaic said UPS had used it for eight years and estimated it saves UPS about $5 million a year (CIO Review).
- A prototype email search tool for a financial recovery audit firm, using a Neo4j vector index and a local language model on the client’s own server, with searches taking about five seconds (case study).
- A review of a customs broker’s tariff classification model, where Mosaic’s experiments lifted chapter-level results by 10% to 20% (case study), and a customer churn model for a propane company that was about 80% accurate (case study).
Scores
| Criterion | Weight | Score | Basis |
|---|---|---|---|
| Production track record | 20% | 6.5 | Runway and taxi-out predictions in pilots’ Aviobook app at three airports; a UPS air hub tool in use for eight years by Mosaic’s account; recent language model projects written up as prototypes. |
| Technical depth | 15% | 7.5 | Optimization, forecasting, computer vision and retrieval systems from a team with MIT, Stanford, Cornell and Duke training; open-source models run on client servers. |
| Industry depth | 10% | 7.0 | Deep aviation and logistics work through its parent company, plus energy, financial services, manufacturing and federal agencies. |
| Delivery model and handoff | 15% | 6.5 | Project, staff placement, training and model deployment options; documentation and staff training on the FAA and audit search projects. |
| Commercial clarity | 10% | 4.5 | No published rates, minimums or Clutch profile; engagement types are described but not priced. |
| Platform independence | 10% | 7.5 | Describes itself as tool-agnostic; has built on AWS, Neo4j, OpenSearch and open-source language models run locally. |
| Stability and scale | 10% | 6.5 | Part of Mosaic ATM, operating since 2004; about 20 people on the data science team page. |
| Independent evidence | 10% | 4.5 | No verified client reviews; business press profiles in CIO Review and The Silicon Review; most case study clients are anonymous. |
| Overall | 100% | 6.4 | Weighted average |
Where Mosaic Data Science fits
| Need | Fit | Why |
|---|---|---|
| Airline, airport or logistics operations models | Strong | Runway and taxi-out predictions for a leading airline; UPS air hub engine start tool. |
| Language model search over internal documents, kept on your own servers or cloud | Good | FAA chat tools in the agency’s AWS environment; on-premises email search prototype for an audit firm. |
| Extra data scientists placed with your team | Good | Rent a Data Scientist program with full-time North American staff. |
| Buyers who need verified references and published prices up front | Weak | No rates published, no verified reviews and mostly anonymous case studies. |
What clients and leaders say
“We were data science before data science was cool.”
“True AI and ML solutions require a trained hand in customizing the tech to the customer environment; there is no single arrow in the analytics quiver to solve all use cases.”
“There are still large risks with not being able to understand what is happening with your model’s mechanics, especially as you think about extending a trained model to data it has not seen.”
Who leads Mosaic Data Science
Chris BrintonFounder and CEO, Mosaic Data Science
Steve AtkinsCo-founder and EVP, Mosaic Data ScienceSources (11)
- Mosaic Data Science: homepage
- Mosaic Data Science: About and team
- Mosaic Data Science: Case studies
- Mosaic Data Science: Airline taxi-out and runway prediction case study (March 2023)
- Mosaic Data Science: FAA enterprise information management case study (August 2024)
- Mosaic Data Science: Financial recovery audit search case study (July 2024)
- Mosaic Data Science: Tariff classification case study (June 2023)
- Mosaic Data Science: Customer churn case study (July 2020)
- CIO Review: Mosaic Data Science profile
- The Silicon Review: Mosaic Data Science profile (April 2023)
- Mosaic ATM: FAA machine learning and AI task order (October 2020)