Palantir customers are being assigned a number.
McCarthy Construction says AI-assisted workflow adopts Palantir Technologies(PLTR) Nvidia (NVDA) technology is saving construction companies money 30,000 working hours per year.
Investors are focused on more than just the number.
L3Harris Technology(LHX) Says internally tested AI setup built using Palantir and Nvidia models reduces costs 95% in its testing environment.
Another implementation reduced latency by a factor of 22 while improving model performance by 7%, according to Tampa General Hospital.
according to Palantirthese three customers are among several companies that have put its artificial intelligence platform, Foundry and Ontology software into actual operations.
For Palantir investors, these numbers may matter more than any other AI demonstration.
In the most recently reported quarter, the company’s U.S. commercial sales increased 149% year over year.
Palantir must prove that customers are doing more than just experimenting with artificial intelligence to stay close to that speed.
In order to continue to be cultivated, they must gain sufficient economic value.
Palantir’s 30,000 hours of data make AI easier to understand
Finding cutting-edge artificial intelligence in the construction industry may seem less obvious.
McCarthy’s case is unique because of its scale.
The 160-year-old construction company used more than 100,000 production requests for information (RFI) to train Nvidia’s Nemotron model.
An RFI serves as a formal inquiry when plans, specifications, or responsibilities need to be clarified during a construction project.
If certain issues cannot be resolved quickly, the project may be delayed or costs may increase.
Before a request reaches the project site, McCarthy uses Palantir’s algorithms to determine whether the request will cause scheduling or cost issues.
Additionally, it has developed an AI-assisted subcontractor agreement writing system.
according to Palantira workflow that saves McCarthy approximately 30,000 hours per year while costing less to execute than the general-purpose AI model McCarthy tested.
Related: Palantir stock faces brutal test after beating estimates
This business case is easier to understand than the benchmark score.
Thirty thousand hours is roughly equivalent to a year’s labor time of a dozen full-time workers.
McCarthy’s ability to kill these jobs wasn’t necessarily the goal.
This means thousands of hours previously spent detecting hazards and converting project information into contracts may be used for other purposes.
Corporate buyers may provide this type of analysis to their chief financial officers when determining whether another AI contract is worthwhile.
Palantir’s other customers are digitally evaluating artificial intelligence
McCarthy isn’t the only client reporting measurable results.
Defense contractor L3Harris uses Palantir to connect more than 30 enterprise resource planning systems, creating approximately 3.5 million data connections and more than 5,000 scheduled data sources.
Instead of relying solely on general-purpose models, the company is now evaluating AI models trained on its own proprietary data.
Heidi Wood, chief transformation officer at L3Harris, said the initial internal tests performed better than leading-edge models in the company’s test environment while also cutting costs. 95%.
These results are reported by the company and have not been independently audited.
Another example is Tampa General Hospital.
The hospital has been working with Palantir since 2021 and uses their software to perform tasks including managing patient flow, staffing nurse practitioners and detecting sepsis symptoms.
Its latest initiative leverages an Nvidia model tailored to Tampa General’s specific risk and privacy needs.
Palantir said that by fine-tuning the model, performance increased by 7% and model speed increased by 22 times.
Another new client set similarly ambitious goals.
Battery materials manufacturer POSCO Future M signed an agreement on October 21 to deploy Palantir’s Foundry and AIP platforms.
The company hopes to reduce product development time by 50%, increase production efficiency by 30%, and improve customer response by more than 60%.
They are not actual results, but goals.
POSCO Future M said Palantir will first be used for sales and marketing before deciding whether to roll out the technology into production.
This difference is crucial.
McCarthy reported that operational savings have been realized.
POSCO outlined what it believes Palantir will eventually be able to offer.
Investors should not treat these two numbers the same way.
Fingomez/Getty Images
Palantir needs customer ROI to sustain extraordinary growth
Palantir is growing at a rate that few large software companies can match.
Second-quarter revenue was US$1.94 billion, an increase of 93% over the same period last year.
U.S. government revenue grew 90% to $809 million, while commercial revenue grew 149% to $764 million.
Additionally, Palantir completed 73 transactions during the quarter, each totaling at least $10 million.
Palantir reported that it had grown from 849 customers to 1,049 as of the end of June.
These numbers illustrate the importance of customer outcomes.
Palantir doesn’t necessarily need to explain its long-term growth plans to continue getting experimental contracts.
When experiments turn into deployments, customers spend more money spreading them across the company, deployments bring quantifiable savings, and the results are better.
The importance of this growth is already evident in Palantir’s own filings.
In the second quarter, average revenue from the top 20 customers over the past 12 months was $124 million, an increase of 67% over the same period last year.
This means that the number of current customer accounts is increasing significantly.
Potential explanations for this growth can be found in the latest artificial intelligence paradigms.
If a military contractor can significantly reduce the cost of operating a specialized artificial intelligence model, or a construction company can save 30,000 hours of manual labor, then purchasing additional Palantir software becomes more reasonable.
However, there is still an important caveat.
Palantir shares most of its latest efficiency statistics with customers, but doesn’t provide enough financial information to determine the ROI of each deployment.
POSCO’s top productivity numbers remain the target.
Additionally, a single effective process does not guarantee that every Palantir customer will achieve comparable savings.
This is the opposite scenario that investors should be concerned about.
Palantir reports third-quarter results after closing bell November 2.
The focus will be primarily on revenue growth.
Underneath that may be the more important long-term question: Are customers getting enough quantifiable value from Palantir’s AI to continue implementing more of it?
Palantir responded convincingly thanks to McCarthy’s 30,000 hour estimate.
Now it must make such results reproducible.
Related: Palantir stock faces brutal test after beating estimates