Writer, the three-year-old San Francisco startup that has, since its inception, focused on building out a secure generative AI platform for enterprises to help them across functions, announced it has raised $100 million in Series B funding today, in a round led by ICONIQ Growth with participation from WndrCo, Balderton Capital, Insight Partners, and Aspect Ventures.

Notably, Writer’s enterprise clients including processional services company Accenture and investment giant Vanguard also joined in the round, signaling strong confidence in Writer’s offerings.

The money will help Writer continue to build out more new features and AI agents for its “full stack” of enterprise-friendly software applications and connections its proprietary AI models.

The company has 14 different AI models available, including Palmyra Small (128M), Palmyra Base (5B), and Palmyra Large (20B), which can already help enterprises do everything from create content and emails to transcribe audio and summarize it. One, PalmyraMed, is specifically designed for medical and healthcare clients.

Writer’s Palymra models assist with functions across enterprises

While the B2B software market has for the last six months been absolutely inundated with new generative AI tools and applications — many of them making use of large language models (LLMs) such as OpenAI’s ChatGPT or Meta Platforms’ Llama 2 — Writer aims to stand out with its in-house AI models.

The company said in its press release today that it is targeting people working in “support, operations, product, sales, HR, marketing, and more,” functions.

Some of the models are tailored for specific industries like healthcare and are fully auditable—a unique selling point in a field often criticized for its “black box” nature.

Trained on non-copyrighted, enterprise-friendly data sources

Writer says its AI models are trained on upwards of “1 trillion tokens of formal and business writing containing no IP or copyrighted content,” according to an email from May Habib, CEO and co-founder of Writer, provided exclusively to VentureBeat by a spokesperson.

What about the performance? There too, Writer has room to brag: The company’s Palmyra “ranked first in three important tests,” using the Stanford HELM benchmarking system for AI models, “scoring 0.609 on Massive Multitask Language Understanding (MMLU), 0.896 on BoolQ, and 0.413 on Natural Questions, outperforming models by OpenAI, Cohere, Anthropic, Google, Microsoft, and more.

“The last mile of quality is the hardest,” Habib said in the release.

Why is the leading model named Palmyra? The name comes from “the ancient city located in present day Syria, where Queen Zenobia rebelled against Rome and established the Palmyrene Empire,” according to Writer’s blog post on the model family, an evocation of female leadership that honors others such as Habib.

More than foundation models

But it’s not just models that Writer is offering: the company also offers a “Knowledge Graph” that helps the models connect with and reference a company’s proprietary data (controllable by customers), and a whole layer of applications including chatbots, prebuilt templates for different business features, and “composable,” user interface (UI) options, meaning you don’t have to be an AI expert to get set up with Writer’s LLM Palmyra.

The software layers do the heavy lifting, and implementation is as easy as drag and drop in many cases, such as with Writer’s “Recaps” application that lets enterprise employees drag and drop audio recordings of meetings, webinars, calls — anything — and automatically summarizes them with key points.

That means that, in addition to competing with other foundation model providers for enterprises, Writer is also fielding apps that compete with specialty software providers like Gong and Otter.ai.

Security and data privacy

The results are impressive, sure, but why should a prospective enterprise customer trust Writer’s Palmyra and other models over more well-known LLMs such as ChatGPT (which recently launched an enterprise tier), Cohere, Llama 2, or others?

In addition to having been trained on a “clean” and enterprise-focused dataset, Habib says that Palmyra also offers the right mix of security and privacy enterprises are looking for, right out of the box.

“Writer LLMs keeps customer data private and never uses it for model training,” Habib wrote to VentureBeat. “Writer also offers customers the option to self-host their own LLM.”

That fact is important, coming after OpenAI CEO and co-founder Sam Altman posted on X (formerly Twitter) earlier this year that OpenAI “never trains on anything ever submitted to the API or uses that data to improve our models in any way,” and told CNBC anchor Andrew Ross Sorkin that the company had changed its Terms of Service to make this explicit.

Palmyra is also Soc-2 Type II, PCI and HIPAA (Health Insurance Portability and Accountability Act) certified — comparable to Anthropic’s Claude 2, which is also said to be HIPAA certified.

Satisfied enterprise customers

Alongside the funding announcement, which Writer plans to use to continue expanding its offerings, the company touted the benefits its clients are already realizing with its platform.

The lingerie/intimates retailer Adore Me reported a 40% increase in non-branded SEO traffic using Writer’s Palmyra, while the robotic process automation platform company UiPath says it cut its content creation time in half.

Writer says its Palmyra enterprise platform has proven its value across multiple verticals, showing a 10x revenue growth in the last two years and a net revenue retention rate over 150%.

With the new funding, Writer is poised to accelerate its mission of embedding generative AI into the enterprise landscape. As AI continues to democratize access to advanced tools and technologies, Writer’s full-stack approach, focused on both flexibility and governance, makes it a compelling choice for enterprises looking to harness the full power of AI.

Correction, Monday Sept. 18, 6:59 pm ET: This article originally underreported the number of models Writer offers, saying it was three when in reality, it is 14. We have since corrected the error and regret it.

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.