While most of us are still just getting our feet wet with generative artificial intelligence, Kiteworks Pte Ltd. is using it to change its business fundamentally.

The company, which makes a secure platform for file sharing, file transfers and email, uses a suite of generative AI platforms for everything from email marketing to sales training. Executives estimate they’ve realized 10-fold efficiency improvements while doubling inbound search traffic and generating a record volume of sales leads.

“Our pipeline is bigger than it’s ever been,” said Patrick Spencer, vice president of corporate marketing and research.

The transformation started six months ago when Chief Strategy and Marketing Officer Tim Freestone “made it a requirement that everyone on my team become a prompting genius,” he said, referring to the instructions used to trigger a response from a generative AI application.

Bouquet of tools

Since then, the company has adopted a growing collection of tools to help with creating search-optimized blog posts, e-books, sales collateral, marketing emails and sales scripts. It uses OpenAI Inc.’s ChatGPT for short-form content and Anthropic PBC’s Claude for long articles and e-books.

Perplexity AI Inc.’s namesake generative engine cites the sources of facts and figures, a feature ChatGPT lacks. “I have not had a single incorrect citation with Perplexity,” Freestone said. Gathering data for a presentation on AI in cybersecurity “would have taken me hours on my own; it took 10 minutes in Perplexity.”

Chatbase from Google LLC’s Area 120 incubator creates a custom GPT model based on the company’s data. “We dump hundreds of thousands of tokens into the knowledgebase and then use prompts to create talking points, emails and scripts to understand compliance regulations,” Freestone said. “It’s guard-railed, so sales reps can only ask questions about what the interface is tuned for, and it won’t answer anything else.” PromptPerfect from Jina AI GmbH helps with writing prompts that coax the best response from generative engines.

Content machine

Careful tool selection and months of practice have turned Kiteworks into a content-generating machine, executives said.

The company poured several hundred blog posts, glossary pages and other content assets into Chatbase and developed custom prompts for each industry the company serves. “When we go to market, there are prompts associated with the different use case personas,” Spencer said. “The result you get today will be different than the result you get next week because additional content has been added.”

Chatbots have automated sales training by enabling representatives to craft pitches and emails specific to the sales scenario. “We used to spend 60 to 100 hours training a sales rep,” Freestone said. “We skip all that. We train them to use the chatbots, which takes one or two hours, and I’m confident they won’t make a mistake.”

Because many Kiteworks customers are heavily regulated, the company must customize its sales briefs to multiple industries. It does that by uploading the entire text of regulations that often “have endless numbers of pages” to Claude, Freestone said.

The summary document the AI engine generates can then be matched to product features and fed back into Claude. “It writes a 2,000-word brief, and we’re done,” Freestone said. “Historically, that would easily take 10 hours per solution brief. It takes about an hour now.” Generative AI engines also create emails, slide decks and video scripts “in hours instead of weeks.”

Translated and localized

Automated translation has converted about 85% of the company’s English-language blog posts into other languages. To optimize for search in the U.K., “We asked Claude and ChatGPT to come up with some topics that would be relevant in the country around secure file sharing,” Freestone said. “It gave us 50, which we culled down to the best 20 and then generated a blog post.” Generative AI can also optimize content to a continually updated list of target keywords.

Prompting is the secret to generative AI success, Freestone said. “With ChatGPT, you have to use phrases like ‘take a deep breath’ and ‘think step by step’ so it doesn’t spit out answers too quickly,” he said. “It’s a black-box tactic, and even engineers don’t know why it does what it does.”

Human oversight of generated content is also essential. “Human in the loop is indispensable,” Freestone said.

Results fall to the bottom line. Kiteworks doesn’t plan to grow its marketing headcount this year as it has in the past. It doesn’t need to. “We have always had a backlog of content requirements,” Freestone said. “We don’t have that anymore.”

And as if to put an exclamation point on the uses of generative AI, executives noted that the email pitch they used to pique the interest of a SiliconANGLE reporter was generated by PRAI Inc.’s “agency in a box.”

“It tells us who to pitch it to, and then it customizes the actual pitch,” Spencer said. “It personalizes the pitch based on your background.”

Image: Kiteworks