Spoke AI uses generative AI to filter workplace noise

As generative AI’s fast-unfolding power to supercharge content-creation raises concerns about what automation might mean for free access to quality information online, Berlin-based startup Spoke AI is preparing to apply it in a more bounded (but still noisy) context: Internally, pitching information workers on tools to automatically summarize inbound communications across third-party tools.

The startup hopes to provide AI productivity tools to all company employees. AI-powered aggregation and summarization tools are initially targeted at project managers.

This group of desk workers uses a lot of third-party software, such as Slack, Jira, Github, Miro, Figma, and Notion, and may need more help managing so many decentralized pings. After the startup refines its tech and gathers new training data, it will launch products for all types of information workers vertically.

This week, Spoke AI released its first beta tools to 20 testers and 500 companies on a waitlist. SMEs with 10–250 employees are showing early interest. Its first feature targets larger companies.

The Q1 2021-founded startup announced a €2 million (~$2.1M) pre-seed round of funding led by the early-stage Northern European focused byFounders fund and Possible Ventures. IBB, Berlin’s regional development bank, received an EU grant. The MVP was built with angel funding before the pre-seed.

“We use [AI] to reduce the noise that people face in their daily work across many different tools and platforms,” says co-founder Max Brenssell. This is initially for product managers, who use eight to 10 tools. We use AI to aggregate, prioritize, and summarize communication across those tools to help them stay on top of it.

The startup’s “workplace operating system” starter package includes a search feature that pulls data from third-party tools, such as conversations or tickets the user has been tagged in, and aggregates it into a “smart inbox” with AI-generated “contextualized summaries.” It will start with a few large companies.

Its “Generative Knowledge Base” (or “intelligent search”) browser plug-in provides AI-powered summarization “in the context of search,” according to Brenssell. Searching across connected tools in their “smart inbox” yields answers “in summarised form, rather than a link to an outdated page.”

Early adopters can use Spoke AI’s automated summarization as a Slack plug-in instead of uploading commercially sensitive content to an unknown platform.

Productivity tools have long offered integrations and aggregations to pull relevant but distributed data into one easier-to-track location. Generative AI creates contextual summaries on top of that to restore context lost when messages are pulled out of their native apps and aggregated into a central repository.

Security and privacy distinguishes. Spoke AI vs. legacy aggregation methods.

The summarization for this use case must be concise and reliable. And securely. “This is how we’re positioning and building it,” says Brenssell. We use pre-trained language models like OpenAI’s GPT. However, we anonymize and clean data before and after processing to improve privacy and safety.

He says Spoke AI envisions selling data anonymization as a service (via an API) to other businesses that want to apply AI models like GPT to their own custom data-sets, in addition to selling AI summarization.

Brenssell also suggests turning the core summarization capabilities into an API to monetize the technology.

Its Slack plug-in summarization tech is currently free. He says the smart inbox feature will initially be offered as a SaaS with tiered pricing based on integrations, security features, etc.

The startup relies on accuracy. Spoke AI may waste users’ time if its summaries don’t reflect the context of notifications.

Brenssell says its beta product has feedback loops so users can rate automated summaries and help it improve. It also shows automation workings so users can trace back the inputs the AI used to create a summary. He says transparency is a priority.

“Obviously, users ask, how can I trust this?” he says. “What we tried to do always is create transparency around where does the data come from that flowed into the summary, and then giving the user kind of a trail, where they can then go deeper if they want to, and really understand where we’re pulling the data that goes into summary.”

Can AI summarization support a standalone business? If productivity giants could add more useful features. (Microsoft is a major investor in OpenAI’s ChatGPT conversational interface for generative AI and has discussed adding it to a variety of its software tools. Google’s email product uses AI, but with mixed results.

Brenssell replies, “We absolutely expect every one of the big [players] like Notion, Slack etc, to launch AI features. I’m glad some have done so. We summarise and boost productivity across multiple tools. Product managers use 10 tools daily, making them the most extreme example. We see that trend in other verticals with more specialized tools.

At some point, you communicate and share information across these tools, right? We see the value in building this flexible, integrations-focused summarization layer so you can use all those tools that make you more productive in specific tasks and areas. By monitoring our system’s communication, you can stay sane.

Because Notion is more flexible for documentation, many Atlassian Jira users prefer it over Confluence. “And these kinds of things we see, increasingly, because there are just more and better tools out there,” he adds. “People don’t want to commit to one product. They want to choose the best tools for their team, workflow, and company. The market looks like that. We obviously also bet on that.”

Generative AI to fight information overload when automation is expected to greatly complicate the signal-to-noise problem seems self-defeating. The movies warn against an AI-AI arms race.

Brenssell calls it “interesting” and says it will be “really interesting to see where, where generative AI tools in the sense of really generating content are going in the next few years”. “Most of the generative AI applications, for example, copy creation, or writing outreach emails, are more external facing,” he says, so Spoke AI’s target market is low risk.

“We’re really focusing on how teams work within their companies and how they can become more efficient. “And there, we haven’t seen so many [generative AI] uses,” he adds.

Notion and Intercom offer feature summaries, he admits. Inbox aggregators can combine WhatsApp, iMessage, b2b email, and Slack, he says. “But nothing that really builds in the summarization, which we believe is just as big the big differentiator,” he says.

Is AI summarization justified? Why not build the same feature with powerful AI models like GPT? He suggests the team’s focus on “data privacy” and ethical data use, as well as product performance to provide “concise and reliable” summaries, will give it an edge. He says summarization is harder than AI-generated content. “Technology is changing. Following that. But we think that by building the right pieces—like anonymizing data—we can keep the AI edge.

The second part is product application. Instead of forcing you to switch from Slack to a new tool, just build a super-integrated user experience that works across all your tools. We oppose this because it causes conflict. So we’re really trying to fit our summarization into existing workflow and take a very integrated [approach].

Brenssell says the startup’s summarization is “currently powered by a combination of fine-tuned pre-trained language models (e.g. Luminous, GPT-3.5), self-hosted open-source technology (e.g. GPT-J, BLOOM, technologies developed by Microsoft and Stanford), and custom models trained in-house (e.g. for Named Entity Recognition, PII Detection, Data Pseudonymization, Question/Answer Identification, Semantic Analysis).

“Having worked with NLP technologies over the recent years and seeing the rapid advancements, we believe that in a space where the core technology will become more and more commoditized, it is still possible and crucial to differentiate,” he also tells. We believe that building with a clear focus on data privacy, responsible, human-centered AI, and augmentation rather than automation differentiates our space. Security- and user experience-enhancing data pre- and post-processing will be crucial to user trust. We want to quickly validate specific use cases with easily accessible technologies and then double down by building proprietary datasets based on implicit and explicit feedback to fine-tune and train our models.”

“As work becomes increasingly distributed and asynchronous, companies need the tools to efficiently share information and create alignment,” byFounders investor Casper Bjarnason said of the pre-seed raise. Spoke is building that, and when we met the founding team, we were blown away by their product vision. We’re thrilled to join Max, Jack, and Carl on their journey!”