Guru expands the reach of its AI-assisted, human-verified enterprise knowledge search app

Phil Wainewright Profile picture for user pwainewright September 27, 2023
AI has the potential to give a welcome boost to the value of enterprise search, but knowledge management vendor Guru believes humans need to be in the loop.

Rick Nucci CEO Guru - Zoom screenshot 2023-09
Rick Nucci, Guru (Zoom screengrab)

Trusted knowledge management vendor Guru last week extended the beta of its AI-assisted Answers product to connect to third-party document stores such as Confluence, Google Drive and Sharepoint. According to Rick Nucci, Guru's CEO, the product brings enterprise search into the AI era, adding a crucial human verification step when knowledge is surfaced using AI. He explains:

The traditional enterprise search experience that we are looking to up-end, watching our customers struggle through this is, you have a search bar, you enter some terms, you get back a bunch of content from all different sources. As the employee, you have to then go through and reconcile that, looking at the different things, trying to figure out what's right and wrong ...

The feedback customers are giving us is, that [process of] search, browse, open, find, is such a big productivity killer. Half the time they don't get an answer they can use anyway, they go and ask an expert. That's been the part they are looking to short-circuit. That's what we've been focused on. It's ask the question, get the recommended answer, and then have a path to be able to confirm, if that answer isn't verified, how do you make sure that it is?

Answers was first released to private beta in May and has been tried out by over 2,100 companies to answer more than 120,000 questions so far. Generative AI helps find answers from within enterprise content, and Guru's verification process ensures those answers are approved as accurate and up-to-date by subject-matter experts. One early customer, online home and auto insurance provider Branch, reports a 30% reduction in the number of questions its agents are asking because Guru is helping them find the answers in existing documentation. Kendall Sipp-Paris, Sales & Support Effectiveness Lead, expects that extending this capability to other knowledge sources will save even more time. She says:

Since implementing Answers, our team is finding answers more easily and not having to ask for help. We’re excited to use it to search across different apps, especially Google Drive since we store a lot of documents in there. This will save our team even more time — they’ll no longer have to find a document and search through it themselves, Answers will do that for them!

Connections are now available to Confluence, Google Drive, Salesforce Knowledge, and Zendesk Guide, while other integrations are currently in testing to expand the list to include Box, Dropbox, OneDrive, and Sharepoint. Others including Slack will follow. The Answers product is now in open beta for all Guru customers with Builder and Enterprise plans. Guru has been able to draw on integration expertise within its team that dates back to its co-founders' history with IaaS platform Boomi.

Organizing enterprise knowledge

Verification isn't necessary for all types of knowledge. Much of what's stored in an enterprise archive is a matter of factual record — what was decided at a certain meeting, or last year's price list. What Guru focuses on is the type of knowledge that changes frequently and needs to be right — how to fix a known issue with a specific product, or what's the current benefits enrollment process? In Guru's typical use cases in a contact center or help desk scenario, the questions that come up tend to cluster around the same recurring topics. Nucci adds:

The problem that teams are typically facing is the repeat question problem. It's the, 'Hey, I've already answered this' problem. That's the part we are looking to intersect with this.

The advent of generative AI has taken out much of the manual process that was previously required to track down and catalog this knowledge. Nucci explains:

So much knowledge isn't anywhere. It isn't in any place, or it's in a messaging system with troves of information of unknown accuracy that someone has to sort through. Before generative AI, it's really manual and labor-intensive to try to address that. We think that becomes now very addressable in a very different way.

Generative AI provides support in organizing, reformatting and translating the content so that it's more useful. He goes on:

We think there's just so much knowledge that's just undocumented flying around, potentially being used inaccurately ... How do you bring that together? By bring it together, I mean, organize it, catalog it, make it discoverable in future searches, future services.

That is a lot of work that used to be done completely manually, that now is going to be handled in a what I would call an AI co-piloting pattern, Guru working alongside teams going, 'Hey, this needs to be organized, this needs to be verified, this needs to be turned from raw bullets into a ... brief narrative, instead of just bullets. We need to remove jargon, we need to translate it into different languages.'

Adding human verification

While the AI provides the tooling to adapt how the knowledge is presented, the Guru Answer app controls access permissions and adds verification of the source material by a knowledgeable human, to ensure the answers surfaced by generative AI can be relied upon. He continues:

'Where did you get your sources from?' I think is probably table stakes. I think there is this transparency that is already critical with generative AI, because I think hallucinations are so top of mind for everyone. We all use ChatGPT and watch it wonderfully make up things. So I think that is critical.

We think the complement to that is, for the categories of knowledge that are high consequence of inaccuracy, there is this added critical need to be able to close the loop with the expert and ensure that it's right — the help desk use cases, support use cases, that span across IT, HR, customer service, even sales questions, even recruiting questions.

Rather than resorting to hallucinations if an answer doesn't already exist, the Guru system consults human subject matter experts to find a more trustworthy response. He goes on:

If you don't get an answer, who should I ask to get that answer fulfilled, so that it never gets asked again? That's the third rung of the Answers experience, we call it 'Ask an Expert', is the ability to route that information. Guru, based on the question, will deduce the likely experts by using the Employee Profile capability we released a while ago, that connects the dots between who you are in the company, what knowledge do you write about and talk about, and therefore, where are you likely topically an expert? Then it will get routed to you.

Having access to answers in the flow of work is also emphasized in the latest announcement, with the Answers feature now rolled out to Guru's Chrome and Slack plug-ins. Nucci elaborates:

That interaction with Guru needs to happen across all the surfaces where you're working. We continue to believe that techniques like browser plug-ins, being built into messaging apps, being built into applications natively, is the other thing that remains critical ... In a given day, you've got to do it in Zendesk, you've got to do it in your HR system, you've got to do it in Slack, you have to do it in email. We intend to be that augmented assistant experience that's following you along all those places.

My take

One of the insights I took away from this conversation is that you can't look at enterprise knowledge as a homogenous body of content that shares the same characteristics. For Guru's purposes, there are three categories of knowledge — static archive knowledge that is settled, dynamic operational knowledge that is constantly changing, and emerging knowledge that has not yet been defined. These three forms all live in unstructured content stores, but there's also transactional knowledge that is stored in transaction systems and I imagine you could also add the taxonomies and relationships that live in graph databases. The more we attempt to harness AI to work with these various forms of enterprise knowledge, the more we learn about the various frames of reference that we need to build so that AI can make sense of it all and deliver back satisfactory answers.

A grey colored placeholder image