Why Clake?

Why Clake?

Content management is at the core of a revenue enablement platform. Therefore, Enablix naturally employed the power of generative AI to deliver a better experience to our customers.

When we started the journey, we knew that data security and privacy was vital to our customers. Therefore, we ruled out training the LLMs on our customer data. And it would have been cost prohibitive to train LLMs with our customer's content.

Generative AI is exceptionally good, and continuously getting better, at enriching and summarizing individual content assets. We scored quick wins on our platform. We could auto-generate summaries and excerpts and auto label assets. 

Next, we focused on streamlining information discovery. Business users are getting used to the simplicity and power of ChatGPT and Claude. We wanted to bring a similar experience to our customers. However, the endeavor proved to be quite challenging. 

Revenue enablement stakeholders manage hundreds and thousands of content assets to drive revenue enablement. And no matter how well one maintains their content library, poor habits and conflicting priorities lead to poor content quality. We knew these challenges when we started out to solve the information discovery challenge. 

We employed different techniques to overcome content scale and content quality obstacles. And it worked but with a lower accuracy. To summarize our learnings (and we discuss our methods and our learnings in this technical deep dive here), 

• The application must identify the user's intent. It is important to identify the user's intent. Without intent, the application cannot provide a high quality response and you are relying on “chance”. 

• We need a structure to define and govern “context”. Everyone speaks to the importance of “context”. But there is no standard or proven method to define “context”. We recognized that we had to come up with one on our own. Here is our take on what is context.

• At scale, enriched information (where enrichment is done for a specific context) will always serve significantly better results for that context than generic information. 

• Knowing a user’s intent helps us identify the “request context”. And we can identify the corresponding enriched information source (the “information context”). 

• Using generative AI at scale requires experimentation and ongoing governance. And experimentation is costly. It should be easy and business-user friendly. (our take on the power of experimentation)

In addition to the above learnings, we also recognized that the shift in Go-To-Market (and in general business landscape) due to AI should be taken into account. We need to ask ourselves, 

• What would revenue enablement look like in the age of AI? 

• If we were building a revenue enablement platform in 2025, how would we build it? 

• Will today’s revenue enablement platforms (including Enablix) stay relevant in 2-3 years? 

• Will AI break down the system silos that have been a source of revenue operations inefficiencies? If it will, what will the new world look like? And how will Enablix play a role in this world? 

Our learnings and us wanting to define a company strategy in the face of generative AI led us to Clake. In the rest of this post, we discuss our thinking that led us to Clake, how Clake fits with Enablix, and our learnings from building Clake and bringing it to our customers.  

Clake: The Initial Thesis

When attempting to solve the “information discovery at scale” problem for our customers, a few glaring issues and opportunities were hard to miss. 

1. There are limited techniques to retrieve information at scale. And these techniques suffer from accuracy issues when employed at a large imperfect information source. 

2. Pre-processing (aka enriching) content for a specific context helps big in two ways: 

   1. Enriched content is a significantly higher quality of information source for a specific context than the original content. E.g. If I need to find out the top customer complaint by querying a pool of 1000 customer call recording transcripts, I would get a significantly better response if I queried an enriched source of customer complaint insights (created by enriching 1000 transcripts) than I would if I queried the 1000 transcripts as is. 

   2. The process of enrichment indirectly enables us to control the business outcome which is a huge unlock. If we continue with the call transcripts example, 

      • When the application builders enable me to query the 1000 transcripts, they can only control what goes into the end query. They will take my question and massage it in some fashion to query the large pool of transcripts. Due to the scale of the information, they have very little control on how the query will be applied across 1000 transcripts. 

      • When the application builders enrich the transcripts for customer complaint insights, they have complete control over processing the call transcripts. They can deal with bad data, duplicate data, information noise, and other elements yielding a high quality customer complaints insights source. 

3. Context definition is not an IT endeavor. It should be owned and managed by business stakeholders with participation from IT. To expand on this point, we will use the 1000 call transcripts use case. We already discussed how enriching helps you introduce context. IT does not have the expertise to define and evaluate the context. When enriching 1000 transcripts for customer complaints, IT does not have the expertise to identify customer complaints. Product marketers and go to market stakeholders do. Business users with domain experience need to own the enrichment process. 

4. Experimentation is vital when implementing AI solutions on large scale information sources. Experimentation enables rapid decision making and helps keep AI token costs down. Enriching 1000 call transcripts to generate customer complaint insights may take a couple of iterations. Stakeholders should be able to experiment with their enrichment process without the burden of AI processing costs. These cost savings will add up fast across multiple AI powered use cases. 

5. API first is a must-have. Clake will power Enablix. But we want Clake to power other agents and applications. 

   • AI has made it 100x easier to build screens and workflows. Organizations are building apps to solve in-house challenges. Salesforce has launched AgentForce to help companies build and maintain Salesforce apps with a prompt. Apps and widgets are a commodity. What they lack is the context. 

   • Organizations are investing in agentic technology for their business problems. We want these agents to use enriched information served by Clake to improve their performance and deliver a more reliable and targeted outcome. 

6. Governance, security, and reusability cannot be ignored. IT may not have domain expertise but they are the primary stakeholders to govern and ensure compliance. IT teams need

   • Visibility of information flow

   • Ensure efficiency use of IT resources

   • Govern and control content security and access

7. With the three learnings, 

   1. business owning context definition 

   2. enabling experimentation is necessary

   3. IT needs to be in the loop

it was clear to us that we not only need to come up with our standards for “context” but we will also need to facilitate the business ownership of context and the ability to experiment without compromising IT oversight. Stakeholders need a workbench where, 

• Business owners can control the context definition process

• Business owners can experiment and iterate when extracting intelligence without dependency on IT

• IT has visibility on the information lifecycle

• IT can exercise information access controls to avoid security gaps

With our roots in content management and our learnings from using AI for information discovery, we addressed this challenge using an AI optimized “Content Lake” approach. The content lake idea led us to the name: Clake. 

As we continue to evolve from our initial thesis, our learnings from our experiences continue to guide our Clake roadmap. 

Clake & Enablix

We have always built software that is highly configurable and extensible. We pride ourselves on repurposing a piece of software to solve business problems in a specific domain.. If you look under the hood, Enablix is similarly highly extensible and can be repurposed to meet industry-specific knowledge management and revenue enablement needs. Therefore, when we started out with Clake, we wanted to build technology that was domain agnostic. 

But we also wanted clear direction on what it means to introduce Clake when we have Enablix.   How will we position Enablix and Clake in the market? 

Clake should be standalone

We want Clake to be a standalone platform. Here are our reasons: 

• We do not know how the revenue enablement space will look like in 2028. Will Enablix customers use Clake as an add-on to their content library? Or will 50% of Enablix workflows be powered by Clake? Or will customers prefer their own agents that will use the content intelligence from Clake? Building Clake as an independent engine was the only way to support the different possibilities. 

• We can tightly integrate Enablix and Clake. We know how to do it very well. So there was no material challenge in integrating the two. 

• When you build two separate (but integrated) systems, you're bound to incur some efficiency and user experience loss. We believe that's a worthwhile trade-off for the upside we can offer the market with a standalone engine.

• Enablix is an enterprise software that follows a sales-led motion. We sign a contract with the buyer and the application owner(s) set it up for their sales users to use the application. If we built Clake as an extension of the Enablix platform, we knew it would be available only to the admins and application owners. That is the nature of our and most enterprise software markets. This outcome went against one of our initial thesis points: Business users should own the context layer. Building Clake as an extension to Enablix would have prevented us from democratizing access to the broader GTM stakeholder community in an organization.

Clake and Enablix are integrated

Clake brings multiple important benefits to the Enablix platform. We started our Clake journey when attempting to activate AI powered experiences in Enablix. With Clake, we can address the earlier gaps in Enablix. 

Additionally, integrating with Enablix gives us quick wins for Clake. We have already delivered value to our customers on real production content. This is vital for Clake’s product journey. Testing AI powered applications is impossible. We are able to prove Clake’s value with real content. 

What does Enablix and Clake integration look like?

• Clake can source content from Enablix

• Enablix can use Clake extracted intelligence to drive AI use cases in the platform

• Clake powered enablement apps are available in Enablix

• Enablix and Clake operate on the same AWS infrastructure to ensure security, privacy, and governance

How is Enablix relevant? 

Enablix is relevant today. But with the advent of vibe coding and many people calling out the demise of traditional SaaS applications, the real question is whether Enablix will be relevant in 2028. 

We believe it will be. It may look different but enterprises will need an enablement system to run their enablement program in 2028 and beyond. 

• Enablix and systems like Enablix cannot be seen only from the lens of screens and workflows. Enablix makes it secure. Enablix makes it centralized. And more importantly Enablix establishes the contract between the “Enabler” and the “Enabled”. Enablix provides the necessary levers for product marketers, marketing leaders, and sales enablement managers to share, communicate and engage their sales teams. We do not see the need for this arrangement going away. 

• Analytics is a big part of an enablement platform. Companies do not want to simply train the sales rep. They need to monitor the performance of their training program.  Enablix will be the central place to monitor the performance of an organization’s enablement program. 

• Enablix offers the governance and control that enablers, like product marketers, sales enablement managers, and other stakeholders need. The details of how this governance layer operates in the generative AI world may change, but organizations will still demand an application where enablers can manage those controls.. 

• Enablement has many use cases. Today AI is changing how these use cases are built, what is the user experience for this use case, and much more. No doubt the use cases are going to be disrupted in the next 2 years. But we believe that organizations will still need a shell to activate these use cases and monitor them. Enablix will be that shell. 

Clake powering Enablix

There are so many ways Enablix can benefit from Clake. We discuss here 4 primary use cases that stand to deliver the highest ROI to Enablix customers. 

Information Discovery

This is a no brainer. Everyone is using ChatGPT and Claude. They are getting used to the smooth conversational experience in their personal world. And professional worlds are attempting to mimic that experience. We are not talking about simply slapping a chat interface in Enablix. We already tried that and our learnings from that exercise started the Clake journey. We are talking about accurate and contextual enablement that moves the revenue needle forward. 

Deal Room Builder

There is a huge opportunity in the sales space to automate Deal Room creation and upkeep. Not only will it improve sales productivity, it will also ensure better buyer experiences. But it is a harder problem to solve. It is a problem that relies on accurate insights and context. Clake is in a perfect position to fill those gaps. 

Content Governance

Content governance is a staple capability of revenue enablement applications. With Clake, Enablix can expand the scope of content governance across other dimensions. Moving from a rules based governance system to an agentic governance model will lower the cost of enablement programs. 

New Use Cases

The most exciting opportunity for Enablix is to activate other enablement use cases that otherwise seemed out of reach. It is common for companies to use sales and customer enablement platforms for one-off problems. Problems like, I want to manage competitive collateral in one place, I want to create playbooks, I want to manage all the SKO content, etc. But as the problem matures, companies shift to a matured point solution to meet their needs. With speed of application development plus the power of Clake, Enablix can expand in these use cases. 

Centralization will drive the future of the revenue enablement market. I doubt organizations can ignore this trend. Enablix, with the power of Clake, has an opportunity to be that central platform.

Clake Learnings

The real test of AI software is in production. Until you run AI powered applications using real world data and information, your software is incomplete. Here are our learnings from running Clake on product content. 

The thesis works

We clearly see improved quality, better answers, and better insights when working with enriched content sets over generic content sets.

We will continue to build on this core tenet. 

Hallucinations

The challenge of hallucinations is real. When I use Claude for work, hallucinations don’t tend to bother me. Yes, Claude messes up once in a while and I have to correct it. There may be more instances of hallucinations for a large research project. But we have all learnt to live with them and we know how to fix them. 

However, when you run the same prompt at scale across hundreds or thousands of unstructured content items, the impact of hallucinations can be severe. First, the hallucination rates go up significantly. When we run the same prompt to extract intelligence from 1000 call recording transcripts, we will see hallucination errors in 5-7% of the responses. Since LLMs are probabilistic, when running the same extract across thousands of records, hallucinations occur more frequently than we would want to.

We have figured out plans to fix these gaps. But note that the gaps exist. 

Visualization is key

We started out Clake with an API first design. But we also implemented a basic App experience to help communicate the value of extracted information. After showing to business users, it is clear that the visualization layer (the Apps) are equally vital to deliver value. 

API first design continues to be a priority. But the App layer has risen in priority too. Here are some emerging use cases: 

• The need for analytical intelligence strengthens the case for Apps. In some instances, business users use Clake apps to give them visibility into the extracted intelligence. 

• Business users use the app to test out the performance. For instance, in one of the projects, an agent uses Salesforce data to extract intelligence from Clake and send an email to the sales rep. It is a headless experience. However, the product marketing team, the business users who own the context and used Clake to extract that context, use the app to try out the performance of the API. 

•  Apps are a must have to showcase end to end value. Apps enable us to demonstrate the use case when equipped with context. It is a great value communication tool. 

Simplicity is vital

We discussed that business should be the owners of context. They should be able to define and manage context. To enable this goal, we want to offer a business friendly experience. The accessibility of Clake resonates with the business audience. 

• Business users should be able to trace the extracted intelligence back to the content item from the source. 

• Users should be able to experiment. Enabling experimentation will streamline the experience. 

Speed is a feature

The speed with which one can extract and analyze intelligence at scale is a big advantage. We have always known that speed enables simplicity. We believe that when customers and prospects see how fast we can convert unstructured content into actionable insights, they are eager to use it for themselves. 

I was surprised how fast LLM APIs respond. We have processed 1000 records in under a minute in many cases. We will continue to focus on speed as a feature. There is a lot to unlock in this swimlane. 

It is not easy

On the surface the concept of Clake is simple. Clake processes unstructured content to extract contextual information at scale. And we often hear the typical IT team's response from prospects: 'We can build this' or 'We are going to build this'. And we know that is not true. 

Yes, IT can build “a enrichment”. They can build “a dashboard”. But even there,

• They will have to address the hallucination challenge 

• Business won’t be leading the context

• The back and forth between IT and business will be drain on achieving the ultimate goal 

We are clearly in the first innings of Clake. And we are super excited to bring Clake to the market to enable future AI powered solutions with better accuracy and reliability.