By Eugene Lee and Julianna Vitolo
A Monumental Transition in the Way We Work
In our last post, we discussed the promise of AI within the application layer and the key hurdles we need to overcome in order to encourage mass adoption. The central question remains: How do we get from where we are today, with promising technology yet little adoption, to ubiquitous, useful AI in the enterprise?
We focus on the application layer, but we acknowledge that a transition of this magnitude will require more than simply migrating customers from conventional SaaS apps to AI equivalents. Workflows optimized for traditional SaaS businesses won’t work for AI-native companies and applications. An AI-first world will require a reinvention of the underlying data models, knowledge graphs, systems of record, and user behavior.
Breaking Down the Transition by Enterprise Function
As a monolith, the concept of ‘AI in the enterprise’ is too big to wrap your thinking around. That’s why we break down the enterprise into its component parts, and frame each workflow within the context of a functional division of the organization. The functions we spend time on are finance; marketing; sales; HR; general productivity; product; and engineering. Within each, we see opportunities for AI-powered automation, and a complete refresh in the organizational structure of the department.
Finance – Within the Office of the CFO, AI is automating away the grunt work. Most of the financial workflows across accounting, expense management, and forecasting are manual, time-consuming, and outdated. They are also formulaic, recurring processes that need to be completed every month and every quarter, making them a perfect application of AI.
Marketing – AI has the potential to be a swiss army knife for modern-day marketing solutions. It paves the way for the personalization of everything, rapidly generating unique ads tailored to the target market – or even to the specific individual. It’s also easier to evaluate reception of specific content, and to manage and scale campaigns. We think there’s a real opportunity for a consolidated marketing platform to enable consistent omnichannel brand messaging, streamline content management, mitigate the risk of LLMs on brand reputation and generate efficient leads for the business.
Sales & Customer Support – AI is helping scale the sales organization; automating inbounds, outbounds, and research; routing and qualifying leads; personalizing messages at scale; and booking meetings with prospects. For instance, Clay recently launched "Claygent", an AI tool focused on data enrichment and sales prospecting. Salesforce notoriously released "AgentForce", out-of-the-box agents that can be deployed by Salesforce's customers. And there are a slew of AI SDRs on the market, supercharging the scale of cold outbound to inhuman levels. Customer support is a particularly interesting application of AI, given the workflows are largely formulaic and consumers are already accustomed to interacting with chatbots. Additionally, there is greater tolerance for error, given the current human-based solutions are often not 100% accurate – far from it in many cases.
HR – Across recruiting, employee lifecycle management, learning and development, and compliance there are many opportunities for the application of AI to make a significant impact on existing processes. There’s a lot of crossover in this area with sales, such as sourcing and recruiting, and scheduling meetings. Finally, given this function revolves around human employees, there are interesting ethical implications regarding the application of AI to processes like recruiting where any bias present in models can create liabilities when applied to people management. Biases in models are, however, easier to investigate, document and correct than those in human interviewers and management professionals.
General Productivity – These solutions aim to speed up the overall productivity and efficiency of the average employee. They target a diverse range of tasks across functions to serve as a generally-capable assistant, augmenting existing workflows with the power of AI.
Product – Within product, “voice of the customer” will take on an entirely new meaning with AI. The tech allows for real-time customer feedback loops, where insights are gleaned from the vast corpus of customer reviews and sales calls to influence the direction of the product. This provides CPOs with a real-time pulse of their product reception, ensuring that they stay aligned with the needs of their users. Ultimately, this should allow them to leverage all data at their disposal to ship high quality products, faster. Generative AI also opens up opportunities to boost product-led sales through interactive and dynamic multi-media content like live product demos.
Engineering – In the future, fewer engineering resources will be required to do the more menial software development tasks. There are both copilots (code completion tools being the lowest-hanging fruit) and autonomous coding agents. Using these tools, engineers will be able to spend more time focusing on higher-level product strategy, rather than the minutiae involved with shipping, debugging and QA testing new code.
Below is an representative view of the enterprise landscape, segmented by functional category:
*Market map is not exhaustive, but rather highlights a few select logos in each category.
We’ll dive into each of these in our function-specific posts to follow. For each category, we’ll walk through the incumbents, the areas where we think there’s room for disruption, and why we believe that. For now, we wanted to include this segmentation to provide a visual representation of our focus areas.
Moving From Single-Function to Multi-Function
Typically, tools we see address one of the functional areas we laid out above, but when we see tools that span functions within an enterprise and encourage cross-functional collaboration, that’s when we get really excited. The primary indicator of a cross-functional product is when we see clear buy-in from stakeholders across different parts of the organization. That means the data generated or product used is valuable enough to attract users from a number of divisions within an enterprise. It also means that the budgetary spend on these tools is coming from multiple parts of the organization, making it less likely to be turned off when times get tough and budgets get strained. We believe the winners will be those that can build as quickly to this kind of impact as possible, but that crucially can do so without also attempting to ‘boil the ocean’ with a too-broad approach.
We refer back to Parker Conrad’s theory of the “consolidated startup”. It’s a powerful guiding force behind many of our assumptions. The number of point solutions has proliferated to a level where now the software stack is so fragmented that most startups use 10-15 tools within each function just to do their basic work. We believe there’s tremendous value in offering up a consolidated platform that performs multiple functions (ideally on behalf of multiple divisions) in order to address a range of needs for a company.
There is a natural tendency for certain divisions to lend themselves to this multi-stakeholder model. For example, the product team already takes input and insights from customer success, marketing, engineering, and sales, so a tool catered to serve that division could be inherently useful to users in other parts of the organization. That isn’t really change management, it’s just automating and codifying the collaboration that’s already taking place. The areas that could lead to interesting overlap where we haven’t yet seen combined solutions are:
HR & Compliance
HR & General Productivity/Collaboration
Go-To-Market – Sales & Marketing & Customer Support all wrapped into one
Finance & GTM
Finance & Product
Sales & Finance (see our post on Quote to Cash)
Automated Workflows: Functional-Specific Agents
What do workflows look like in a world of AI-native companies? Our best guess: automated (performed without user input) and autonomous (able to act independently on behalf of users). The best form factor for autonomous workflows are AI agents – the task-oriented workhorses of the LLM era. It’s clear to us that agents will be the key value unlock for enterprises, not just enhancing the efficiency of existing human-centric workflows, but taking certain low-level workflows entirely off employees’ plates.
We think these agents will need to be function-specific to be effective, at least at the outset. The problems faced when dealing with LLMs, primarily their non-deterministic nature and predisposition to hallucinations, make it difficult to create highly reliable general purpose agents. It is easier to train an agent to do one specific task extremely well, than it is to train it to do 50 tasks moderately well. By going deep within a specific function, agents will benefit from specialization much like their human counterparts do. They can train on a specific subset of tasks and learn from repetition, refining their accuracy over time.
There’s also the issue of compounding error. When a chain of agents are required to complete a task, each subsequent agent loses fidelity and accuracy as they generate output and then use that output to prompt another agent. Since each prompt generated in the chain varies, there is no way to precisely influence what the subsequent prompts will be. Thus, if the first prompt has an error, each of the following prompts will have an even greater margin of error, compounding until the end result is entirely incorrect. It’s difficult to control for this error accumulation in the current environment, given agent quality and testing solutions are still in their infancy.
For these reasons, we think there is real potential for function-specific agents. By building an agent designed to be an expert in one task, there is a higher likelihood that it will be reliable when deployed in production. Building trust in this agent is key to expanding to multi-agent workflows. This is a similar development cycle to software companies, where innovation occurred first within specific verticalized functions and then broadened to all-encompassing horizontal solutions. We assume this trajectory will naturally lead each enterprise to deploy an army of functional-specific agents, each fine-tuned to complete a narrow scope of tasks within their set domain. This will be the initial wave of agents used to automate workflows. Once the functional-specific capabilities are proven out, there will be room for the more horizontal, general-purpose agents to emerge.
A Shifting Landscape
As we continue to track and monitor the new enterprise AI technology, we will provide updates to our views and the areas we think are most ripe for disruption. For now, we are focused on meeting with AI-first founders and companies at the forefront of the knowledge work revolution, changing the nature of workflows and the foundational software systems that support them. In our upcoming posts, we’ll do a deep dive on each of the functional categories we’ve shared here to articulate our thinking around the opportunities within each part of the enterprise.
