“See, I own everything around it, so, of course, I get what’s underneath it. I drink… your… milkshake. I drink it up!”
In perhaps the most memorable sequence from the Paul Thomas Anderson masterpiece “There Will Be Blood,” Daniel Day-Lewis’s oil prospector Daniel Plainview tries to buy a parcel of land he knows sits atop a large deposit of oil. When the land owner declines to sell, Plainview walks away from his offer and instead purchases all the land around the parcel he wants. This allows him to drill underground to the edge of his neighbor’s oil field and surreptitiously capture the oil he wants and, as he famously sneers, “I drink your milkshake.” But how do milkshakes and Oscar-winning actors relate to new wedges for vertical software? Read on.
We have written extensively on vertical software and wedges. As a reminder, a wedge is a product that helps start a relationship with a customer and (hopefully!) is able to capture valuable downstream data. To extend our “There Will Be Blood” analogy, wedges have historically focused on capturing data in a way that’s similar to buying land directly above the oilfield: they’ve generally enabled human-generated information to be entered into or extracted from a system. Think sales reps entering prospect information into Salesforce, ServiceTitan plumbers inputting scheduling information, or Procore construction supervisors managing project information. Employees input and track all of this data to help drive or track performance, collaborate with coworkers, and streamline operations.
What’s exciting about this moment in time is that LLMs can enable experiences that capture data at the point of creation in ways that were impossible for traditional software. Often these experiences are upstream of historical systems of record, meaning these new products can capture, structure, and store previously inaccessible company data and make it usable. This is the functional equivalent of “taking someone’s milkshake,” and it has created what might be a golden era of vertical software — particularly in categories that have been dominated by incumbent platforms for decades, like sales (Salesforce, HubSpot), insurance (Guidewire / Duck Creek), IT service management (ServiceNow), and call centers (Genesys).
These categories have historically been “slow boil” industries, which is to say new entrants have traditionally struggled to grow revenue rapidly, and they often get stuck at subscale revenue profiles due to the defensive posturing of incumbents. We are now seeing AI wedge products shift these industries to a “fast boil,” whereby startups are growing at breakneck speeds. Below are some of the most compelling versions of these products we’ve seen thus far:
Voice: For industries where most sales or customer service interactions still happen over the phone or in person, vertically trained voice agents might empower a new system of record to capture data before the existing systems of record need to be populated. This eliminates the need for a human to interact with the system of record over time and captures data that, in the past, was challenging to collect.
Liberate, for example, has an insurance-specific voice agent that’s paired with an enterprise software platform. Together, they function as an automation platform that sits on top of legacy systems for insurers such as Guidewire or Duck Creek, as well as for home-grown systems. The key for Liberate is its middleware platform, which allows its voice agent to have read and write capabilities into legacy systems, effectively augmenting or even replacing human employees with processing submissions, accepting payments, and resolving claims — all areas where insurance carriers, agents, and brokers haven’t seen any real innovation in years.
This playbook could then be re-run for other categories with similar high-call volumes and legacy infrastructure like banking and travel.
Messaging and email: If sales and support is driven via email or messaging, it’s another way to capture data before it reaches the legacy system of record. Again the critical consideration is context for the text based agent. If it’s outbound marketing or sales, who is the ideal customer and what is the appropriate email address? If it’s synthesizing a quote, proposal, or claim, what are the pricing or claims guidelines to respond to the customer? We believe we’ll see both vertical and horizontal instances of this.
11x, for example, has built a new AI-enabled sales development rep that can automatically generate leads and book meetings for companies. It has built both a voice and email product that can automatically identify, contact, and triage prospect interactions — effectively replacing the entire SDR workflow. In doing so, it creates new leads before they reach the CRM, and as models and training improve, it could end up serving as the source of truth for all sales data.
Meanwhile, hx’s pricing intelligence tool helps insurance companies more effectively ingest and model pricing data. Insurance carriers, particularly commercial insurance carriers, are notoriously slow in turning around submissions because it’s time intensive to structure and assess the messy data they receive. hyperexponential leverages LLM-based tools that can help structure data from submissions more effectively and then stores that data in their Python-based tool for actuaries and underwriters. This creates a system of truth for pricing data.
Internal communications: So much of what makes running a large firm’s technology operations challenging is dealing with the constant issues around onboarding and offboarding employees, and around isolating and managing helpdesk requests. ServiceNow has built a flexible $190+ billion company-building software that can help companies manage all of these requests. But, like in sales or support, the most time-consuming element are the conversations that kick off or happen throughout these support aspects. LLMs can solve those conversational elements and potentially even resolve some of the back-end automation problems.
Fixify, for example, is building a new tech-enabled services platform that augments existing IT infrastructure. They have an LLM-based chat interface that helps identify and triage IT issues, leading to faster ticket resolution times. This platform is all built on an enterprise-grade ticketing platform that learns how various recurring tickets might be automated and reduces the need for human analysts over time.
These companies are great examples of some initial wedge products that are changing slow boil industries to fast boil industries. But with technology simplifying this process, the obvious question is: how can companies leverage these wedge products into building a durable and defensible business?
- Integrations: None of this works without integrations into the existing systems and enabling read/write capability. Building integration software alongside whatever LLM enabled wedge is being utilized is critical for the success of these businesses. In many cases where integrations and implementations can be challenging (like financial services) this may also prove to be a moat over time, particularly as foundation models improve and barriers to entry drop.
- Workflows: While it’s the AI holy grail to imagine a world where AI can eliminate all workflow, it’s likely there will always be some UI layer that allows a user to interact with the data, collaborate with coworkers, and modify outputs. These workflow layers are currently owned by systems of record. Bringing them alongside wedge products key in making a wedge a milkshake.
- Data storage: Owning the foundational customer unit has historically been the method of capturing the most value. If AI enabled tools can start to build their own data repositories, they can eliminate the need to maintain legacy systems of record and capture even more spend from customers.
Like Daniel Plainview’s oil strategy, vertical software companies looking to disrupt established players must strategically position themselves to extract value. With LLMs enabling new ways of capturing and structuring data, these companies are poised to reshape industries and trigger a golden age of vertical software — provided they can integrate seamlessly and build robust moats around their data ecosystems. It’s time for startups to start drinking incumbent milkshakes.
