Commoditized industries suffering from price compression will have new, high margin winners powered by AI.
My first ever investment back in 2018 was Invisible Technologies. After seven years of linear execution, the company went from ~$20M to $100M of revenue in one year. They’re growing 100% a year on 25% EBITDA margins. Invisible offers services that would typically be outsourced like data structuring and enrichment, automating those tasks using AI agents. Even for Invisible, these operations started off as labor intensive, hiring thousands of offshore humans to complete tasks according to a checklist. Then came LLM’s, and Invisible is now servicing customers like Microsoft and OpenAI. Model training data sourcing, structuring and cleaning is handled by Invisible’s humans-in-the-loop, powered by AI agents that handle everything up to verification.
Thesis Overview
Services businesses are labor intensive, slow-growth and typically low-margin. This has made them historically difficult to sell traditional software to. Willingness-to-pay is low, labor wants to be out servicing customers instead of using another tool, and ACV growth through seat-expansion is limited.
However, they represent a large and under-penetrated $2.3T industry that employs over 100M people, 59% of the total US workforce, and contributes 77.6% of total GDP. This dwarfs the SaaS market size in comparison. In some ways, the promise of software failed. US corporate profits have flatlined at ~10% since 1960. Software has barely made a dent in end-customer efficiency, instead increasing top-line efficiency and user experience.
Our thesis is that while the software era sold access to tooling, the rate-limiting factor was the end customer still being responsible for execution tool usage. LLM’s will enable startups to sell access to the desired outcome itself, executed by the company that’s selling the service. This will both increase productivity, unlock willingness to pay and end customer’s bottom lines.
Levante believes there will be multiple $10B companies built across specific service verticals. From accounting and finance, HR, administration and asset-intensive services, work will be automated to a scale that skilled workers and previous iterations like RPA couldn’t handle.
The first generation of services companies like CliQr in cloud management (acquired by Cisco for $260M in 2016), SpringCM in document management (acquired by DocuSign for $220M in 2018) and UiPath ($11B market cap) did a good job organizing specific workflows. The potential of these businesses today is much larger. Agentic AI are able to both organize workflows and execute through both complex decision-making and autonomy.
Business Model Unlock
At Levante we’re seeing founders using LLMs to transform services companies themselves. From agents to run the operations of a commercial cleaning company, to agentic AI flight-school instructors. Since LLMs are excellent at replacing text-based clerical work, they can unlock value in industries previously weighed down by high-cost labor. That allows them to do two things: increase the bottom-line of end customers by automating expensive back-office overhead, and charge as a percentage of labor replaced which increases ACVs.
Levante portfolio company Mura is using humans-in-the-loop powered by AI to handle scheduling, ordering, payroll and bookkeeping functions for commercial cleaning companies. SaaS solutions in the space have struggled to grow meaningful share or ACVs. Swept scaled to $3M ARR before stalling, with per-seat SaaS pricing between $50 - $200 per month. In contrast, Mura is able to charge 10-30% of the labor replaced: their first customer TCP America is paying $3,500 per month, or $42,000 annually, a 17.5X ACV increase. Mura is both unlocking 15-20% efficiency gains for customers and higher end-customer spend in a $75B market where SaaS struggled.
In Europe, 11x just closed a Series A with Benchmark at $100M valuation. While not operating in a pure services industry, it shows the power of the model. The company went from $0-$2M in ARR in just six months. 11x allows customers to build AI SDRs, which execute on everything from lead sourcing, personalized outbound and qualified prospect prioritization. The power of this model is this: instead of selling another sales prospecting tool to SDRs for $29 a month, 11x can price itself as a 5x cheaper ($70k average salary), 2x more productive, 24/7 SDR hire that doesn’t need as much onboarding or training to get started. There are 667k SDRs in the US, implying a $250M-$1B market for seat-based SaaS (depending on pricing), and a $47B market for AI labor replacement.
Pricing is tied to outcomes, in this case number of qualified leads, which allows for ACV growth in line with customer success. Today, 11x has reportedly high churn. Our assumption is that since the product has only been fully live for six months, the agents themselves are not fully trained on relevant SDR data and probably surface irrelevant leads. However, it’s clear the demand-pull is strong, with the company growing 50% monthly. We expect that as companies like 11x acquire more training data, their agents will perform closer to the humans themselves, tying ROI closer to outcomes and closing the churn gap.
Opportunities
We think about markets worth pursuing as ones where current software spend as a percentage of total salary spend is low single or double digit. In HR for example, the leading software company Workday generates ~$2B of revenue while total HR salary spend is $200B: 1% penetration. We’re particularly excited by automation of outsourced services and the salaries tied to them, which Gartner estimates across IT and business processes to be $2.3T with AI expected to inject over $28B over the next five years.
Similar to Invisible, business processes services like data entry, audio transcription, financial reporting, bookkeeping and others are repetitive, rules based and execution-orientated: a great fit for LLMs and for business models that encourage usage. Some recent examples include ConverzAI which raised $5M from Foundation to take over recruitment calls, Mechanical Orchard which raised $31M from Emergence to automate the transition of mainframe data to cloud, and Tennr which raised $23M from A16Z to handle fax reading and data entry for physicians.
We argue that every service offered by firms like Deloitte by thousands of human consultants can be replaced by agentic AI, and we plan to find outlier founders who want to take on the Big Four’s $200B combined revenue. Importantly, the more vertically focussed these agents are, the more sustainable the businesses will be. GoNavi for example, while starting in a niche $5.5B flight school industry, will be able to gather and train its models on flight audio instructions and responses.
