Enterprise AI spending platform Larridin has secured $30 million in Series A funding led by Pelion Ventures, the company announced on 2 November 2026. The San Francisco startup tracks how corporate spending on software agents, token consumption and seat licences translates into real employee output across 50 enterprise customers.
Key points
- Pelion Ventures backed Larridin with $30 million in new Series A capital, pushing its overall funding to $47 million.
- The company audits more than 2 billion model interactions to connect software invoices directly to delivered commercial output.
- Automated coding assistants helped generate 62% of merged changes by September, while waits for peer review jumped by half.
- The 20-person startup plans to double its workforce as enterprise finance chiefs demand clear proof of software utility.
The Series A round and customer expansion
The Series A round was led by Pelion Ventures, with previous backers Andreessen Horowitz and Gradient Ventures participating after funding Larridin’s $17 million seed round. New investors in the round include GFT, W23, Thayer, DG Daiwa Ventures, Alumni Ventures and Cougar Capital. Total capital raised by the business now stands at $47 million.
Commercial demand accelerated sharply during the second quarter of 2026 as corporate executives began auditing software bills, chief executive Russ Fradin told Business Insider. The company expanded from a single enterprise account in December 2025 to 50 active clients, including SurveyMonkey, Vertiv, Klaviyo, Gainsight, EcoVadis and Sundt. Fradin confirmed plans to double the company’s 20-person headcount to support engineering and commercial expansion.
The investment arrives during a broader wave of capital entering the software automation sector. Startups such as Serval, which raised $150 million for enterprise agents, and Crux Labs, which raised $6 million to capture human judgment, have deployed autonomous systems across corporate workflows. As those tools proliferate, corporate leadership teams have faced increasing difficulty verifying their utility.
Jeff Kearl, the Pelion general partner who led the investment, said generative software costs have climbed rapidly from minor departmental trials into primary balance-sheet liabilities. “All of a sudden they’re like, ‘Wow, that’s a lot of money going to AI,’” Kearl said.
Connecting software bills to corporate output
Measuring enterprise adoption has traditionally meant counting monthly active user accounts. Larridin argues that tracking seat utilisation fails to show whether an organisation is receiving useful work in return. A department can exhaust millions of model tokens without improving delivery schedules, customer resolutions or finished products.
According to the company’s AI Impact product documentation, the platform aggregates billed user seats, token consumption and cloud inference invoices. It then maps those expenditures against observed tasks across departments to calculate the financial return on each tool.
The platform separates estimated capacity from direct cash savings. While automated systems can return theoretical hours to employees, that capacity only generates financial value if an organisation redirects staff time toward productive work or reduces external contractor expenses. Available hours do not automatically reduce departmental budgets or payroll requirements.
Variations in internal spending patterns can be stark. Larridin reported that at one customer organisation where 92% of software engineers used automation tools, a single engineer generated roughly 65% of the team’s entire model bill during one week. At another client, two different coding assistants performing equivalent tasks showed a 66-fold difference in cost per completed code change.
How workflow intelligence observes work
Larridin collects operational data through a desktop background agent and a browser extension. Its Workflow Intelligence platform monitors activity inside software applications chosen by corporate administrators, recording the order of operations, active time and application switches.
The technology is structured around administrative privacy boundaries. Chief technology officer Ameya Kanitkar said the software produces high-level scores and metadata without capturing or retaining message text, document files, keystrokes or screen recordings. Default dashboards display aggregated team information, keeping individual employee tracking disabled unless an administrator changes the configuration.
These observations allow companies to view complete operational sequences. Teams can see where repetitive friction occurs, which stages consume the most time and how automated systems alter employee behaviour. The platform highlights prospective automation opportunities and tracks them as they progress from initial trials to completed installations.
Distinguishing between prospective and completed projects remains central to the platform’s accounting model. Larridin notes that an identified workflow candidate has not delivered any tangible financial benefit until it is functioning reliably in a live operational setting.
Engineering output and review bottlenecks
For software development departments, Larridin provides a dedicated Developer Intelligence offering. The system correlates code commits with model expenses, software defect frequency and pull-request completion speeds against internal baseline data.
The tool also tracks autonomous agent sessions, logging token usage and cost per step. Built-in routing capabilities permit engineering leads to establish cost and performance thresholds, directing routine programming tasks to less expensive models while reserving premium models for complex architectural work.
Data published in Larridin’s Engineering AI Impact Report showed that generating raw code faster does not guarantee faster software delivery. Drawing on data from more than 1,000 developers, the study tracked code changes between April and September 2026: automated tools helped write 35% of merged pull requests in spring, expanding to 62% by autumn. Over the same six-month window, median wait times for initial human peer reviews lengthened by 50% across the measured cohort.
The finding indicates that faster generation can shift the operational bottleneck rather than eliminate it. When autonomous systems generate more code than senior staff can review and verify, deployment pipelines slow down despite higher initial output numbers.
The measurement layer for software agents
Fradin and co-founder Jim Larrison previously worked together at digital audience measurement provider Comscore. Fradin said the emergence of heavy corporate investment in generative software mirrors earlier shifts in digital advertising, where third-party measurement became necessary to verify vendor performance claims.
Larridin claims its current customer cohort has recorded an average 47% increase in measurable team output and a median first-year spend optimisation of $2.4 million. The company compiles these figures directly from client deployments across more than 2 billion tracked interactions and 82,000 observed operational workflows.
The shift toward financial accountability reflects wider industry behaviour. Financial institutions including BNY have established formal auditing teams to monitor token consumption, while Amazon has warned its internal engineering teams against unrestrained model querying. Larridin will use the new capital to expand its data science staff and build additional direct integrations into enterprise resource planning software.
