Date: 01.11.2025 (2025-11-01)
Prathmesh Thergaonkar (01.11.2025) • https://www.linkedin.com/in/prathmesh-thergaonkar-a3810438/ • Global Director Finance Analytics at Fractal (large AI company) • I have developed products with Agentic capabilities on AR / AP / R2R / controllership etc • Based in Bengaluru
Notes: • Knows Varun from college • Quai.ai • What he is doing: • Focus: AP and AR -> they started a few years back - AR and order to cash specifically • Focus on large CPG companies • They are now switching to agentic technology • Check website -> agentic I2C • They build across techstacks: AWS and Azure • They modularised it • O2C modules: • Cash • etc. • -> people do not talk to each other • They use orchestrator agents • Collections: • Worklist • Risk models: potency of default • Agents are looking across multiple systems • Multiple ERPs, etc. • Real time information about any action a human might take • They replaced human in the loop for 60% of activities -> low criticality, escalation to human in low confidence cases • Also did AP • A lot of focus on invoices -> read it -> IDP fails for 20% of cases (Intelligent Document Processing) • No duplicate payments • Fraud • They also do it • Duplicate in AP • Complex problem to solve -> as AI gets smart, people also get smarter • You have to stay ahead • Cases are complicated -> important to get deep technically • Sell it later as an add-on • Not solved really • CF management: what to pay and what to hold off • Challenges? • Copilot: easier said than done -> huge mistake rate -> complex to build • Text to sequel is easy -> high level stuff • Detailed queries are more complex • Tech behind it: they use multiple LLMs (they are an OpenAI partner) -> GCP, OpenAI, Azure, AWS - also cureAI and other • AP: • Simpler than AR -> loss competitor • SAP, Blacklight etc. in AR • Sellex or sth is leader -> ask Varun for intro for Europe • Document parsing capability • It is solved in his POV for 80% of documents -> large players do that (MSFT) etc.; ChatGPT is really good • Other players: Hypersense etc. • Problem: last few % -> handwritten notes, localised documents (niche language: dutch etc.) • Procurement space is complicated -> contract analysis -> payment, risk terms is the challenge • They built procure.ai -> it is a product of theirs; agentic to sth is also their company -> check what else they have • They work with Fortune 500 companies • Team: • Service company for a long time -> now they go into product (IPO in a few months) • Services will become smaller • Finance: 40 to 45 people -> 20 are on product • Focus on a small part of the process and nail it vs. horizontally • Heidi (some other name): got really good at building a model for collections -> 2nd module for cash, then credit etc. -> a lot of organic growth • A lot of pre-built connectors -> that is expensive -> they build everything on their own-> they will look at partnerships (Palantir, Big4 etc.) • Off topic: Migration / integration process is a great use case (a lot of tech debt) • Only integration is a hard sell -> you need something on top • Highest value: • R2R • O2C • P2P • Checkpoints: • He does the same thing on his own separately • He is technical -> but only vibe coding • He does consulting on the side -> brand building -> is interested in working with us as an advisor or sth. • Maybe he could be a cofounder