No PriorsNo Priors Ep. 76 | With Ramp Co-Founders Eric Glyman and Karim Atiyeh
CHAPTERS
- 0:00 – 3:38
From Paribus to Ramp: building “self-driving money” for businesses
Sarah and Elad welcome Eric Glyman and Karim Atiyeh and set up Ramp’s core thesis: automate financial operations so companies spend less time and money on busywork. Eric recounts the founders’ earlier company Paribus and how its “agent-like” savings idea evolved into Ramp’s spend and finance platform.
- •Paribus origin: scanning purchases to get price-drop refunds
- •Insight: customers want money in the bank, not points/cashback gimmicks
- •Ramp’s mission: aligned incentives—help businesses spend less
- •Vision of “self-driving money” through integrated financial primitives
- •Ramp’s scale: tens of thousands of businesses across many segments
- 3:38 – 5:17
Why the shift to businesses: waste increases with success + consumer-grade UX at work
Karim explains how early conversations with startup friends revealed that business spend problems mirror consumer savings problems, but grow worse as companies scale. The founders also saw a major gap between delightful consumer apps and clunky workplace finance tools.
- •Early-stage founders surfaced the pain: spend becomes less visible as companies grow
- •Waste scales with company success, making the opportunity bigger at the top end
- •Consumer product expectations (speed, UX) weren’t met by legacy business tools
- •Ramp’s bet: bring consumer UI/UX obsession to complex B2B finance workflows
- 5:17 – 8:03
The original “savings hook”: stand out by cutting redundant spend and time
Eric describes the initial go-to-market wedge: differentiate from ego-driven credit card marketing by focusing on savings and time. Early Ramp product value included identifying duplicative SaaS spend and removing receipt/expense-report friction.
- •Positioning against traditional card rewards/luxury framing
- •Core belief: time is the modern luxury; focus on saving hours, not perks
- •Early product: analyze statements to find redundant software spend
- •Automate receipts and coding to give time back (especially Fridays/close)
- •Savings + automation as the simple, memorable mechanism
- 8:03 – 11:05
Applying AI to finance workflows: context, constraints, and “observability”
Elad asks how Ramp uses generative AI, and Karim explains a pragmatic approach: gather context from statements, inboxes, and ERP connections, then use AI to structure unstructured data. The focus is less on “AI branding” and more on accuracy, speed, and operator control.
- •Context layering: card data + inbox + ERP to understand transactions
- •AI for repetitive finance tasks: close workflows, categorization, bill timing
- •Risk/fraud: three-way matching (PO, invoice, delivery) aided by AI
- •Finance users want correctness and control—not flashy AI labels
- •Job-to-be-done framing drives which AI features get built
- 11:05 – 13:57
Internal AI leverage in go-to-market: “Ironman suits” for SDRs (not agents first)
Eric details how Ramp boosted sales efficiency by instrumenting the best SDR’s workflow, then building automation and tooling around it. Instead of replacing people immediately with autonomous agents, Ramp automated the manual steps that made top performers effective.
- •Shadowing an elite SDR to identify the real workflow bottlenecks
- •Automate signals (fundraises/job changes), list-building, and email discovery
- •Use A/B testing and data to iterate messaging at scale
- •Outcome: multiples higher meeting-booking productivity vs competitors
- •General principle: decompose roles and automate the repetitive steps
- 13:57 – 18:26
How Ramp resources this: engineers own business outcomes and time-based bottlenecks
Sarah probes why engineers would work on sales/marketing tooling, and Eric explains Ramp’s organizational design: engineering is tied to P&L outcomes, not treated as a pure cost center. The company prioritizes “where the hours go” as the core lens for productivity automation.
- •Engineering goaled to business problems (P&L), not just technical milestones
- •Contrast with legacy orgs where engineering is only a cost center
- •Productivity framing: optimize for hours saved, not just dollars cut
- •AI as augmentation to make each hour go further, not just headcount reduction
- 18:26 – 21:12
Build vs buy (and why not sell the internal AI tools): staying focused on the finance buyer
Elad asks why Ramp doesn’t productize its internal SDR/marketing automations as separate SaaS. Eric explains the go-to-market simplicity of selling an expanding suite to the same economic buyer (finance) and the friction introduced by selling across multiple buyer groups.
- •Temptation: internal tooling could be standalone SaaS products
- •Ramp prioritizes a cohesive suite: cards/expenses → bills → procurement → travel
- •Avoid GTM complexity from selling to different functions/buyers
- •Vendor selection philosophy: favor “slope” and progress over checkbox compliance
- •Open question acknowledged: maybe they should explore it more
- 21:12 – 24:31
Selling productivity: “we’re a productivity company” and the finance mandate to do more with less
Sarah asks how to sell productivity when many companies don’t value it explicitly. Eric argues finance teams are uniquely pressured—G&A must shrink as a percentage—so they’re motivated buyers for automation; Ramp “shows, not tells,” focusing on measurable outcomes.
- •Ramp reframed: primarily selling time (automation) more than fintech features
- •Finance is constrained: asked to scale without proportional headcount growth
- •Free-to-try and savings visibility help prove ROI quickly
- •Position on AI: don’t lead with “AI,” lead with outcomes (hours/dollars saved)
- •Karim’s view: successful customers grow more, benefiting Ramp over time
- 24:31 – 27:50
Risk mitigation in financial services AI: constrain tasks, prefill forms, ask only what matters
Sarah challenges the “AI is too risky” argument common in large banks. Karim explains that many finance tasks are not open-ended; by constraining the output space and designing UI that keeps humans in control (prefilled, editable), the risk becomes manageable and measurable.
- •Most finance AI tasks have bounded answer sets (e.g., known categories)
- •Design pattern: prefill and rank suggestions; user can correct
- •Use AI to ask the minimal disambiguating question (context-aware forms)
- •Known-risk vs unknown-risk framing: build workflows where failure modes are understood
- •Translation focus: bridge finance/accounting speak to business operations
- 27:50 – 28:43
Model strategy: stop fine-tuning, harden infra, and swap models fast with evals
Elad asks about fine-tuning versus waiting for better models. Karim describes Ramp’s shift away from heavy fine-tuning toward infrastructure that supports rapid model evaluation and switching, enabling quick adoption of cheaper or better models as they appear.
- •Early fine-tuning experiments, then realization: time better spent elsewhere
- •Invest in infra to swap models quickly and safely
- •Task-specific evals as the gating mechanism for production changes
- •Example: rapid rollout of newer, cheaper models once “good enough”
- 28:43 – 30:45
What the AI stack is missing: better interfaces, controls, and deterministic-style reviews
Karim highlights gaps: end-to-end app navigation agents are hard in fast-changing products, and most model control today is still prompt-based. He sketches a vision for AI-powered “test suites” for marketing copy—automated brand/CTA checks—plus interest in new interfaces like Claude Artifacts.
- •Hard problem: AI agents navigating complex, frequently changing apps
- •Need better “knobs” than longer prompts to guide model behavior
- •Idea: deterministic-ish AI review pipeline for outbound emails/copy
- •Interest in interface innovation (e.g., artifact-like outputs as editable mini apps)
- •Shift from ad hoc prompting to structured, repeatable quality control
- 30:45 – 32:28
Designing a modern marketing team: systems thinking, bottlenecks, and leverage
Sarah asks what marketing looks like when a technical leader owns it. Karim describes mapping marketing like an engineering system—dependencies, bottlenecks, parallelism—then using AI to remove repetitive work so humans can focus on taste and differentiation.
- •Marketing’s enduring human moat: taste and judgment of “good”
- •Reframe marketing as a system with bottlenecks and interfaces between roles
- •AI used to harden processes and increase throughput without slowing releases
- •Goal: keep pace with rapid product shipping while maintaining quality
- •Systems thinking becomes a company-wide competency to hire for
- 32:28 – 41:09
Creative freedom + AI production: the Andy Warhol “Factory” model for brand
Eric uses Andy Warhol’s Factory as an analogy: industrialize production so creators focus on what’s striking. The group discusses how AI collapses the cost of making creative, enabling teams to produce far more variations while preserving a coherent brand system.
- •Warhol’s insight: abstract away production complexity; focus on the striking idea
- •AI lowers creative production costs for visuals, copy, music, and video
- •Brand systems matter: outputs can scale yet remain recognizable
- •Modern marketing challenge: managing massive parallel creative generation
- •Need for new organizational norms to harness “100-person output” via APIs
- 41:09 – 46:03
Augmenting bookkeeping and the future of finance work: automate the tedious, elevate the strategic
Eric addresses job disruption through a bookkeeping example: routine finance work is shrinking while advisory/strategic work grows. Ramp’s thesis is to remove anxiety-inducing chores (receipts, reconciliation) and enable finance teams to spend more time on higher-value decisions—enabled by better “primitives” and orchestration across systems.
- •Labor shift: fewer bookkeepers; different finance roles growing
- •Automation target: monotonous tasks like chasing receipts and tagging spend
- •Finance teams actively want automation due to G&A pressure
- •Bigger blocker than models: fragmented tool stacks and lack of orchestration
- •Build primitives + pipes first, then overlay intelligence for real execution
- 46:03 – 48:20
AI-generated internal podcasts: compressing 10,000 hours of customer calls into 5 minutes
Karim shares an internal applied-AI project that turns huge volumes of customer conversations into short, digestible podcast-style summaries for teams across Ramp. Eric notes an unexpected founder preference: not just highlights from happy customers, but surfacing angry feedback to avoid bad news being filtered out.
- •Problem: tens of thousands of hours of customer conversations are unscalable to consume
- •Solution: auto-generate a weekly ~5-minute “podcast” of key customer moments
- •Enable drill-down by segment/persona/topic to speed internal coordination
- •Sentiment analysis to find highlights—then demand for “angry customer” cuts
- •Use case: ensure leadership sees unfiltered problems as the company scales
