No PriorsNo Priors Ep. 83 | With Rippling COO Matt MacInnis
CHAPTERS
- 0:05 – 1:11
Rippling at a glance: unifying HR, IT, and finance operations
Sarah and Matt set the stage with Rippling’s core promise: reducing the administrative burden of running a company by bringing key back-office functions into one platform. Matt shares a few scale indicators to contextualize Rippling’s footprint.
- •Rippling as an all-in-one workforce management platform (HR, IT, finance)
- •Common entry point: payroll, then expansion into adjacent needs like device management
- •Company scale: ~3,500 employees and tens of thousands of customers
- •Framing the episode’s “spicy” topic: AI-generated performance signals
- 1:11 – 2:37
The compound-startup model: 25 SKUs and high-velocity product shipping
Elad and Matt discuss Rippling’s approach to bundling: many products that can be adopted together and expanded over time. Matt attributes product velocity partly to a deliberate strategy of hiring experienced entrepreneurs.
- •~25 distinct SKUs offered across suites
- •Regular cadence: small features quarterly, bigger launches every few quarters
- •Examples of expansion: scheduling and applicant tracking
- •Hiring strategy: 150+ former founders working at Rippling to accelerate new product creation
- 2:37 – 4:00
Why bundling compounds financially: cross-sell drives SaaS unit economics
Matt connects the compound product strategy to SaaS financial mechanics, arguing that cross-sell into an existing customer base becomes the dominant driver of unit economics at scale. He frames Rippling as multiple subscale businesses that become more efficient as they interlock.
- •Rippling resembles “many startups” growing in parallel under one roof
- •Efficiency gains emerge as product suites reinforce each other
- •Cross-sell economics converge and can outweigh new-logo motion over time
- •Bundling creates a long runway of additional products to sell to existing customers
- 4:00 – 5:00
Platform “vibranium”: the employee graph as Rippling’s core advantage
Matt explains that consolidation’s real leverage comes from shared underlying rails—especially a deep, structured understanding of employees and their relationships across systems. This foundation sets up why Rippling believes it can build AI capabilities others can’t.
- •Consolidation is more than sales efficiency; it’s about shared platform leverage
- •“Vibranium advantage”: a defensible superpower at the platform core
- •Rippling’s differentiator: deep employee data/employee graph across apps
- •AI becomes more powerful when paired with structured, unified workforce data
- 5:00 – 6:01
Talent Signal: using work output to generate performance-management signals
Matt introduces Talent Signal, an AI product that analyzes employee work product and combines it with job-history context to produce a calibrated performance signal. The goal is to provide actionable, evidence-backed insights that complement human judgment.
- •Talent Signal reads work product (e.g., code, support interactions) to infer performance
- •Combines HR context (e.g., role/job level) with outputs produced in tools like GitHub
- •Produces three categories: high potential, typical, needs attention/pay attention
- •Positions AI evaluation of performance as an emerging inevitability
- 6:01 – 8:31
Reducing “manager vibe” bias: facts-first evaluation and privacy boundaries
The conversation turns to the shortcomings of traditional performance reviews, where managers lack time to review everything and “vibes” can dominate ambiguous cases. Matt argues Talent Signal improves fairness by focusing on work artifacts and intentionally excluding demographic attributes.
- •Performance reviews often rely on incomplete evidence and manager calibration meetings
- •“Manager vibe” can amplify bias when performance is ambiguous
- •Model uses work product only; excludes demographics like race, age, location
- •Outputs include concrete examples to support coaching conversations
- 8:31 – 11:13
Calibration and rollout design: localized distributions, early access, and 90-day signals
Matt describes how Rippling calibrates signals by job level and presents results in a company-localized way to keep them usable. To manage risk and adoption, the initial release is constrained: one signal at the 90-day mark, supported by backtesting for credibility.
- •Job level is included for calibration; results shown as localized (company-specific) signals
- •Distribution appears normalized within groups (some high-potential, some needs-attention)
- •Early access program limits scope to avoid stretching the “Overton window” too far
- •V1 generates one signal at day 90; older employees can be backtested using first 90 days
- 11:13 – 13:19
What success looks like: protecting against bad management and surfacing hidden talent
Sarah asks about the aspiration for performance management, and Matt anchors on mitigating the damage of “bad managers.” He shares an internal example where the model surfaced a high-potential engineer who might otherwise have been overlooked, illustrating the intended fairness and performance benefits.
- •Primary motivation: counteracting lazy or biased management practices
- •Model can “lift people from obscurity” by highlighting strong contributions
- •Concrete internal example: underrecognized engineer flagged as high potential
- •Also flags underperformance early to trigger support and improve team outcomes
- 13:19 – 14:56
Dogfooding learnings and guardrails: no AI-only decisions, humans evaluate the whole person
Matt explains how internal use shaped policy: Talent Signal can’t be used as the sole basis for promotions, terminations, or other major employment decisions. The model acts as a “cheat sheet,” while managers remain responsible for holistic evaluation and context.
- •Internal feedback influenced policies and employee-rights considerations
- •Prohibition on making significant decisions based solely on the model
- •Managers must independently assess evidence and context
- •Talent Signal is framed as assistive—supporting, not replacing, human processes
- 14:56 – 18:21
What the model actually evaluates: predictive inputs, code-quality reasoning, and misattribution risks
They discuss limitations and the rationale for focusing on roles with measurable outputs, starting with individual contributors. Matt explains that work product was the strongest predictor in studies, but also shares a customer story where coaching caused misattribution—highlighting why inspection and judgment are required.
- •Initial scope: individual contributors (engineers, sales, support), not managers
- •Work product showed the best predictive power for outcomes like promotion/termination
- •LLMs can assess code quality dimensions (maintainability, extensibility, security)
- •Real-world failure mode: misattribution when a manager heavily co-authors work
- 18:21 – 23:29
Choosing this AI product first: skip chatbots, build what only Rippling can build
Elad probes how Rippling selected Talent Signal amid many AI possibilities. Matt argues that Rippling already has a strong roadmap and didn’t need AI “window dressing,” so they prioritized a high-leverage product where their unique data foundation creates defensibility and revenue impact.
- •Rippling evaluated obvious AI patterns (chatbots/copilots) but deprioritized them initially
- •Philosophy: allocate scarce AI talent to highest-opportunity-cost projects
- •Critique of “AI rebranding” for companies lacking a clear next-product roadmap
- •Decision principle: build what’s uniquely enabled by Rippling’s platform data
- 23:29 – 25:15
Data platform and integration: securely replicating GitHub/Salesforce and resolving identity
Matt outlines the infrastructure required to make Talent Signal work: a scaled data platform that can ingest and structure large external datasets while reliably mapping them to employee identities. This integration layer is positioned as a key enabler for advanced AI products.
- •Talent Signal runs on Rippling’s internal data platform (“our AWS moment”)
- •Integration example: installing a GitHub app to replicate repos into a secure environment
- •Scale challenges: replicating large systems like Salesforce data
- •Critical capability: consistently dereferencing/connecting identities across systems (knowing who’s who)
- 25:15 – 31:28
Early adopters, adoption principles, and responding to “AI pitchfork” concerns
The episode closes with who adopts Talent Signal first and what responsible adoption requires. Matt emphasizes understanding the stakes, avoiding blind deference to AI, and welcoming external critique to reduce harm and improve the product’s integrity over time.
- •Early adopters tend to be performance-oriented and competitively driven cultures
- •Best initial fit: sales, support, engineering—domains already rich in coaching workflows
- •Principles: understand consequences, inspect context, don’t defer decisions to AI
- •Acknowledges critics as valuable accountability; commitment to learn and mitigate unintended consequences
