How Orium’s AI Playbook Turned Complexity into 5x Growth

Modern enterprises run 20-40 different systems across commerce, fulfillment, B2B, and wholesale channels. Composable architecture allows companies to plug in best-of-breed vendors that interoperate well, enabling faster innovation than legacy all-in-one platforms. This approach is now the absolute reality at scale, not just an ideal.

• •Enterprise commerce now requires 20-40 different systems to handle complexity across channels
• •Composable vendors are built to plug in/out and work well with other systems, unlike legacy platforms that build walls
• •Partners in composable ecosystems move faster because they can build for specific domains of expertise
• •Harry Rosen case study: 9-month transformation led to 5x digital commerce growth by replacing monolithic core

Successful AI implementations use specialized agents in specific domains rather than one massive agent. Companies that deploy narrow, fine-tuned agents with specialized vendors succeed, while those investing millions in broad use cases tend to fail. The approach mirrors hiring and managing people—multiple specialists working together.

• •MIT and Google studies show specialized domain agents succeed while broad AI investments fail
• •Think of agents like employees: many specialized agents for many specific jobs, not one all-knowing AI
• •Composability enables swapping AI providers as the landscape evolves (Game of Thrones-style competition)
• •Build systems with iteration in mind—ability to change one input/output without overhauling everything

ChatGPT's ACP protocol launch with Stripe and Shopify enables direct purchasing inside AI interfaces, fundamentally changing customer discovery and buying patterns. Brands must optimize for AI-mediated shopping journeys where products are part of larger problem-solving conversations, not standalone purchases.

• •Orium was SI launch partner for ChatGPT's ACP (Agent Commerce Protocol) with Stripe/Shopify
• •Customers now solve problems through AI conversations, with product purchases as one component
• •AI unlocks new markets: consumers tackle DIY projects they'd previously avoided due to research friction
• •Brands winning early by optimizing content for agent discovery, not just human buyers

The MACH Alliance (122 vendors, major SIs, AWS, GCP) is pushing to become the first true agentic ecosystem with agent-to-agent and agent-to-system interoperability. Focus is on productizing 100+ point agentic capabilities into real use cases within 12 months through standards like MCP, ACP, and Agency consortium work.

• •122 software vendors plus Accenture, Deloitte, AWS, GCP focused on composable interoperability
• •Members already building 100+ point agentic capabilities—goal is linking them together sooner
• •Partnering with Agency (AGNTCY) consortium on discovery, identity, trust, and agent-to-agent communication
• •Target: productized cross-vendor agent use cases available to enterprises within 12 months

While Model Context Protocol (MCP) shows promise, many implementations lack proper guardrails, observability, and access controls. Successful teams use MCP with custom endpoints and controlled integrations, but underlying systems often weren't built for safe, scalable agent interactions.

• •Advanced teams build custom MCP endpoints with Zapier, connecting to code generators and CRM agents
• •Major concern: systems underneath MCP lack appropriate governance, observability, and permission constraints
• •Products that inherit access controls at granular scope level enable safer MCP deployment
• •Need agent frameworks or governance layers wrapped around MCP for enterprise-safe implementation

Orium implements a 'three-times rule': if you do something more than three times, default to progressive automation until fully automated. This cultural shift moves teams from 'why should we automate?' to 'why isn't this automated?'—a key differentiator of AI-native firms.

• •If doing a task more than 3 times, progressively automate it—doesn't need to be perfect on attempt 4
• •Cultural shift from questioning automation to expecting it by default
• •AI-native firms build self-optimizing loops that turn months-long processes into automated workflows
• •Push deterministic automation as far as possible, then strategically insert LLMs where reasoning is needed

Employees progress through stages (activated, adaptive, transformative) when adopting AI—no one jumps to the end. Peer-to-peer sharing drives early adoption more effectively than CEO mandates. Leaders must help teams rethink how entire functions evolve, like QA engineers transitioning to agent evaluation roles.

• •Framework: activated → adaptive → transformative (based on Zapier model). Everyone moves through stages, taking weeks to a year
• •Peer sharing of wins ('I never have to do this again') drives adoption better than executive mandates
• •QA role evolution example: automated QA still needed, plus new eval skills for agent building (30-40% of team vs 10%)
• •Leaders paint vision of how functions transform, giving permission to bring new capabilities to teams

Ethan Mollick's insight: if you had one unassailable advantage, it's gone now—but something never before possible is now possible. For consulting, the billable hour model is dying, replaced by fixed-price outcome delivery enabled by AI's ability to handle variability and uncertainty.

• •Old advantages disappear, but new capabilities emerge—job is finding them before competitors do
• •Consulting shift: from billable hours/utilization to fixed-price outcomes customers can count on
• •AI enables pricing defined outcomes despite variability that was previously unmanageable
• •Optimize for expertise quality and speed of delivery, not just people in seats

As a B Corporation, Orium considers employees, customers, and society alongside shareholders. AI enables modeling circular economies, personalized medicine, and ecological outcomes—potentially more disruptive than the internet but faster. The key is making conscious choices about partners, workload power sources, and ensuring teams benefit from being on the forefront.

• •AI will be as or more disruptive than the internet, happening faster
• •New capabilities: circular economy modeling, personalized medicine, ecological outcome prediction
• •B Corp approach: choose partners based on how workloads are powered and broader impact
• •Ensure team benefits from being on forefront of transformation, not just shareholders