McKinsey, BCG, and Goldman Sachs leadership. PhDs in AI. Not just modeled, but shipped. One team from strategic diagnostic through deployment.
We find that most fail in one of a few structural ways — none of which have anything to do with whether the models work.
Most teams know to narrow. Few teams stay narrowed. One use case quietly expands into dozens of features, chasing perfection instead of the version operators will actually adopt. Or the program runs the other way: locked into today's architecture, today's vendors, today's tools, until a new competitor, a new payer rule, or a new release from OpenAI shifts the ground, and the build can't absorb it. Either way: budget gone, nothing shipped.
Models work in a lab and never reach the operations they were built for, because they weren't co-created with the operators who'd use them or the executives who'd back them. Data science delivers. The operators don't know how to use it. The executives can't tell whether narrow focus is the right call or a failed program. None of it becomes Functional AI.
Every transition is a failure point. From strategy team to build team, from build team to operators, from the people who built it to the people who'll maintain it. Each handoff resets institutional knowledge and creates rework. More effort goes into keeping the system running than into leveraging or refining it to maximize its value.
The tool ships, then can't deploy. The data the model needs lives in three systems that don't talk to each other. The compliance review wasn't started until launch week. The security exception never gets approved. The asset sits behind a wall the program never planned for.
Avoiding these failures starts before execution, with a strategy built around the P&L, designed for adoption, staffed for the work, and ready to adapt.
Emerald Key was built to do both.
I spent years leading strategy, transformations, AI and analytics for Fortune 500 clients as an Associate Partner at McKinsey, with a PhD in engineering behind it. I founded Emerald Key because the AI moment finally made it possible to do both halves of the work, the strategy and the build, inside a single senior team, measured in the business outcomes that matter. No handoffs. No proofs of concept that never ship. The same people you scope it with are the people who build it.
If you're thinking about AI seriously, let's talk.
"The biggest risk is not acting... someone in your industry will get this right."
Nephi Johnson · From a recent interview
The team you'd hire if you could hire anyone.
Partners who ran $1B+ transformations across Fortune 500 and PE-backed companies. Bringing strategic clarity, analytical rigor and executive communication to complex problems.
Dealmakers and operating partners who have built, integrated, and exited platforms. They run AI programs the way a sponsor runs a portfolio, measured in EBITDA, not in slides.
Founders, C-suite executives, and PhDs actively running companies and building AI products. Not consulting from the sidelines. The kind of expertise most firms hire to consult once and never retain.
Build the agents, models and analytics the use cases depend on. Production-grade, not lab-grade. Tested against the workflows they're meant to support.
Turn "this is technically possible" into "this changes the P&L." Deep problem solvers who own the connection between AI capability and operating outcomes.
Design the systems that scale past the pilot. Define the user experience, tech stack, tools, and the architecture that holds up at production volume. Make the build-vs-buy and now-vs-later calls that keep the program moving.
Compliant integration across business management systems (CRM, ERP, EMR, BI) without breaking what's already running. Monitoring, governance, and the model-ops layer that keeps production stable after launch.
Build the production code behind the AI: backend services, integration glue, agent runtime, API layer. The engineering that turns models into running software, not lab demos.
Build the interfaces and dashboards your team actually uses. Custom production tools, not generic dashboards retrofitted to your data.
A library of innovative, custom, production-grade AI agents we've already built. New engagements launch with a head start, not a blank slate.
Data quality matters. So does momentum. We don't choose between them. We test POCs against the data we have, standardize on the system that works today, and let the infrastructure mature alongside the agents. Each stream proves the other. The result is faster impact without infrastructure debt.
Two streams, one team, no blockers.Each engagement is staffed for the phase you're in, with senior leadership consistent throughout.
The diagnostic maps the current state of your business, audits your data and systems, and identifies the highest-leverage AI opportunities, sized to the P&L, sequenced by value, and pressure-tested for feasibility. You leave with a defensible view of where AI moves the business and what it's worth. Four weeks for a roadmap diagnostic, faster when due diligence demands a shorter cycle.
The design phase turns the diagnostic into a buildable plan: reference architecture, prioritized use case sequence, build-vs-buy decisions, integration map, and a milestone plan with named owners. We design the change management plan alongside the technical plan, how the people closest to the work will learn the system, adopt it, and own it. Built around your existing stack and the people who will run it, not a template solution applied to your business. Every architectural decision is made against the value baseline from Diagnose, so the design optimizes for the outcomes that matter, not the technology that's trendy.
We build working software in your environment, instrumented from day one. Weekly demos with real outputs, not slides about future outputs. Your executives stay in the loop on what's shipping, what's working, and what's being adjusted. As operators start using the system, we adapt the build to ensure adoption. Every milestone is measured against the value baseline: if a use case isn't moving the number it was scoped to move, we adjust the build or the scope before we keep going. What works scales, across teams, across business units, across the next AI opportunity in your roadmap.
Production AI systems need ongoing care, especially as they scale: monitoring, governance, model drift management, evaluation against new edge cases, and continuous improvement as the business evolves. We run it with you, and we train and co-work alongside your team so the capability lives inside your organization, not in a vendor relationship. By the time you're operating it without us, your people understand the system the same way our people understand the system. The result is durable: AI that keeps producing value long after the engagement ends, because the team running it knows why every decision was made.
Built on the core platforms enterprise AI runs on.
Resulting in Functional AI.
Shipped work across financial services, ed-tech, biotech, non-profit, and multi-industry advisory.
Listens to inbound leads to learn what each prospect actually values. Calibrates a propensity-to-buy score from this AI listening, machine learning and advanced analytics. The score drives downstream decisions — which leads to escalate and which marketing channels to invest in. A purpose-built UI puts the decisions into the CMO's hands.
An interactive avatar trained on the firm's products, sales playbook, and compliance requirements. Qualifies prospects, consults on fit, and closes transactions within authorized dollar thresholds. Integrated with the CRM and order management systems. High-value transactions get a warm handoff to the human concierge desk. Does substantially more than answer FAQs — it sells.
Recruits alumni mentors using compliant public-data scraping, matches them semantically to current students using LLMs (hometown, athletic background, volunteer history, career path) — going far beyond traditional rule-based matching that breaks when fields are missing. Automates the full program lifecycle: outreach, scheduling, engagement tracking, follow-up. We built the SaaS, the GTM strategy, and supported the raise.
Multi-function speech AI and analytics tuned precisely to the firm's call types, customer base, products, and regulatory environment — not an off-the-shelf transcription engine pointed at sales calls. The system scores leads, openers, closers, and compliance. Outputs roll up into the metrics, dashboards, and alerts that desk managers use to coach brokers, save at-risk deals, and route high-potential conversations to top performers. Has supported the firm's sales floor through ~50% headcount growth and revenue scaling from ~$100M to ~$300M.
Ran technical due diligence on the product, then translated the findings into the business thesis and capital strategy that drove the raise. Built the data-room documentation, advised on operating and commercial model, and joined investor meetings over the course of the raise. The team closed $45M Series A. Still actively advising on growth strategy.
Pro-bono engagement modernizing the organization's technology stack, brand presence, and growth strategy. New website, new CRM, refreshed brand identity, and operating model designed to scale impact alongside donor growth. Functional capability transferred to the team for ongoing ownership.
Workshops and feasibility scoping engagements for organizations evaluating AI opportunities. Includes brainstorming, wireframing and value prop definition of potential tools, testing of available AI products against business needs, market sizing, and process mapping. Conducted across architecture, legal services, luxury fashion, tax services, and consumer packaged goods. Deliberately lighter-touch than full strategy engagements — designed to help organizations decide where AI fits before committing to a transformation.
Strategic diagnostic, technical build, or anywhere in between. One team in the boardroom and in the build.