By Vijoy Pandey

When McKinsey introduced the Three Horizons of Growth model in 1999, it gave enterprises a time-based vocabulary for thinking about innovation portfolios: the three-to-five-year Horizon 3 for the long-range bets, the medium-term Horizon 2 for adjacencies, and the immediate- and short-term Horizon 1 for the core business.

The Three Horizons model worked well for roughly two decades, but eventually many innovation leaders sensed that the organizing axis, time, had become unreliable. In the age of rapid software development, a Horizon 3 idea could become a Horizon 1 business within a quarter.

If leaders couldn’t organize growth around time, then what should replace it? The answer is risk.

Why Time Broke

In 1999, risk reduction was broadly sequential. Solving the technology problem preceded the market problem, which preceded scaling inside the parent company. That sequence demanded calendar time, so time mapped cleanly onto horizon assignment. Long development cycles meant technology uncertainty, nearer-term work meant market uncertainty, and immediate work meant integration.

Today the pivotal factor in growth models is risk, categorized in three horizons:

Horizon 3 (H3) is technology risk: whether the underlying technology can work at scale under production conditions. But between demonstration milestones and the packaging, pricing, and go-to-market machinery enterprise scale demands, there is an engineering gap that more engineers and faster software cycles cannot close.

Horizon 2 (H2) is market risk: whether a sufficient buyer base exists at price points that make business sense, and whether design partner validation generates a repeatable signal. Even if the technology works, enterprise use cases at scale and with stability may remain elusive. Progress here requires joint discovery with customers; the binding constraint is reality-response time.

Horizon 1 (H1) is platformization risk: whether a validated solution can consolidate into a company-wide platform, integrating cleanly with existing products, go-to-market motion, and customer relationships. Large vendors cannot fight point battles the way startups can; they need platforms that compound.

The three risk horizons are about what you don’t yet know, not about when you expect to know it.

Three Cases

This framework spans six years of incubation work and dozens of initiatives at Outshift by Cisco, the company’s incubation engine. Three current ventures illustrate each risk horizon.

Horizon 3: Quantum Networks. Cisco Quantum Labs is building the networking infrastructure for distributed quantum computing, from silicon through the software stack. Working prototypes exist, but hard physics problems remain, including entanglement fidelity across distance and quantum memory at useful timescales. This initiative stays at H3 because the binding constraint is physics, not market adoption. Asking for revenue at this stage would produce the wrong behavior.

Horizon 2: The Intelligence Layer. Scaling intelligence across distributed agent systems through the Internet of Cognition is the H2 frontier, whether that means agents reasoning together or acting reliably in the physical world. The models keep improving, but enterprise use cases for collective reasoning remain unclear, and embodied AI still wrestles with perception under uncertainty and the simulation-to-reality gap. Progress requires joint discovery with customers, and the constraint is how fast the market can validate the use case.

Horizon 1: Agent Infrastructure. Multi-agent coordination has crossed into H1 because market risk has cleared. AGNTCY, the open-source Internet of Agents framework now at the Linux Foundation with more than 80 member organizations, and the Agentic Network Validator, running in production with a major European carrier, answer the same product-company fit question in two ways. AGNTCY graduated into an open ecosystem and the Validator into Cisco’s portfolio. Both moved because the market validated the use case, not because time passed.

What AI Compresses

Over the past year, leading AI-native enterprises have reorganized teams around multi-agent–human groups, with each employee augmented by frontier models and orchestrated agents. The delivery layer has compressed dramatically; code that once took months to write now takes days.

If risk-based horizon classification were a function of team size or delivery speed, AI would have broken the model the same way it breaks time-based classification. But problem discovery moves at a different pace entirely, and that distinction keeps the risk-based framework intact.

The temptation is to treat execution speed as license to run more H3 bets simultaneously. And for the delivery layer, that logic holds. Eventually, the convergence of quantum computing, embodied AI, and agents that reason collectively may compress innovation across all three horizons in ways that aren’t possible today. That’s the horizon Outshift is building toward.

What the Risk Framework Surfaces

The question of whether and when to go to market depends on whether a parent company can deliver and scale a venture better than an independent startup. Technology fit and market fit are necessary but not sufficient. When company fit is weak, spinning the venture out is the right outcome, not the consolation prize.

Governance has to match the risk type. H3 ventures need executive leadership willing to sustain patient capital. H2 ventures need a general manager with founder-level authority, making acceleration calls on real traction data. And H1 ventures need product management and quota-carrying sales. Misaligning governance with horizon is the most common incubation failure.

A healthy portfolio skews to H3 and H2 because the incubator’s job is eliminating technology and market uncertainty, then handing the venture off. Holding H1 ventures too long misallocates capacity that should be doing the work only an incubator can do.

The three horizons were always real. The categories have always corresponded to technology risk, market risk, and platformization risk. But time is no longer the right sorting variable. Replacing time with risk changes every governance decision that follows: when to graduate, when to spin out, when to hold, and when to push.

Different innovation bets need different treatment, but time doesn’t tell you which bet is which. Risk was always the point.

Vijoy Pandey is SVP and GM of Outshift by Cisco.

---

Find Outshift by Cisco’s frameworks, ventures, and research at outshift.cisco.com and in The Shift, Outshift’s practitioner newsletter.