Every leadership team eventually meets the same trap. The core business is doing well, so it earns more attention, more headcount and more of next year's budget, and the small, weird bet that might matter in five years gets postponed one more quarter, then quietly dies. Nobody decided to kill the future. The numbers just argued for the present, every single time.

Two ideas, used together, are the way out. Three Horizons gives you a map of where to place growth bets. Organisational ambidexterity tells you how one company can run a stable core and a messy experiment side by side without either one destroying the other.

The quick version

  • Three Horizons (McKinsey, 1999) sorts your bets into three buckets, defend the core, build emerging businesses, seed long-shot options, so you fund all three at once instead of just today's.
  • Ambidexterity is the organisational skill of exploiting a mature business and exploring a new one at the same time, with different rules for each.
  • Read the horizons as maturity and certainty, not as calendar years, the original "Horizon 3 takes a decade" framing has aged badly.
  • The hard part is structural, not analytical: protect the new bet from the core's metrics, or the core's certain returns will starve it.

The idea in depth

The Three Horizons model was introduced by McKinsey consultants Mehrdad Baghai, Stephen Coley and David White in their 1999 book The Alchemy of Growth, drawn from a multi-year study of how companies sustain growth. Their finding was uncomfortable: durable growers do not pick between today and tomorrow. They fund both, and a third thing, the bets that look like nothing yet, at the same time. McKinsey later canonised the framework in its "Enduring Ideas" series, where it remains one of the firm's most-cited strategy artifacts.

The three buckets are simple. Horizon 1 is the core that pays the bills and must be defended and extended. Horizon 2 is the emerging business, real, growing, demanding investment, not yet profitable at scale. Horizon 3 is the option: research projects, pilots, small stakes in things that may become large or may become nothing. The point of the map is that a company always has H1 pressure shouting the loudest, so H2 and H3 need to be named, funded and managed deliberately or they vanish.

flowchart LR
    H1(["Horizon 1
Defend & extend the core
· certain · fund & optimise"])
    H2(["Horizon 2
Build emerging businesses
· promising · invest to scale"])
    H3(["Horizon 3
Seed future options
· uncertain · many small bets"])
    H1 --> H2 --> H3
    H3 -.->|graduates as it proves out| H2
    H2 -.->|becomes the new core| H1
Three Horizons as a maturity portfolio, bets move left as they prove out, not on a fixed clock. Leaders Loop

So the move is: run a deliberate portfolio review that forces three columns, and protect a ring-fenced slice of money and people for H2 and H3 before the H1 owners divide up the budget. If your planning process can quietly route all the funding to the core, it will.

An honest limitation. The original model tied each horizon to a time band, H1 now, H2 in a few years, H3 in five to ten. That assumption is the part that broke. In a 2019 essay, Steve Blank argued the time-based reading is now a "fatal flaw": disruptive H3-style ideas, he points to the likes of Uber and Airbnb, can now be built and shipped almost as fast as an H1 feature, so treating disruption as comfortably far away invites incumbents to be blindsided. He and others made the same case in Harvard Business Review. The fix is not to bin the model but to read the horizons as degrees of maturity and certainty, not as distance on a calendar. An H3 bet is uncertain, not necessarily slow.

Why the map needs ambidexterity to work

Drawing the three columns is the easy part. The trouble starts the morning after, when the same leadership team, the same budget and the same metrics have to nurture a fragile experiment and squeeze an efficient core. Those two jobs want opposite things, and that tension is exactly what ambidexterity addresses.

The vocabulary comes from Stanford's James G. March, whose 1991 Organization Science paper "Exploration and Exploitation in Organizational Learning" drew the line every leader feels. Exploitation refines what you already do, it rewards efficiency, predictability and control. Exploration searches for what you don't yet do, it needs slack, autonomy and tolerance for failure. March's warning is the one to tape to the wall: adaptive systems refine exploitation faster than exploration, which makes them effective in the short run and self-destructive in the long run. Left alone, the certain returns of the core out-argue the uncertain returns of the new, every budget round, forever.

Adaptive processes, by refining exploitation more rapidly than exploration, are likely to become effective in the short run but self-destructive in the long run. James G. March, 1991

Charles O'Reilly (Stanford) and Michael Tushman (Harvard) turned that warning into a managerial design. In their 2004 Harvard Business Review article "The Ambidextrous Organization," they studied 35 attempts to launch breakthrough innovations across nine industries. The winning pattern, structural ambidexterity, gives the exploratory unit its own processes, structure and culture, separated from the core so the core's metrics can't smother it, while keeping the two tightly linked at the senior-executive level so the new business can still draw on the parent's cash, brand and capabilities. More than 90% of the organisations that used this separated-but-integrated design hit their goals, far ahead of the alternatives. (One caveat: that 90% is a single study of selected cases, not a law of nature, strong evidence for the design, not a guarantee.)

flowchart TB
    CEO(["Senior team
holds the tension · allocates · integrates"])
    CEO --> EX(["Exploit unit
efficiency · control · incremental
= Horizon 1"])
    CEO --> EXP(["Explore unit
autonomy · experiment · slack
= Horizon 2 / 3"])
    EX -.->|shares cash, brand, capability| EXP
Structural ambidexterity: separate the explore unit's rules from the core, integrate them at the top. Leaders Loop

So the move is: when an H2 or H3 bet starts to matter, stop asking it to live by H1 rules. Give it a different scoreboard (learning and validated demand, not quarterly margin), a different cadence, and a senior sponsor who owns the tension personally rather than delegating it to whoever loses the budget fight. O'Reilly and Tushman also describe a lighter version, contextual ambidexterity, where the same people switch between exploiting and exploring depending on the task; smaller teams often start there and only carve out a separate unit once the experiment needs real protection.

An honest limitation. Separation has a failure mode of its own. Push the explore unit too far from the core and it loses the very advantages, distribution, customers, capital, that made doing it inside a big company worthwhile in the first place; it becomes an orphan startup with a worse cost base than a real one. The skill the senior team can't delegate is calibrating how separate: insulated from the core's metrics, still plugged into its muscle.

A worked example

O'Reilly and Tushman's own go-to case is Ciba Vision, the contact-lens maker, in the 1990s. (The figures below are illustrative, used to show the shape of the decision rather than to report the company's exact accounts.) Picture a profitable H1 business selling conventional lenses, throwing off reliable cash, with every incentive to keep optimising it. The leadership instead stood up separate exploratory projects to chase breakthrough products, daily-wear and extended-wear lenses, that risked cannibalising the very core funding them.

Run it through the toolkit. Horizon 1: the conventional-lens business, defend, optimise, harvest the cash. Horizon 3 graduating to Horizon 2: the new lens programmes, uncertain, threatening to the core, potentially the next core. The trap was obvious: in a normal budget round, an H1 manager protecting a £100m cash engine will always out-argue a punt that might return £0 or £300m in some uncertain future. So the ambidextrous design ran the new programmes as separate units with their own leaders and rules, so they weren't judged on this quarter's lens margin; kept them integrated at the top, so they could draw on the parent's manufacturing and distribution; and gave them a senior sponsor who owned the awkward fact that the new business's success meant eating into the old one. The question stops being "which bet wins the budget?" and becomes "how do we fund both, on different scoreboards, and let the new one grow up?"

Frequently asked questions

Are the Three Horizons and ambidexterity the same thing?

No, and conflating them is a common mistake. Three Horizons is a portfolio map, it sorts your growth bets by maturity. Ambidexterity is the organisational answer to the problem the map exposes: how one company actually runs a steady core (H1) and a messy experiment (H2/H3) at the same time without one strangling the other. The map tells you what; ambidexterity tells you how.

Isn't the Three Horizons model out of date?

Only the calendar reading of it. The original framing implied H3 disruption takes years, and critics like Steve Blank are right that this is now dangerous, disruptive ideas can ship fast. Drop the time bands and read the horizons as levels of maturity and certainty, and the map is as useful as ever. An H3 bet is one you're uncertain about, not one that's necessarily far off.

Do you need a separate unit to be ambidextrous?

Not always. Structural ambidexterity gives exploration its own unit with its own rules, strongest when the new bet needs protecting from the core. Contextual ambidexterity asks the same people to flip between exploiting and exploring depending on the task. Small teams often start contextual; the structural split usually arrives the moment the experiment starts losing budget fights to the core.

Why do exploratory bets keep getting starved?

Because a certain near-term return almost always beats an uncertain longer-term one in a head-to-head budget contest, March (1991) named this the tendency of exploitation to drive out exploration. The cure isn't a pep talk about innovation; it's structural. Fund and judge the new business by different rules so it never has to win the same argument the core wins by default.

Where does most of the money still go?

To Horizon 1, and that's correct, the core pays for everything else. The discipline is to ring-fence a deliberate, non-trivial slice for H2 and H3 before the core claims the pool, and to accept that most H3 bets will fail. The portfolio works because a few don't.

Related in the Toolkit

Where to go next