Quick Wins, Strategic Bets and Transformation Initiatives

Bansi Nagji and Geoff Tuff studied how companies split their innovation effort across industrial, technology and consumer goods firms. The firms whose share prices outperformed their peers put about 70 percent of their innovation resources into the core business, 20 percent into adjacent moves and only 10 percent into transformational ones. Yet when the authors traced where the returns came from, the ratio was roughly inverted: the core contributed about 10 percent of long-term returns, and the transformational 10 percent of effort produced about 70 percent¹.

So should an organization put most of its money into the core, or into the ambitious? The question assumes one kind of investment. What separates the leaders in the study is that they treat different kinds of investment differently, fund the ambitious kinds on purpose, and decide the balance instead of letting it happen.

Three kinds of investment, one portfolio

Prioritizing the AI Portfolio chose which initiatives to fund within a finite organization. The funded set is never uniform. It holds three kinds of investment, and they are distinguished by what each is meant to buy, not by the technology, the size of the budget or the department that asked.

A quick win has a clear problem, available data, known users and limited process change, so it can return measurable value within months. A strategic bet has high upside and high uncertainty: a possible new product, service, revenue model or proprietary capability whose economics are not yet known. A transformation initiative changes how a part of the enterprise creates value: its processes, roles, decision rights, data and systems, usually across several functions and several years.

The framing has an older ancestor. Mehrdad Baghai, Stephen Coley and David White of McKinsey described three horizons of growth: extending and defending the core business, building emerging businesses, and creating options for the future. Their central point was that a company must manage all three at the same time, not one after another². What Is an Enterprise AI Strategy? used the horizons to shape the strategy. Here the lens moves down to the initiative, and the mapping is not one to one. A transformation can sit entirely inside the core business while a quick win can be the first visible step of a strategic bet.

Quick wins: momentum, and a trap

Quick wins earn their place. They return value while larger work is still uncertain, they teach the organization how to run AI in production, and they give skeptical colleagues something real to judge. John Kotter’s study of corporate transformations found that most people will not stay with a long change effort unless they see compelling evidence within 12 to 24 months that it is producing results³. Quick wins are often that evidence.

Kotter added a distinction that matters more than the timing. Creating short-term wins is different from hoping for them. In one manufacturer he described, the guiding team chose its visible win about six months into the effort, because it met several criteria at once: it could be delivered fast by a small team committed to the new direction, and it showed the direction working³. A quick win chosen that way serves the larger agenda. A quick win chosen only because it was easy serves itself. How to pick a first project is the subject of Where Should We Start With AI?; the portfolio question is what the whole set of quick wins adds up to.

Measure quick wins against a baseline: time to value, adoption, unit cost, quality and the user’s experience. Fund them in small, short tranches, with the security, ownership and controls any production system needs. Small does not mean uncontrolled.

The trap is a portfolio made only of them. Each department buys or builds its own assistant, each with its own vendor, data connection and security review, and the enterprise ends up with dozens of modest tools and no shared capability. Everyday productivity tools are also among the easiest kinds of AI for competitors to copy, because the same products are on sale to everyone. A portfolio of quick wins can optimize today’s business very well and still leave tomorrow’s unfunded.

Strategic bets: paying for options

A strategic bet is best understood as the purchase of an option: a modest sum that buys the right, but not the obligation, to make a much larger investment later if the evidence supports it. That changes what success means. An early bet that ends with a clear “no” at low cost has done its job, because it stopped the enterprise from making the large investment blind.

It does not change the need for evidence. Rita McGrath and Ian MacMillan’s discovery-driven planning, built for ventures whose outcomes cannot be forecast, asks the team to write down the assumptions the plan depends on, test the most important ones first and release money only as milestones are met⁴. For an AI bet the assumptions are usually about customer response, the quality the model can reach on real data, unit economics and the controls the use will require. Prioritizing the AI Portfolio set out staged funding and stop rules; here they are what turns an uncertain idea into a bet rather than a hope.

Two questions keep a bet honest. What future option are we buying? What evidence would change our mind? If the team cannot answer the second before the money moves, the bet has no exit, and the main risk of this lane follows: the permanent experiment, renewed every year because nobody defined what failure would look like.

Transformation: changing how value is created

A transformation initiative is not a large technology project. Its outcome is a different way of working: an end-to-end process redesigned around what AI can do, with new roles, new decision rights, new data and new incentives. A model is a component of it. Technology deployment is never its final measure; business outcome, process performance and adoption are.

Transformation has a shape that confuses ordinary financial reviews. Erik Brynjolfsson, Daniel Rock and Chad Syverson showed that general-purpose technologies such as AI need large complementary investments, mostly intangible: new processes, skills and business models. Because these are poorly measured, productivity looks worse than it is in the early years and better than it is later, when the benefits are harvested. They called the shape the productivity J-curve. The understatement is not small: adjusting US data for the intangibles tied to computer hardware and software put total factor productivity 15.9 percent above the official measure by the end of 2017⁵. This is the firm-level face of the productivity paradox that The Four Industrial Revolutions described: the lag is the normal pattern, not a sign of failure.

The practical consequences are three. Fund transformation over several years, but still in stages where uncertainty remains. Judge the early stages on leading indicators, such as process steps redesigned, data made usable and people working the new way, rather than on profit that cannot yet appear. And give it a business owner, not a technology owner, with the product, data, operations, finance, risk and change leaders around the table. Kotter’s warning applies at the other end too: new ways of working take five to ten years to settle into a culture, and declaring victory at the first clear improvement invites regression³.

Different money, gates and measures

If the three types buy different things, they cannot pass the same test. A single return-on-investment hurdle, applied to everything, approves quick wins, starves bets that cannot yet show a return, and kills transformations in their second year, at the bottom of the curve.

The disciplines underneath are the same in all three lanes: value, feasibility, risk, cost and adoption. What differs is the threshold and the question. Governance follows the same logic. Quick wins need speed with the minimum controls the risk requires. Bets need evidence discipline and someone other than the sponsor reading it. Transformations need enterprise coordination, because they touch many functions at once.

Initiatives change type

The categories describe what an investment is for at a moment, not a permanent label. A drafting assistant for one team can reveal that the whole process it sits in should be redesigned, and become the first stage of a transformation. A strategic bet that passes its gates stops being a bet and becomes either a scaled product or the core of a transformation. A transformation, once it has built shared data, platforms and skills, makes the next round of quick wins cheaper.

That loop is possible, not automatic. It needs a review that asks, for each initiative, whether its type has changed, and moves its funding rules when it has. A quick win that has quietly become a transformation, still funded and governed as a small tool, is a common reason large AI programs drift.

Choosing the mix

There is no correct percentage. Nagji and Tuff were explicit that their 70-20-10 finding was an average across industries and regions, not a formula. In their data the right balance varied with the industry, with a company’s competitive position and with its stage of development: technology companies spent less on the core, consumer goods companies had little transformational activity, and a lagging company might rationally take more risk than a leader¹.

For AI, four factors usually set the mix. Strategy decides how much the enterprise needs new options rather than a better core. Maturity decides what it can carry: as The AI Maturity Model showed, an organization early on the ladder lacks the production disciplines that large bets and transformations need, so it may lean toward quick wins and controlled experiments while it builds them. Capacity decides how much can run at once; a single transformation can absorb the same data and change specialists as a dozen quick wins. Risk appetite decides how much uncertainty the board will fund, and it should be stated, not inferred from whichever sponsor argues best.

Nagji and Tuff’s practical advice translates directly. Decide the ratio you believe the strategy needs, measure how far the current portfolio is from it, and plan to close the gap¹. Many organizations have never counted. When they do, the actual mix is often whatever accumulated from individual approvals.

Five ways a portfolio fails

Unbalanced AI portfolios tend to fail in five recognizable ways.

The first two are mirror images. A portfolio of only quick wins improves today’s processes and creates no differentiation. A portfolio of only transformation spends heavily for years before anyone can see a result, and loses support before the benefits arrive. The third, ungated bets, confuses uncertainty with a license to keep spending. The fourth is common in practice: nobody decided the mix, so it reflects which sponsors were most persuasive. The fifth is the quietest. The technology is live and the dashboards are green, but people still work the old way, so the business outcome never comes. At each review, ask the same questions of every lane: what changed, what we learned, what value was proven, what it cost, and whether funding should change.

Story: the fertilizer maker’s three envelopes

A fertilizer maker with five plants is in its second year of serious AI work. Three quick wins are live and measured: a tool that turns maintenance technicians’ free-text notes into coded fault entries, a query assistant for shift supervisors and an assistant that matches freight and bagging-supplier invoices. Each paid back within a year. The budget committee has room for one major new commitment, because the shared data and engineering team of about twenty people is already stretched. Three proposals arrive.

Proposal A is the safe choice, and the departments asking for it can point to the three that worked. Proposal B could change how the company sells to the farm-supply dealers it serves, by recommending each dealer’s seasonal order and pricing it for that dealer, but nobody knows how dealers will respond or what the model will cost to run at scale. Proposal C would move maintenance on the granulation and bagging lines from fixed intervals toward planning based on the condition of each machine, using sensor data the lines already record. It would change maintenance planning, spare-parts stocking and the work of the engineering teams, and no interval on a line handling ammonium nitrate could change without the process-safety team signing off. Before reading on, decide which one you would fund.

The committee funded none of them as proposed. It approved three of the ten quick wins, chosen because they would also build pieces the maintenance work would need, such as cleaner equipment histories. It funded the offers bet only for a twelve-week test with a few hundred dealers in one region, with the stop threshold written in advance: a minimum lift in revenue per dealer and no rise in cancelled orders. And it started the maintenance transformation with its slowest work first: making equipment data usable and running one type of line as a first stage, owned by the company’s director of engineering rather than by the technology function.

A year later, the offers test had missed its threshold and was stopped; its dealer-data work was kept. The three new quick wins were live. The first maintenance stage was in a controlled trial on the bagging lines, still years from its full benefit, with leading indicators the committee could read each quarter. The decision that mattered was not which envelope to open. It was the mix, and the different rules attached to each part of it.

What this means for leaders

Treat the AI portfolio as three kinds of investment, not one list. Name the type of every funded initiative by what it is meant to buy: value now, an option, or a new way of working. Then give each type its own funding rhythm, its own gate question and its own measures, so that quick wins are held to proven value, bets to validated assumptions and transformations to evidence that the enterprise is actually changing.

Then decide the mix on purpose. Start from strategy, temper it with maturity and capacity, and state the risk appetite out loud. Choose quick wins that also build what the larger work will need. Reclassify initiatives when they change type, and rebalance when evidence, technology, competition or capacity moves. The executive question is simple to ask and hard to answer: what value do we need now, what options must we create, and which changes must we start today because they take years?

Check yourself

1. Every AI initiative should clear the same return-on-investment hurdle.MythFact
2. A portfolio of only quick wins is the safest AI strategy.MythFact
3. A strategic bet that ends in a clear, cheap “no” can be a success.MythFact
4. Research shows a 70-20-10 split is the right allocation for AI.MythFact
5. An initiative can move from quick win to transformation.MythFact
6. A transformation is finished when the new system goes live.MythFact

Reflection: count your own mix

What comes next

Balancing the mix decides what kinds of investment the enterprise makes. It does not move any of them out of the pilot. A quick win still has to reach production, a strategic bet still has to produce its evidence, and a transformation still has to operate every day. The next chapter, From AI Pilot to Production to Scale, sets out the path that turns a promising experiment into a dependable enterprise capability.