The Ansoff Matrix 2026: Put a Probability on Each Box
Four growth routes ordered by risk, which is really a claim about success rates. Calder Instruments, and the exact rate at which that ordering reverses.
- The four cells sort growth by how far it moves from what you already sell. Ordering them by risk is a claim about success rates, and the matrix never asks for one.
- Calder Instruments has four routes. Ranked on expected value net of capital, penetration returns 4.4 million dollars and diversification loses 9.0 million.
- Raise the diversification success rate to 35.8 percent and it matches penetration. One unstated input decides the whole ordering.
- Deloitte publishes a case where a car maker enters the bus market, which is a new product and a new market at once. The cells are not exclusive in practice.
- Kearney publishes the instruction to avoid forcing a solution into a standard framework, and names the three things the structure must fit.
Existing product in the existing market, existing product in a new market, new product in the existing market, new in new. The grid sorts growth by distance from what you sell today. Its famous property is that the four are ordered by risk, and risk is a number the grid never asks you to supply.
Four Ansoff matrix routes, and the claim hidden in the order
The four routes come from Igor Ansoff's article "Strategies for Diversification", published in the Harvard Business Review in 1957. We did not read a copy hosted on a primary domain, so the attribution stops at author, title, publication and year. The name Ansoff matrix came later and we do not put it in his mouth.
The two axes are definitional. A product either exists in your line or it does not. A market either has your product in it or it does not. So the grid cannot be wrong about where a route sits, which is unusual among strategy frameworks and is the source of its durability.
The risk ordering is a different kind of statement. Saying diversification is riskier than penetration is saying that the probability of the intended outcome falls as you move away from the diagonal. That is an empirical claim about success rates, and the grid supplies neither the rates nor a way to get them.
This matters in a case because the ordering by risk and the ordering by value are different orderings. A route can be the riskiest and still be the best, if the payoff scales faster than the probability falls. Deciding that requires three numbers per cell, and the grid asks for none.
Reading a worked example is not producing one.
Score my answerCalder Instruments, and three numbers per Ansoff matrix route
Calder Instruments makes industrial pressure sensors, with 80 million dollars of revenue in one region. The board wants growth and has four proposals on the table. Each carries a capital request, an annual contribution if it works, and a success rate estimated by the sponsoring team.
| Route | Ansoff cell | Capital, $M | Annual contribution if it works, $M | Success rate |
|---|---|---|---|---|
| Win share from two local rivals | Existing product, existing market | 4 | 2.4 | 70% |
| Enter three neighboring countries | Existing product, new market | 9 | 5.0 | 50% |
| Add a wireless sensor line | New product, existing market | 14 | 8.0 | 40% |
| Buy into process analytics software | New product, new market | 26 | 17.0 | 20% |
The success rates are the sponsoring teams' own, and in a real case they are the number you ask the interviewer for. Nobody publishes them. Our research file says so plainly, because the temptation is to treat a plausible figure as a fact.
Note the pattern in the table. Capital rises by a factor of 6.5 from the first route to the last, and annual contribution rises by a factor of about 7.1. The payoff really does scale with distance. Whether it scales fast enough is what the next section settles.
The expected value of each Ansoff matrix box, over five years
Take a five year horizon, no discounting, so the arithmetic stays checkable by hand. Expected contribution is five times the annual figure times the success rate. Subtract the capital to get the net.
| Route | 5 x annual, $M | Times success rate, $M | Less capital, $M | Net per dollar of capital |
|---|---|---|---|---|
| Penetration | 12.0 | 8.4 | 4.4 | 1.10 |
| Market development | 25.0 | 12.5 | 3.5 | 0.39 |
| Product development | 40.0 | 16.0 | 2.0 | 0.14 |
| Diversification | 85.0 | 17.0 | -9.0 | negative |
Every line by hand. Penetration: five years at 2.4 million a year is 12.0 million; 70 percent of 12.0 million is 8.4 million; less 4 million of capital leaves 4.4 million; 4.4 over 4 is 1.10 per dollar. Market development: five years at 5.0 million is 25.0 million; 50 percent of 25.0 million is 12.5 million; less 9 million leaves 3.5 million; 3.5 over 9 is 0.39, rounded to two decimals. Product development: five years at 8.0 million is 40.0 million; 40 percent of 40.0 million is 16.0 million; less 14 million leaves 2.0 million; 2.0 over 14 is 0.14, again to two decimals. Diversification: five years at 17.0 million is 85.0 million; 20 percent of 85.0 million is 17.0 million; less 26 million leaves a loss of 9.0 million.
At these inputs the risk ordering and the value ordering agree, and the agreement is the point worth saying out loud. The matrix looked like it had ranked the routes. What ranked them was the success rate column, and the matrix does not contain that column.
Can you hold the framework under a clock?
The success rate at which the order reverses
Now push on the one input nobody knows. Hold everything else and raise the diversification success rate.
At 35 percent, expected contribution is 85.0 million times 0.35, which is 29.75 million. Less 26 million of capital, the net is 3.75 million. That already beats market development at 3.5 million and product development at 2.0 million.
Solve for the rate at which it matches penetration. Set 85 times the rate, less 26, equal to 4.4. That gives 85 times the rate equal to 30.4, so the rate is 0.3576, which is 35.8 percent.
One number, unstated in the prompt and unrequested by the grid, moves the riskiest cell from worst to best. The band that matters is narrow: below 30.6 percent diversification loses money outright, since 26 over 85 is 0.3059, and above 35.8 percent it leads the field. Between those two figures it is profitable and still ranked last.
Oliver Wyman publishes the instruction that makes this concrete in an interview. Under its own step called pick an approach, its preparation page says to "Show your client why it's better than the alternatives", checked 30 September 2026 at oliverwyman.com. A candidate who lists four cells has shown the alternatives and compared none of them. A candidate who names 35.8 percent has done the comparison and named the number the client would have to believe.
A published case interview that sits in two cells at once
Deloitte publishes a growth case in which an Asian car maker considers entering the bus market, after five strong years and a weak forecast, checked 30 September 2026 at deloitte.com. A bus is a new product for a car maker, and bus buyers are a new market. The prompt occupies the top left and bottom right of the grid simultaneously.
That is a structural problem rather than an edge case. The cells describe two independent binaries, so a real growth move usually lands in more than one, and the risk ordering then has nothing to say about it. The firm's own model answer does not reach for a grid. Its great answer tier adds a multi year contract commitment, a joint venture, expansion beyond Asia, and higher margin services. Those are entry modes and hedges, which is the dimension the matrix has no axis for.
The data sheet behind that case's margin question is an image, so we reconstruct none of its numbers. What the case supports is the shape of the problem, which is enough.
Deloitte's general advice points the same way. Its five published steps are "Understand the issue", "Identify the underlying assumptions", "Summarize specific issues and findings", "State your recommendations", "Outline next steps and expected results/impacts", checked 30 September 2026 at deloitte.com. Step two is where the success rate belongs, and it is a step about assumptions rather than about domains.
Using it without losing the points
The weights below are ours, read from the rubric code in src/lib/grading. No consulting firm publishes a rubric of its own.
- Coverage45
- Separation25
- Depth20
- Concision10
| Case math | Answer 40 · Working 25 · Units 15 · Interpretation 20 |
|---|---|
| Idea generation | Breadth 30 · Buckets 30 · Specificity 20 · Prioritisation 20 |
| Market sizing | Scope 25 · Assumption chain 30 · Arithmetic 25 · Sanity check 20 |
| Exhibits | Observation 25 · Implication 25 · So what 30 · Next test 20 |
| Synthesis | Answer first 35 · Evidence 25 · Risk 20 · Next step 20 |
Four steps that keep the grid useful on a growth prompt.
- Name the four cells once, in ten seconds. They are a completeness check on the option set, and they are good at that.
- Ask for capital and contribution per route. Two numbers per cell, and the interviewer usually has them.
- Ask for or assume a success rate, out loud. State it as an assumption and say you will test it.
- Solve backwards for the rate that flips the answer. That number is your recommendation's hinge, and it is what a partner remembers.
Coverage 45 is where the grid earns its keep, because it is hard to miss a growth option with it on the page. Depth 20 is where an unnumbered version dies: four labeled boxes with no arithmetic cannot reach a recommendation, and a growth prompt is a resource allocation question. Kearney's page puts the same warning in its own words, saying to avoid forcing a solution to fit a standard framework and naming the three things it must fit instead, which are the problem, the business and the objectives, checked 30 September 2026 at kearney.com.
For the portfolio question that gets confused with this one, see Ansoff against the growth share matrix and the growth share matrix itself. For the wider family of prompts, growth strategy cases and revenue growth cases both start from the arithmetic. Our structuring drill times the habit above and returns the four dimensions separately, and how an answer is assessed sets out the weights.
See your answer scored, dimension by dimension.
Reading a worked example and producing one are different skills. Work a full case and see it scored, dimension by dimension.
Sources
Primary sources only: the firm, the test vendor, or the institution concerned. Each carries the date we last read it.
- usrecruiting.deloitte.com/leadautoa published growth case in which an Asian car maker enters the bus market, occupying two Ansoff cells at once, and the firm's good answer and great answer tierschecked 2026-09-30
- deloitte.com/us/en/careers/join-deloitte/recruiting-tips/case-and-scenario-interview-tips.htmlthe published five step approach, and the fact that none of the five steps names a domain of analysischecked 2026-09-30
- kearney.com/careers/interviewing/crack-the-casethe instruction to avoid forcing a solution into a standard framework and the three things the structure must fitchecked 2026-09-30
- oliverwyman.com/careers/entry-level/interview-preparation.htmlthe published instruction, under pick an approach, to show the client why the chosen approach beats the alternativeschecked 2026-09-30
Edits the firm and case guides on this site. Reads each claim about a recruiting process back to the firm's own page before it is published, keeps the source ledger dated, and owns the corrections.
How this guide was made
- Written against a published editorial process, not from memory. Every claim about a firm's process is either linked to that firm's own page or absent. We say what we could not verify rather than filling the gap.
- Each page carries a dated source ledger, below. When a firm changes a programme or a deadline, the date tells you how far to trust what you just read.
We look for first-hand candidate accounts for every firm guide. Where we found none we could source, the page says so and carries the firm's own words instead. Nothing here is dressed up as testimony, and if your own experience contradicts what you read, tell us and we will correct it and date the change.
Something wrong or out of date here? Tell us and we will correct it and date the change.