ROI Analysis for Major Investments: The Stress-Test You Skip

ROI Analysis for Major Investments: The Stress-Test You Skip

In Q3 of 2024, a $4M Australian skincare brand approved a NetSuite migration. The proposal said $180K all-in, nine months to live, and a payback inside eighteen. The CEO believed the numbers. The CFO believed the numbers. The deployment partner had a deck full of three brands that had done it cleanly. The board signed off in twenty minutes.

By month nineteen, the project had cost $410K, eaten the head of ops, killed two reporting cycles, and forced an emergency line of credit through Q4. NetSuite eventually went live. The brand had survived. Nobody on the leadership team had any appetite for a second major capital project for at least two years.

The CEO told me this story over a flat white in Surry Hills. He kept asking what they should have caught. The honest answer, which I gave him, was that they had stress-tested the wrong question. They had pressure-tested the vendor. They had not pressure-tested their own model. ROI analysis for major investments is supposed to gate this kind of decision, and theirs had not.

This article walks through the failure mode that produced that $230K overrun, the hidden costs that single-point ROI calculations systematically miss, and the playbook I now run with operators before any project above $100K gets a signature.

The $230K Overrun That Looked Green on the Spreadsheet

The standard ROI model for a major capital project plugs in one expected cost figure, one expected benefit figure, one expected timeline, and treats the resulting payback number as the answer. That is the spreadsheet that produced the NetSuite go-ahead in Surry Hills. It is also the spreadsheet that has produced almost every blown capital project I have audited in the last decade.

The math is straightforward. The cognitive failure underneath is not.

Daniel Kahneman and Dan Lovallo, writing in Flyvbjerg base rates, describe the pattern as the "inside view." Operators forecast the project in front of them by walking through its internal logic: vendor scope, internal team availability, the milestones they expect to hit. They do not anchor against the base-rate failure data of similar projects executed in the wider economy. The inside view feels rigorous because it is detailed. It is wrong because it ignores the only data set that actually predicts the outcome.

That base rate is brutal. The Standish CHAOS report tracking software project outcomes finds that fewer than one in three IT projects deliver on time, on budget, and with the originally promised functionality. Roughly one in five fails outright. The remaining half land somewhere in the wreckage of overrun, descoped, or limped to live. The pattern is so consistent across two decades of CHAOS reports that you can plan to it.

The Surry Hills CEO had not built a bad spreadsheet. He had built a spreadsheet that asked the wrong question. "What does this project cost if everything we plan goes to plan?" is not the question. The question is, "What does this project cost given what we know about projects like this in the wild?"

Single-point ROI is a polite confession that you have not asked that second question. The villain here is not the operator. It is the spreadsheet pattern that pretends a major capital decision can be reduced to one cost line, one benefit line, and one payback number.

Why the Math Doesn't Work: The Hidden Cost Stack

When a single-point ROI model fails, it does not fail by 5 percent. It fails by 50 to 130 percent. The reason is that three layers of hidden cost compound on top of each other, and the standard model only sees the first one.

The first layer is direct cost overrun. Vendor invoices come in higher than scope. Custom development chews up budget. Internal team time gets re-routed and the opportunity cost shows up as missed revenue elsewhere. This is the layer most operators eventually reconcile, three months too late, in a quiet meeting with the CFO. Roughly a third of capital project failures live entirely in this layer.

The second layer is timeline slippage. Every month the project runs past its original go-live is a month the promised benefits do not materialise. If you modelled an eighteen-month payback and the project lands six months late, you have lost a third of the cumulative cash benefit before the system even works. That benefit is real, immediate, and almost never priced into the original business case. The Standish data shows that delay is the most common form of project pathology, more common than direct overrun.

The third layer is adoption lag. The system goes live. The team has not been retrained. The data migration has small but persistent gaps. Reporting confidence drops for two quarters while everyone reconciles. The promised benefit, the one that justified the project in the first place, takes another six to twelve months to actually arrive. The Panorama ERP report finds that fewer than half of ERP customers report achieving more than 50 percent of expected benefits within a year of go-live. McKinsey's work on large capital projects in McKinsey capital projects sees the same shape in physical infrastructure: the average megaproject delivers late and over budget, and the realised benefits trail planned benefits by years.

Stack those three layers and the failure mathematics get savage. A project quoted at $180K and an eighteen-month payback that overruns by 80 percent, slips by six months, and underdelivers on year-one benefits by 40 percent is not running 20 percent over. It is running closer to 130 percent over the original NPV expectation. Most operators have never seen the math laid out that way because their spreadsheet stops at line one.

This is the gap The Major Investment Stress-Test Playbook is built to close.

The Major Investment Stress-Test Playbook

I deploy The Major Investment Stress-Test Playbook with operators before any capital project above $100K gets approval. The playbook does three things the standard ROI model does not. It forces three explicit scenarios. It applies a base-rate adjustment from comparable real-world projects. It commits the operator to a written kill criterion before the project starts.

Scenario one is the Base Case. This is the model your vendor or internal lead built. Take it as given for the moment. Do not torture it. Just label it.

Scenario two is the Slipped Case. Apply explicit base-rate adjustments drawn from comparable projects. For ERP work, I use Panorama and Standish numbers as the anchor: assume cost runs 25 to 40 percent above plan, timeline runs 30 to 60 percent above plan, and year-one benefit realisation lands at 40 to 60 percent of plan. For physical capex like a warehouse build, McKinsey's overrun data anchors a 20 to 40 percent cost adjustment. For private-label launches, I use the operator pattern that fewer than half of new SKUs hit their year-one volume target. The Slipped Case is not pessimism. It is what the wider economy tells you to expect.

Scenario three is the Failed Adoption Case. Assume the project goes live but the promised behavioural change does not happen. The new system is used for the old workflows. The new SKU launches and sells through at half the velocity. The new warehouse runs but the team never restructures the pick paths to use the layout. Now run the ROI math. This is the case nobody wants to model. It is also the case most likely to actually happen, and the only one that tells you whether the project is worth doing if the team underexecutes.

Three scenarios. Three NPV outputs. Three honest payback periods.

Now apply the kill criterion. Before the project starts, the leadership team agrees in writing to a single number: the trip threshold beyond which the project will be paused, descoped, or killed outright. For software work it is usually a cost-overrun percentage past which the original business case fails the Slipped scenario. For inventory or capex projects it is usually a timeline slippage threshold past which the working-capital cost eats the original return.

The Major Investment Stress-Test Playbook works because it converts hope into a contract. The pre-committed kill criterion is the single most powerful artifact in the entire process. It is the thing that lets a CFO walk into month fourteen of an over-running project and say "we agreed in writing we would pause if we hit this number," instead of having a fresh political fight about whether the project is still worth finishing.

Warren Buffett has been writing about this discipline for fifty years. His Buffett shareholder letters repeatedly hammer the same idea: the discipline of rejecting projects that fail a clearly-stated hurdle test is more valuable than the discipline of picking the right project. Most operators have never read the letters and reinvent the lesson the hard way.

Execution: Day 0 to Day 90

The playbook lives or dies in execution. Here is the cadence I run with operators.

Days 0 to 14 is the Comparable Projects File. Before any model gets built, the project sponsor has to assemble five to ten comparable projects in operator-grade detail. For an ERP migration that means five recent NetSuite or Microsoft Dynamics rollouts in physical-product brands of similar size, with documented cost and timeline outcomes. For a warehouse move it means five 3PL transitions or owned-warehouse builds. For a private-label launch it means five recent line extensions in adjacent categories. The sponsor pulls these from peer networks, vendor case studies, and published research like Panorama and Standish. The artifact is a one-page table with columns for planned cost, actual cost, planned timeline, actual timeline, and a one-sentence note on what went wrong. If the sponsor cannot produce this file, the project is not ready for a model. Period.

Days 15 to 45 is Model Construction. The sponsor builds the three-scenario ROI model. Base Case gets the vendor numbers. Slipped Case applies the base-rate adjustments derived from the Comparable Projects File. Failed Adoption Case strips out the behavioural-change benefits and reruns the math. Each scenario produces an NPV at the operator's hurdle rate. Aswath Damodaran's body of work on capital budgeting, surveyed in Damodaran on hurdle rates, is the cleanest reference for setting that hurdle rate honestly: it should reflect the brand's real cost of capital, not a wishful number pulled from a textbook.

Days 46 to 60 is the Decision Memo. The sponsor writes a two-page memo. Page one summarises the three scenarios, the assumptions in each, and the resulting NPVs. Page two states the kill criterion in writing: the cost threshold, the timeline threshold, and the adoption threshold beyond which the project pauses for a full re-review. The CEO signs. The CFO signs. The project sponsor signs. The kill criterion goes into a calendar reminder for monthly check-in.

Days 61 to 90 is the Pre-Mortem. Before any contract is signed, the leadership team runs a single ninety-minute pre-mortem session. The framing question is the standard one: "It is eighteen months from now. The project has failed. What killed it?" Each leader writes their top three answers privately, then the group consolidates into a single risk register. Each risk gets an owner and a leading indicator. The leading indicators feed the monthly tripwire check.

Three months of work. One signed memo. One kill criterion. One risk register. That is the prep cost for a $100K-plus capital decision. Operators always think this sounds like too much process. The brands I have seen reverse a bad project in month four are the ones who paid that cost in months one through three.

From Single-Point Hope to Base-Rate Honesty

The brands compounding through the $1M to $10M band are not the ones with the best vendor selection. They are the ones who model their major investments the way a credit officer models a loan. Three scenarios. A base-rate adjustment grounded in real comparable data. A kill criterion signed before the contract. A pre-mortem on the calendar.

The Surry Hills CEO has run two more capital projects since the NetSuite story. Both of them used a version of The Major Investment Stress-Test Playbook. The first was a 3PL transition originally costed at $90K and twelve weeks. The Slipped scenario said $135K and seventeen weeks. It landed at $124K and sixteen weeks. The model was honest. The team did not blink. The kill criterion never tripped.

The second was a private-label launch the team eventually killed at month two. The Failed Adoption scenario showed that even at full execution, the SKU paid back inside three years only if it stole shelf from a higher-margin existing line. The team paused the launch, redesigned the formulation, and relaunched eight months later in a different category. The Stress-Test Playbook did not pick the second SKU. It just refused to greenlight the wrong first one.

That is the shift worth funding. Not faster decisions. Not more decisions. Better-priced decisions, anchored in base rates, defended by a written contract. The same pattern Buffett describes across the Buffett shareholder letters reads as common sense at $50B and as hard discipline at $5M, but the math is identical at both ends.

If your next capital project above $100K does not survive a Slipped scenario at your honest cost of capital, you do not have a project worth signing. You have a vendor pitch you have not yet stress-tested.