The Carr And Erwin Method How Two Mavericks Rewrote The Rules Of Product Strategy And Execution

What the Carr and Erwin Approach Actually Looks Like in Practice

Most people encounter the Carr and Erwin Method through the book they co-authored on business modeling and financial forecasting. The core idea is straightforward: build bottom-up financial models where every line item traces back to a driver, and connect those drivers to strategic scenarios instead of treating them as static assumptions. The reason it caught traction in product circles is that traditional spreadsheets tend to treat revenue and cost as guesses wrapped in percentages. Their framework forces you to decompose everything into operating-level mechanics. Units sold, conversion rates, average selling price, churn, ramp time per hire, infra cost per unit. Each driver becomes a testable hypothesis instead of a placeholder number.

The Carr And Erwin Method How Two Mavericks Rewrote The Rules Of Product Strategy And Execution

Setting this up starts with choosing your time horizon. For most product strategies you are working with monthly granularity over at least twenty-four months, sometimes thirty-six if you have long sales cycles. Quarterly data smooths over too much operational reality. Next you build a driver map. This is not a chart you put in a slide deck. It is a working document, usually in a spreadsheet or a lightweight tool like Ableflesh or even a well-structured Airtable base, that links your top-line output to underlying inputs. Revenue becomes units shipped multiplied by ASP. Units shipped becomes pipeline qualified leads times close rate times seat capacity. Seat capacity ties back to hiring plans, training duration, and attrition. Each node has a source. That source should be recent, specific, and cited. Scenario logic follows after the driver map is solid. You define at least three scenarios: base, upside, downside. The trick that most teams skip is explicitly defining the triggering condition for each scenario shift. If pipeline growth drops below fifteen percent month over month for two consecutive months, you move from base to downside. That condition needs to be measurable from your existing data tools. If you have to manually check five dashboards every week to know which scenario you are in, the model will not be used.

Rolling forecasts replace annual budgets here. You re-run the model monthly, keeping drivers that have empirical backing fixed and allowing forward-looking drivers to adjust based on new data. This cuts budgeting cycles down from about six weeks of meeting-heavy negotiation to roughly four days of model updates and review. One detail that matters more than people admit is versioning. Your model should have a dated snapshot for every scenario change. When product leadership asks why revenue guidance shifted, you need to show the exact model state from three weeks ago with the changed driver clearly marked. Without versioning, you end up in conversations where no one can agree on what assumption was actually used. I spent about fourteen months building a Carr and Erwin style model for a B2B SaaS product that sat in a mid-market niche. The breakdown went smoothly until we hit the channel partner commission structure. Partners were paid on gross bookings, not recognized revenue, and the payment terms varied by tier. My initial driver mapped commission as a flat percentage of projected bookings. That produced a cost line that was roughly twenty-two percent too low in months four through nine, which completely skewed the runway calculation and made us look like we had cash to spare when we did not.

The fix was granular. I segmented partner tiers by contract date, pulled actual payment terms from the legal repository, and built a lag function that mapped booking-to-cash timing per tier. It added about six hours of work upfront and removed an entire category of forecast error. The commission line stabilized within three percent of actuals for the rest of the year. Here is something people rarely mention about this approach. The biggest bottleneck is not the modeling itself. It is getting engineering and product to agree on driver ownership. Every driver needs a clear owner who can update it monthly. In my experience, about forty percent of driver slots go unowned because the person responsible changes roles, leaves, or considers their job complete once they deliver the initial number. I started using a RACI column inside the model file itself and added a soft review cadence where each owner had to confirm their driver value before the monthly roll-forward. This cut stale-driver issues from roughly one in five nodes down to about one in twelve. Another counter-intuitive point. Bottom-up models perform worst when you try to model everything at the same level of detail. You should deliberately oversimplify low-impact drivers and overspecify high-impact ones. In one engagement, the team modeled customer support cost per ticket down to the hour with worker rate assumptions, while using a single percentage guess for cloud infrastructure cost, which was the largest variable line item. That is backwards. Cloud cost should have been built from compute hours, storage volumes, and pricing tiers. Support cost could have stayed a ratio. The final model accuracy improved noticeably once we reversed that allocation.

There are real limitations to this method. It does not handle sudden market shocks well. If a competitor drops price by half or a regulatory change alters your TAM overnight, your driver map still contains last quarter's assumptions and the model will produce confident but wrong numbers. You need a separate shock layer, usually a manual override that flags the affected drivers and recalibrates them externally before feeding back into the base model. Adding that step takes about thirty minutes per major event. The method also requires a baseline level of financial literacy across the team. If your product managers treat the model as a finance artifact instead of a living operational tool, you will get model drift within sixty to ninety days. I have seen teams abandon the approach after three months because the original architect left and nobody understood the driver relationships well enough to maintain it. If your organization has fewer than fifteen people in product and go-to-market combined, a full Carr and Erwin implementation usually costs more than it saves. A simplified version with five to eight core drivers and quarterly reviews often delivers the same insight with a fraction of the overhead. Tools like Plainly or even a carefully structured Google Sheets file with scenario tabs can cover that need without the rigidity of a dedicated modeling platform.

For teams ready to go further, the practical stack tends to be a driver map in a shared spreadsheet or database, a rolling forecast schedule built into the same environment, and a simple dashboard that shows actual versus model variance by driver. That last piece is what keeps the model honest. If your closest driver has a twelve percent variance for two months running, you either adjust the driver or you acknowledge the model gap and stop treating the output as truth. The Carr and Erwin approach is not a strategy replacement. It is a discipline layer that makes your strategy testable. It works best when leadership is willing to let the model surface uncomfortable answers instead of editing assumptions to match desired outcomes. Used well, it removes about seventy percent of the argument from quarterly planning sessions. The remaining thirty percent is usually about priorities, not math.