Software teams should be judged not only by delivery speed or cost, but also by the business changes their software produces, argues software development author James Shore. In the fourth part of his series on AI’s impact on development, he proposes modeling and forecasting the business outcomes of major initiatives.
Shore’s article was originally published on his Successful Software blog. Earlier installments examined the principles of assessing impact, delivery speed and the unintended consequences of using AI. This installment instead considers how to plan and prioritize the expected value of development work.
Cost is not the same as value
Companies often spend considerable time estimating development costs while giving much less thought to how software could create more value, Shore argues. There is no guaranteed relationship between the two: a $20 million project can produce worthless software, while a $2 million project can transform a business. In his example, cost reductions generally deliver incremental improvements, whereas increases in value can make a much larger difference.
The question may become particularly important as AI spreads. Engineering leader Lee Eason suggests that customers have a limited capacity to absorb new features. If writing code becomes faster, the bottleneck may shift from development to customer attention: even if more features are released, users have their own work and may not have time to engage with them all. Shore therefore stresses selecting work that is genuinely valuable.
Why model business impact?
A company’s financial results can be measured at a broad level, but attributing a particular share to one development initiative is difficult. If purchases rise by 10% after a feature launch, sales incentives, a marketing campaign and partners’ activities may also have contributed. Revenue figures can also take months to arrive, particularly where business-to-business sales cycles are long.
Shore therefore recommends modeling value rather than measuring it directly. Parts of the model can be paired with proxy metrics that provide faster feedback and help refine forecasts over time. As one example, he cites Dave McClure’s “Pirate Metrics”: acquisition, activation, retention, referral and revenue, abbreviated AARRR. For a new feature, a team might track how many users try it and how many continue using it.
Other indicators can also be useful. If a usability initiative is expected to reduce customer churn, a team could watch changes in support contacts and whether customers cite usability problems less often when canceling.
Organizational support and a financial model
Building a model requires participation beyond the development team. Sales can help estimate the effect on new business; customer success can assess retention and upselling; and marketing can address lead generation. Partners may contribute to any of those areas.
In a previous engineering leadership role, Shore took roughly two years to establish the approach, with most of that time spent winning people’s support. Working with the product leader, he first involved the chief product officer and then the chief executive. With the chief executive’s backing, financial models became an official part of prioritizing major initiatives, which the company called “product bets.”
The first forecasts were exaggerated and insufficiently thought through. A chief financial officer who joined later helped critique and improve the models and obtain the necessary business information. Executive backing and personal relationships with colleagues helped overcome reluctance to make forecasts. Shore expects the process to become routine over time as teams develop templates and experience.
The proposed model is a spreadsheet that calculates net present value over a set period. Shore’s team used five years: it estimated each year’s income and expenditure, discounted future amounts and added the results. The aim, he says, is not to get every detail perfect but to clarify what financial benefit is expected and why. Given the uncertainty, teams can prepare optimistic, pessimistic and best-guess versions.
For a sales-oriented business, he suggests examining the following items:
- New sales: how many customers the initiative might gain, how many the company might lose without it and what each customer is worth.
- Upselling: how many customers might upgrade and how much additional revenue those upgrades could bring.
- Retention: customer value, the churn rate and the proportion of at-risk customers the initiative might retain.
- Cost savings: for example, reductions in internal work or support tasks. Shore urges caution, arguing that sales and retention often offer more value than further cost optimization.
- Ongoing expenditure: operating costs after launch, such as token costs for an AI-powered service, estimated from the number of sessions and their average cost.
- Development costs: primarily labor and AI token expenses.
For development spending, Shore favors a ceiling and milestone approach. The net present value of the rest of the model can indicate roughly how much the company can spend while keeping the project a worthwhile investment. The team can then work iteratively using a “build-measure-learn” approach and consult an executive steering committee at each milestone on whether to continue or cut its losses. In his view, this provides more opportunity to steer the work toward value than a traditional up-front estimate.
A model informs decisions; it does not make them
Shore says a model is useful when it forms part of prioritization for substantial initiatives because it brings assumptions into the open, including those proponents often leave unstated. Advocates of different initiatives can critique one another’s models, provided the discussion does not become dysfunctional. Finance can help provide a clear assessment and demonstrate how criticism can improve a model without damaging working relationships.
After launch, the model can also help demonstrate product development’s value and accountability. Teams should continue watching proxy metrics, actual sales results and savings, then revise forecasts accordingly.
The method can also support an assessment of AI’s productivity impact. Measured speed gains can be used to model what the initiative would have cost without AI, while models of unintended consequences can help forecast maintenance costs. Shore emphasizes that the results remain based on estimates and assumptions.
He regards software priorities as a practical expression of company strategy: the software an organization builds or buys helps determine what it can do in the coming months and years. Financial models, however, do not replace judgment about product or organizational strategy; a company should not automatically select the project with the highest net present value.
Shore also warns that models can be manipulated if unfavorable metrics or sales evidence are not incorporated. He recommends using them to support decisions rather than assess individual performance. Because the approach is relatively demanding, it is unsuitable for small features and bug fixes and does not replace measures of speed or unintended consequences. Modeling value takes longer, but it can help organizations understand development’s overall business contribution and execute their strategy.



