AI tools, composable architecture, and the low-code revolution have rewritten the economics of custom software — but the graveyard of failed projects has never been larger. The build-vs.-buy decision in 2026 is no longer a binary choice. Companies like Volkswagen have torched over $7.5 billion trying to build what they should have bought, while firms like Duolingo are using AI to produce 4–5x more output with the same headcount. The real winners are those embracing a hybrid "buy the platform, build the differentiator" strategy — and the data shows they're achieving 40–75% faster time-to-market. For business leaders, the stakes have never been higher: Forrester estimates 67% of software projects fail because of wrong build-vs.-buy choices, and failed digital transformations now cost organizations $2.3 trillion per year globally.
The $7.5 billion lesson: when "build" goes catastrophically wrong
The most expensive build failure in recent memory belongs to Volkswagen's CARIAD software division. Launched in 2020 to create a unified operating system across all 12 VW Group brands — from Porsche to Skoda — CARIAD accumulated over $7.5 billion in operating losses by 2024. The division ballooned to 6,000 employees without clear structure or authority, as staff from VW, Audi, and Porsche each built parallel systems. The result was architectural deadlock: 17 status meetings per week, delayed launches of the Porsche Macan Electric and Audi Q6 E-Tron by 18+ months, and ultimately a $5.8 billion investment in Rivian just to license working EV software. VW tried to compress a 10-year development journey (Tesla built its stack over 15 years) into five. The total estimated damage exceeds €20 billion.
CARIAD isn't an outlier. The Australian Securities Exchange spent seven years and AUD $250 million attempting to replace its clearing system with blockchain technology before scrapping the project entirely — ASIC later sued ASX for misleading the market about progress. The Pentagon's DCHRMS project, begun in 2018 with a $36 million budget and one-year timeline, was abandoned in 2025 after consuming $280+ million across eight years — a 780% cost overrun. And the UK's NHS National Programme for IT, history's largest civilian IT program, burned through £10–12.7 billion before cancellation, delivering barely a quarter of that in actual benefits.
The pattern is consistent. Technology is rarely the root cause of build failures. VW's problem wasn't code — it was organizational chaos. The NHS program was imposed top-down without consulting doctors. The Pentagon lacked oversight. Hertz paid Accenture $32 million for a website redesign that never launched because the assigned team lacked expertise in the contracted technology. These are leadership failures, not engineering ones.
When "buy" backfires: the SAP and Oracle graveyard
Buying off-the-shelf software carries its own catastrophic risks, and enterprise resource planning implementations dominate the failure list. Lidl, the €80 billion grocery giant, spent €500 million over seven years on an SAP implementation before abandoning it entirely. The fundamental problem was deceptively simple: Lidl managed inventory by purchase prices, SAP uses retail prices. Rather than adapting their processes, Lidl demanded massive customization — getting the worst of both worlds.
Target Canada represents perhaps the most devastating buy-side failure. When Target expanded north, it chose to adopt SAP fresh rather than extend its proven U.S. custom systems. Nobody on the team had strong SAP expertise. Data entered into the system was riddled with errors — wrong dimensions, currencies, missing fields. The SAP, JDA forecasting, and Manhattan warehouse systems couldn't communicate properly. Warehouses overflowed while store shelves sat empty. Target Canada reported $2.1 billion CAD in losses before shuttering all 133 stores and laying off 17,600 employees. Total costs reached an estimated $5–7 billion.
The list goes on: Revlon lost $70.3 million in a single quarter when a botched SAP S/4HANA migration prevented its North Carolina facility from fulfilling orders, triggering shareholder lawsuits. National Grid spent $585 million remediating an SAP implementation that went live during Hurricane Sandy, hiring 850 contractors at $30 million per month. LeasePlan spent €100 million on an SAP system that never went live. Birmingham City Council declared effective bankruptcy after an Oracle ERP implementation left it unable to manage finances.
The cross-cutting lesson from both build and buy failures is that the decision itself matters less than the execution. Bad data, compressed timelines, unfamiliar technology, missing user involvement, and absent executive alignment destroy projects regardless of whether you're building or buying.
AI is rewriting the cost equation — but not the way vendors claim
The most disruptive force in the build-vs.-buy calculus is AI-assisted development. GitHub Copilot now has over 20 million users, with 90% of Fortune 100 companies adopting it. A controlled study of 4,800 developers by GitHub and Accenture found 55% faster task completion and 75% reduction in pull request cycle times (from 9.6 days to 2.4 days). Across the industry, 41% of all code written in 2025 is AI-generated, and at Y Combinator's Winter 2025 batch, 21% of startups reported codebases that were 91% AI-generated.
The named-company examples are striking. Duolingo's CEO Luis von Ahn reported that "with the same number of people, we can make four or five times as much content." The company's first 100 language courses took 12 years; the next 148 took roughly one year using generative AI. Shopify's CEO Tobi Lutke issued a company-wide memo requiring teams to "demonstrate why they cannot get what they want done using AI" before requesting additional headcount. Shopify's headcount dropped from 8,300 to 8,100 while revenue grew 20–40% annually. Freshworks engineering teams reduced coding time by 30% and improved code quality by 61%, with one team compressing a 70-hour project to just 8 hours.
However, the productivity narrative has important caveats. A rigorous randomized controlled trial by METR with 16 experienced open-source developers found they actually took 19% longer to complete tasks with AI tools — even though they believed AI sped them up by 20%. Developer sentiment toward AI tools has dropped to 60% positive (from over 70% in 2023–2024), and only 29–46% of developers trust AI outputs. Code churn is increasing as teams discover that AI-generated code can be harder to maintain. One company built an AI-assisted internal analytics dashboard in two weeks; six months later, nobody could modify it, and they scrapped it for a SaaS product.
The pragmatic takeaway: AI dramatically lowers the cost of the first 90% of a build project — the tedious but straightforward work — but barely touches the remaining 10% that consumes 90% of the time. For prototyping, internal tools, and MVPs, AI has genuinely shifted economics toward "build." For complex, mission-critical systems requiring long-term maintenance, the advantage is more nuanced.
The hybrid model is winning: composable architecture in practice
The most successful companies aren't choosing build or buy — they're doing both through composable, API-first architectures. The MACH Alliance's 2025 survey of enterprises with 5,000+ employees found 87% have widely implemented microservices, API-first, cloud-native, and headless (MACH) technologies. Nine in ten report the approach met or exceeded ROI expectations. IDC data shows composable ecosystems lower total cost of ownership by as much as 37% and reduce innovation lead time by up to 50%.
The concrete results are compelling. L.L.Bean migrated its $1 billion+ digital business to Commercetools' composable commerce platform, onboarding 250,000 SKUs, migrating 5.5 million customer accounts, and deploying 100+ enhancements — generating $10 million in incremental revenue with 100% uptime during peak events. Pet Valu achieved 40% faster time to market and 70% faster homepage loads. Ulta Beauty launched buy-online-pick-up-in-store capability in just 7 days across 1.3 million SKUs. BMW Group shipped 100 releases in 8 months with 99.99% availability.
In e-commerce specifically, the hybrid migration wave from monolithic platforms to Shopify's headless Hydrogen framework is producing consistent results. NYDJ cut total cost of ownership by 65% after migrating from Salesforce Commerce Cloud, launching Afterpay integration in a single day versus an estimated three-month project on the old platform. Bauer Hockey saw an 18% conversion increase and 60% revenue growth. Digital agency CQL, which migrated 17 brands from Salesforce Commerce Cloud to Shopify, documented $800K to $11 million in total cost of ownership savings per brand, with 50–75% faster implementation timelines.
Gartner predicts that by 2027, more than 60% of new cloud-based commerce solutions will use MACH principles, and organizations using composable architecture will outpace competitors by 80% in feature implementation speed. The strategic logic is clear: buy the infrastructure where it's commoditized (payments via Stripe, communications via Twilio, commerce via Shopify or Commercetools, content via Contentful), and build only where your business creates unique competitive advantage.
Low-code platforms have crossed the enterprise threshold
Low-code and no-code platforms are no longer just for prototypes and departmental tools. The market will exceed $30 billion in 2026 according to Gartner, and 97% of Fortune 500 companies now use Microsoft Power Platform. Gartner's 2025 Magic Quadrant names OutSystems (Leader for the 9th consecutive year), Microsoft, Mendix, Appian, ServiceNow, and Salesforce as the enterprise leaders. The prediction that 75% of all new applications will use low-code by 2026 appears to be tracking toward reality: already 70% of new enterprise applications leverage these technologies, up from less than 25% in 2020.
Real enterprise deployments demonstrate the maturity shift. Coca-Cola Beverages Vietnam digitized 60+ workflows in six months using Power Apps and Power Automate, from purchase orders to warehouse management. Aviva unified 22 different systems into a single interface for call center operations using Appian. Northern Trust replaced an end-of-life case management system serving 10,000 employees across 8,000 case types in 50 functional areas. Deutsche Bahn licensed every employee for Power Platform, enabling rapid app creation across the entire organization.
The citizen developer movement is accelerating this trend. Gartner projects citizen developers will outnumber professional developers 4:1 at large enterprises by 2026, with 80% of low-code users sitting outside formal IT departments. Nearly 60% of custom apps are now built outside IT. The most common use cases — form building (58%), workflow automation (49%), and data visualization (33%) — reflect the operational reality that business teams know their needs better than IT queues can serve them.
The risks are real but manageable. 47% of organizations cite scalability concerns, 37% worry about vendor lock-in, and governance remains a challenge. One security CEO predicts the typical enterprise will run 4,500–6,000 AI-generated apps and automations in 2026, with 66% remaining undiscovered by security teams. The emerging best practice is governed citizen development through Centers of Excellence that provide guardrails without killing agility. The warning from TXP's Andy Beardshaw is worth noting: "The growth of low-code and citizen developers will give rise to the next legacy crisis" if organizations don't maintain what citizen developers build.
A decision framework grounded in evidence
The 2024–2026 data points toward a clear framework. Build when the software is your competitive differentiator — Netflix's recommendation engine drives 80% of viewership; Uber's real-time matching is the product. Buy when the function is commoditized — the ERP failures at Lidl, Target Canada, and Revlon show that attempting to reinvent solved problems is expensive and dangerous. Go hybrid when you need both reliability and differentiation — the L.L.Bean, NYDJ, and BMW examples demonstrate that buying infrastructure and building experience layers delivers the best risk-adjusted returns.
Three new factors make 2026 different from any prior year. First, AI tools have reduced prototyping and MVP costs by 50–80%, making "build a quick version first to understand your needs, then decide" a viable strategy for the first time. Second, composable architecture means "buy" no longer means "accept whatever the vendor ships" — API-first platforms let you assemble best-of-breed components and customize the edges. Third, low-code platforms have matured to enterprise grade, creating a middle path where business teams can build 60–70% of what they need without professional developers.
The companies paying the highest price in 2026 are those treating build-vs.-buy as a one-time, binary decision rather than a continuous portfolio strategy. The evidence shows that organizations using formal decision frameworks achieve 40% better project outcomes. The question is no longer"should we build or buy?" It's "for each capability in our stack, what's the right mix of build, buy, and compose — and how do we govern that mix as it evolves?"
