Shelf Life | Vol. 41 – Some Assembly Required: Measuring ROI for AI in the SDLC
📅 March 2026 | ✍️ Shelf Life
The pieces are multiplying. The real challenge is deciding which ones belong in the build.
TOP SHELF INSIGHTS
• 🧱 AI coding tools are in the enterprise toolkit. Teams now generate 30 to 50 percent of new code with AI assistance. The build got faster. The strategy hasn't caught up.
• 🛒 In retail and consumer goods, the most powerful revenue drivers (pricing, promotions, loyalty, checkout) live inside software. AI compresses the time from idea to deployed capability. That's a commercial advantage, not just an engineering one.
• 📦 When building gets faster, deciding what to build becomes the harder problem. The backlog is quietly becoming the most strategically important document in your technology organization.
• ⚠️ Dumping all the pieces on the floor and building fast without a plan produces something that looks impressive until someone steps on a piece. Governance needs to scale with speed (and Band-Aids).
• 📊 Lines of code is the wrong metric. Conversion lift, experimentation velocity, and time to market are the right ones. Measure what the software does to revenue, not how much of it you shipped.
The New Set: Your Revenue Lives in the Box
Every retailer has a version of this story. A recommendation engine that nudged basket size up two percent. A checkout redesign that cut abandonment. A dynamic pricing model that stopped leaving margin on the table. A loyalty mechanic that finally made the right offer at the right moment.
None of those things live in a store. They live in code. And for years, building them required months of lead time, significant engineering capacity, and the patience of a senior leader willing to defend the roadmap through three budget cycles.
AI is collapsing that timeline. Development teams across industries now report that AI tools generate 30 to 50 percent of new code. Certain tasks complete up to 55 percent faster. Prototypes that once took weeks materialize in hours. The pieces are all in the box. They’re cheaper than they’ve ever been. And there are a lot more of them than you’ve ever had access to before.
For CIOs, that’s a productivity story. For everyone else in the C-suite, it should be a much more interesting one: if software features drive revenue, and AI dramatically accelerates how quickly those features can be built, software delivery velocity just became a growth strategy. The question is whether the organization is structured to treat it like one.
Speed to market for digital capabilities is starting to look a lot like shelf availability for physical products. Miss the window and the demand goes somewhere else.
Open the Instructions: What AI in the SDLC Actually Does
The question most leadership teams are sitting with right now is not whether AI belongs in their software development process. It is what that actually means on a Monday morning.
The short answer: AI is being embedded across the entire development lifecycle, not just in the "writing code" part. The longer answer is that different entry points suit different organizations, and the ones getting the most traction started narrow and specific rather than broad and ambitious.
Here is where teams are seeing real, measurable impact today:
Code generation and acceleration. Developers use AI to generate boilerplate code, draft functions, and complete repetitive logic that previously consumed significant time. This is the most visible use case and the one most organizations start with. The productivity gains are real but the strategic value only materializes when that freed-up capacity gets redirected toward higher-value work.
Automated testing. AI can write and run test cases faster than any human team. For retailers with complex checkout flows, pricing logic, or loyalty mechanics, this means bugs get caught earlier and release cycles shorten. Less drama at launch. Fewer 2am calls.
Documentation. Nobody loves writing documentation. AI does it without complaint. Teams that have adopted AI-assisted documentation report that codebases become easier to maintain, onboard new developers into, and audit, which compounds value over time even when it feels invisible in the moment.
Code review and vulnerability detection. AI tools can scan code for security risks, inconsistencies with existing architecture, and patterns that have historically caused problems. This is where governance gets teeth and where organizations can actually scale review capacity alongside development speed rather than letting the gap widen.
Natural language to working prototype. This is the one that tends to stop non-technical executives mid-sentence. Business teams can now describe a capability in plain language and have a working prototype surface in hours. A merchandising team wants to test a new promotional logic. A loyalty team wants to model a new redemption mechanic. The distance between the commercial idea and something they can actually look at has collapsed.
Where to start — the practical version:
Most organizations that have successfully adopted AI in their development workflow began with one of two entry points:
The first is developer tooling — giving engineering teams access to AI coding assistants like GitHub Copilot or Cursor, establishing guardrails, and measuring the impact on delivery speed over 60 to 90 days. Low disruption, fast signal, relatively low governance overhead to start.
The second is a single high-value workflow — identifying one development process that is consistently slow, expensive, or error-prone (automated testing is a frequent choice) and deploying AI specifically there. The ROI is easier to isolate, the risk is contained, and the organizational learning is transferable.
What does not work well as a starting point: buying enterprise licenses across the board, deploying without policy, and hoping velocity metrics improve. That produces the bucket-dump problem before the organization has any practice cleaning it up.
Dumping the Bucket: The Backlog Is Now Everyone’s Problem
Here is the part of the AI-in-engineering story that does not make it into the press release.
When building gets dramatically faster, the number of things competing to get built multiplies just as fast. Feature requests arrive from merchandising, marketing, supply chain, loyalty, and digital teams simultaneously. The perceived cost of adding something to the backlog drops. So everyone adds to the backlog. And suddenly the constraint has shifted from “can we build it” to “do we have any idea what we should build next.”
This is the Lego bucket problem. Dump all the pieces on the floor, hand a child unlimited time to build, and you get something spectacular and structurally questionable. What you needed was someone to read the instructions first.
The organizations that navigate this well will develop sharper prioritization frameworks, tighter connections between engineering roadmaps and revenue targets, and explicit criteria for what earns a place in the queue. The ones that don’t will build more things that matter less, faster than ever before.
"Gartner research consistently finds that organizations with faster software delivery cycles outperform peers on digital revenue metrics over a three-year horizon. The advantage compounds — but only when delivery speed is paired with strategic prioritization discipline."
In practical terms, this means the conversation about AI in software development needs to move out of engineering forums and into commercial leadership. CEOs, CMOs, and Chief Digital Officers need a point of view on what gets built next. Because the decisions happening in backlog grooming sessions now carry more commercial consequence than they did two years ago.
Step 47 Is Missing: Speed Without Governance Is a Liability
Anyone who has built a Lego set knows the specific dread of reaching step 47 and realizing something went wrong around step 12. You can keep building. The thing will still look roughly right from a distance. But it will not hold together the way it should, and fixing it later costs more than pausing earlier would have.
AI-generated code has a version of this problem. The volume of code being produced in many organizations has already outpaced the rate at which it can be thoroughly reviewed. AI writes fast. Human review does not automatically scale with it. The gap is where risk accumulates — quietly, then suddenly.
Technical debt: Rapid code generation produces fragile systems when architectural discipline slips. What ships fast can cost significantly more to maintain, extend, or replace.Security vulnerabilities: AI-generated code can include patterns that introduce exploitable weaknesses. In retail environments, where customer data, payment flows, and loyalty databases are at stake, this is not a theoretical concern.Architectural fragmentation: Without oversight, teams build solutions that connect poorly to existing platforms. The result is a patchwork that slows future development more than it accelerated the current sprint.Governance gaps: Most organizations adopted AI coding tools faster than they updated development policies or accountability structures. The tools moved. The operating model did not.
The fix is not complicated, but it requires intention: human review checkpoints for critical code paths, automated testing pipelines, and explicit ownership of AI-assisted development decisions. Governance is not the thing that slows AI down. It is the thing that makes the speed sustainable. Looking at AI in the SDLC? Make sure someone like Gartner is in the room helping you think through how to incorporate the right process from the get-go.
Where Do the Extra Pieces Go?: You’re Measuring the Wrong Thing
Every Lego set comes with extra pieces. Small ones. The kind that end up in the carpet and announce themselves at 2am. They were never the point of the build. But if you counted them as progress, you’d think you were ahead.
Most organizations evaluating AI in their development workflow are doing a version of this. Lines of code generated. Developer output per sprint. Time saved on documentation. These metrics are not wrong. They are just incomplete. They tell you how the engine is running. They say nothing about where the car went.
The commercial case for AI in the software development lifecycle is built on a different set of questions: Did the new pricing capability improve margin? Did the checkout redesign reduce abandonment (and by how much? How many experiments ran this quarter versus last, and what did the team actually learn? How much faster did a revenue-driving feature reach customers compared to twelve months ago?
Revenue impact of new features: Did the capability improve conversion, basket size, or retention?Experimentation velocity: How many hypotheses did the team test this quarter? How many produced actionable results?Speed to market: How quickly can the organization move from commercial idea to deployed capability?Cost per feature delivered: Are development costs declining as AI tools scale across more of the workflow?
Shifting from engineering throughput to commercial impact is an organizational change worth considering. It requires product, engineering, and commercial leadership to agree on what “good” looks like. That alignment is harder than the technology. It is also more valuable.
On the House
AI in SDLC doesn't sound like a sexy topic, but the implications are certainly exciting. The organizations that learn to convert that speed into customer value — with the governance, the prioritization discipline, and the commercial measurement to back it up — will pull ahead in ways that are genuinely hard to replicate.
The ones that get the technology without the operating model will generate a lot of code. They’ll have very impressive sprint velocity metrics. And they’ll wonder, at some point, why the revenue didn’t follow.
The ones that don't prioritize the right use cases will achieve speed, but lose potential return.
The instructions were in the box. Someone just needed to read them first.
The Last Look
When AI makes building dramatically faster, the real competitive advantage moves to the organization that knows what to build next.
Is your prioritization framework ready for that conversation, or is it still buried in a backlog nobody owns?
More to come in the Shelf Life series.
Follow me here for sharp takes on the trends shaping retail, fashion, and consumer product companies.
Want to talk more about how Gartner Consulting can help your organization? Follow me on LinkedIn or @ShelfLifebyJKS on Instagram or reach out!
📍 Jackie Swanson is a Managing Partner at Gartner Consulting, specializing in retail, consumer products, and utilities. She advises companies on large-scale transformations spanning strategy, operations, and technology. Jackie lives in New York with her husband and their three children.
#ShelfLife #AISoftwareDelivery