Crack the AI scaling wall and redefine business success via Services-as-Software

This Market Impact Report is for CIOs, CTOs, heads of AI, and senior operations leaders in retail, CPG, healthcare, insurance, and industrials evaluating how to move AI from pilot purgatory to enterprise scale and adopt Services-as-Software™ as a standard operating model.

Introduction

This report outlines how enterprise leaders in retail, CPG, healthcare, insurance, and industrials are deploying AI to compress decision time, improve customer experience (CX), and industrialize operations, and why most efforts stall at scale. We test what ‘good’ looks like for governance, measurement, and sourcing in the emerging Services-as-Software (SaS) era. The findings translate directly into buyer actions, KPIs, governance gates, and contract levers.

Insights were drawn from several interviews with CIOs, CTOs, heads of AI, and senior business leaders in healthcare, retail, CPG, insurance, and industrials across North America and Asia (anonymized where requested).

Executive summary

Three years ago, it took half an hour. Now the same person can do it in five seconds. That’s not a fad. That’s transformation.

— Head of AI at a US-based healthcare company

AI can be a buzzword. It’s not a magic wand. Sometimes it kills one issue and creates another.

— CIO of an Asia-based insurance company

The AI scaling wall: Debt, drag, deficit, and doubt

For all the enthusiasm around AI, a sobering fact remains: most enterprise AI initiatives do not fully scale or deliver sustained impact. Multiple responses indicate that only 10–15% of AI proof-of-concepts make it to production. Our interviews confirmed this ‘scaling wall’ of debt, drag, deficit, and doubt.

Every executive had stories of promising pilots that fizzled out or stalled, pointing to a similar pattern: no shortage of AI POCs but a lack of fully deployed AI solutions enterprise-wide.

So why is scaling AI hard? The leaders we spoke with cited four interlocking challenges: technological, organizational, cultural, and trust (see Exhibit 1), which must be overcome to break out of pilot purgatory.

1. Data and technology debt: Legacy stacks, mainframes, fragmented ERP/CRM, and ‘spaghetti’ data complicate integration and slow deployment. Even when a model works, plugging it into live systems is arduous.
2. Process and organizational drag: Stage-gates and annual budgeting designed for pre-AI project cycles stall promising pilots. Decision rights across CIO, digital, and business owners are unclear. Bolting AI onto old workflows only yields marginal impact. Organizations must address process debt explicitly.
3. Talent and culture: Resistance arises from fear of replacement and lack of ‘trilingual’ talent (domain, data, product). Teams need show-and-tell demos and internal academies to build trust and pull.
4. Trust and governance gaps: Leaders want to know when an agent is ready for production and how to version and monitor it. Some firms simultaneously ban consumer AI tools while driving enterprise AI, creating policy contradictions.

Legacy tools weren’t built with AI in mind. Data is ‘spaghetti’ with no consolidation.

— Retail and insurance leaders

Exhibit 1: The four intertwined obstacles to AI scaling

Source: Interviews with several enterprise leaders across the US and Asia

Breaking through this scaling wall requires an all-fronts assault. Companies must modernize their data and IT environment (cloud migration, data lakes, APIs, modular architectures) so AI pilots aren’t stuck in integration hell. They should revamp processes to be more agile—adopting DevOps/MLOps for model deployment, implementing agile project methods, and changing approval workflows to empower faster experimentation. Culturally, they must invest in upskilling and change management. This involves positioning AI as a tool to augment employees, not replace them, and training staff to confidently use AI-driven insights in their day-to-day decisions. A retail executive said they host ‘AI demo days’ where teams showcase what they built and learned. Such sessions demystify AI and drum up cross-company support while building grassroots support to scale successful projects to other units.

Crucially, the tone from the top matters. Several executives noted that without C-level support, AI initiatives languish in innovation labs. One described how their CEO and business unit heads actively ask “How are we using AI on this?”—a clear signal that scaling AI is a priority.

A retail product executive said their CIO always insists on an ‘AI plan’ for any new budget request. This kind of mandate forces teams to bake scalability into project design (they know it’s headed for real deployment, not just a sandbox). It also helps combat the ‘random acts of AI’ syndrome by aligning AI efforts with business strategy and ensuring resource commitment beyond the prototype phase.

Where is AI delivering now? Bets beyond cost positively impact decision latency and CX

Enterprise leaders no longer ask “why AI?” but “where and how can AI deliver the most agility?” Across our interviews, executives described a range of AI initiatives aimed squarely at making their businesses more responsive and impressive. These initiatives span decision support, customer-facing innovations, and operational efficiencies.

Here are some of the prominent bets enterprises are placing:

Decision support in healthcare: A healthcare AI head described complex, cross‑modal queries dropping from 30 minutes to five seconds, enabled by governed pipelines and a blend of proprietary/open models and Gemini for Google Cloud—an example of measurable decision‑latency collapse.

These tools are really great. I love the human support I get with them.

— Head of Applied AI at a healthcare company

Retail personalization at scale: A global sportswear brand operationalized expert‑on‑demand and AI coaches to provide 24×7 product‑fit guidance by drawing on shopping history, validating the results through A/B testing on add‑to‑cart, conversion, and checkout rates.

We plan each season because we must respond to trends and consumer insights on Nike.com and our app.

— Digital business director of a sportswear brand

Supply chain and frontline optimization in beverages: A global beverages leader uses store‑specific selling stories that factor demographics and weather while running 17,000 daily routes with AI‑assisted routing (Ortec). Data governance is managed using a system integrator, while vendors provide planning, manufacturing, and CRM stacks.

We like to use the word “smarter” as in “how do we make every process smarter?”—suggestions our people can act on.

— Supply chain transformation lead, CPG

Meal kit operations: Machine learning and synthetic data help improve forecast accuracy. Barcoding and scan‑sort automation reduce manual touches. Pilot agentic AI augments customer support.

Insurance operations: Straight‑through underwriting and KYC checks combine machine learning, computer vision, and HITL for compliance. Conversational bots support customers.

How is success getting measured?

Technology programs (especially automation and outsourcing) have traditionally been justified with cost savings and efficiency gains. The classic KPI might be an ‘X% reduction in processing time’ or a ‘Y headcount reduction leading to $Z saved.’ While cost and efficiency remain important, our research found a notable shift: leading organizations are redefining success in terms of agility, resilience, and CX rather than cost alone. KPIs are expanding beyond savings to include decision latency, forecast accuracy, time to detect/resolve, conversion, CSAT/NPS, and error rates.

Exhibit 2: AI bets are shaping enterprises’ overall KPIs and value proposition across sectors

Source: Interviews with several enterprise leaders across the US and Asia

One might wonder if there are trade-offs between cost and these new metrics. In some cases, yes—improving resilience or CX means investing more in redundancy or service, which can increase cost. However, the interviewed leaders mostly believe that AI lets them transcend the old trade-off. For example, it can simultaneously improve CX and reduce cost by automating routine tasks (customers get faster service; company saves money on manual labor) or it can improve resilience and efficiency by optimizing inventory (less stockouts and inventory holding cost). These win-win outcomes are the ideal, and AI makes them more achievable than traditional methods. Therefore, focusing on the broader agility metrics often yields cost benefits, even if indirectly.

Moreover, as enterprises recalibrate their success metrics, they’re also reassessing their expectations for providers and partners. This mindset is driving the Services-as-Software evolution that we will discuss in the next section.

Don’t come to me selling 30% cost savings through outsourcing. Tell me how you’ll improve my NPS by 10 points or reduce my turnaround by 50%—that’s how you’ll get my attention now.

— CIO of a US-based specialty retailer

Pragmatic governance can infuse speed and trust

The one theme that nearly every executive hammered home was this: without proper governance, your AI ambitions will run amok or aground. Leaders want guardrails that accelerate scale. Readiness criteria (accuracy thresholds, robustness, explainability), versioning, always-on monitoring, and a rollback plan are emerging as standard gates.

How do you know when a tool, technique, or agent is ready for production? How do you manage and version it?

— Head of Applied AI at a healthcare company

To identify which AI projects are the most promising and ready to scale to the next level, enterprises are designing stage-gates, with dedicated teams responsible for moving them to the next stage.

Exhibit 3: Enterprises are leveraging similar governance stage gates in their AI pursuits

Source: Interviews with several enterprise leaders across the US and Asia

Here’s a breakdown of how enterprises are implementing governance and why it’s so critical to infuse agility.

1. Centralized oversight and strategy: Many companies have formed some type of AI council or steering committee to coordinate efforts. These typically include stakeholders from IT, data science, risk/compliance, and business units, often chaired by a C-level leader (Chief Data Officer, CIO, etc.). The mandate is to set AI strategy and policy, prioritize use cases, allocate resources, and monitor progress/risks.
2. Standards and policies (responsible AI): Another key function of governance is establishing policies for ethical and responsible AI use. Generative AI and machine learning bring concerns about bias, transparency, privacy, and security. Firms are drafting guidelines to address key questions: “What data is permissible to use for AI? How do we validate models for fairness and accuracy? When must a human be kept in the loop? How do we handle AI-generated content? For instance, a bank might restrict the use of customer personal data in AI models without anonymization or mandate that any AI decision (like for loan approval) affecting a customer must be explainable and subject to human override. A CIO from the insurance sector noted that compliance and risk teams are involved early in their AI projects. “Trust in the system is the hardest challenge,” as he put it.
3. Cost and investment governance: A perhaps less exciting yet important aspect of governance is controlling costs and avoiding duplicate spending. When every team experiments with AI independently, this often results in inefficiencies. Central governance can approve budgets in a way that encourages reusing platforms and focuses on investing smartly.
4. Talent and ethics oversight: Some governance bodies also take on the role of monitoring workforce impact and ethical implications. For example, if an AI will displace roles, the governance group might work with human resources on retraining programs or determine how to handle that transition responsibly. They might also put boundaries on use cases; for e.g., deciding that fully autonomous decision-making is off-limits in sensitive areas or that AI should augment rather than replace some interactions (like a luxury brand deciding an AI stylist will assist human stylists to preserve brand experience). Such principle-based decisions are increasingly codified in AI ethics charters that boards or C-suites ratify.

In summary, governance is the keystone because it holds together all other pieces. It ensures that the AI bets align with business strategy, attacks the scaling barriers by creating consistency and trust, and bakes in the redefined success metrics and ethical considerations.

Services-as-Software is the next frontier

Amid the tactical work of scaling AI and retooling metrics and governance, our conversations also touched on a more strategic, disruptive trend on the horizon: the blurring line between technology services and software products. We call this the Services-as-Software (SaS) paradigm, where services traditionally delivered by humans (often via outsourcing or consulting engagements) are increasingly being codified into software platforms, frequently powered by AI. In parallel, software vendors are infusing more services-like capabilities into their products, effectively encroaching on a territory once dominated by service providers. This two-way convergence, sometimes dubbed ‘reverse SaaS’ or ‘everything-as-a-service,’ is poised to redefine how enterprises procure and consume solutions and how providers deliver value.

Why is SaS the need of the hour? Because SaS promises agility on a fundamentally different scale. Instead of lengthy service engagements or labor-intensive contracts, enterprises could tap into on-demand, intelligent platforms that deliver outcomes via automation and minimal human oversight. It’s the difference between hiring a team of analysts to continuously monitor the supply chain versus subscribing to a supply chain control-tower software that automatically flags anomalies to managers. The latter is faster, more scalable, and outcome-focused.

Three core factors, i.e., technology maturity, need for speed and resilience, and workforce and customer expectations, are fueling the SaS agenda.

Exhibit 4: Buyers seek faster time to value and portability; providers are pushed to codify IP and stand behind outcomes

Source: Interviews with several enterprise leaders across the US and Asia

However, SaS is a continuum and HFS (in collaboration with the leaders) contemplated a SaS maturity ladder. This buyer-side framework shows how ‘services become software’ in four rungs—each rung changes what you buy, how value is proven, and how you contract. It helps you locate every vendor engagement today and plot the next step toward paying for outcomes with portability and guardrails.

Exhibit 5: From tools to outcomes, where does your current provider stand on the maturity ladder?

Source: Interviews with several enterprise leaders across the US and Asia

SaS is changing the core principles of enterprise-provider relationship

The role of providers is shifting from labor arbitrage to AI-first partners. They are required to codify IP, embed AI across delivery, and measure metrics by outcomes, not hours. Increasingly fatigued by AI‑washing, enterprises are prioritizing partners for their embedded intelligence and willingness to co‑own outcomes. Buyer expectations include:

• Co-innovation with outcome guarantees and shared-risk models (especially industry-specific)
• End-to-end AI-infused delivery (codegen/testing agents, process automation, AI-assisted engagement)
• Governance champions to build explainability and ROI metrics
• Ecosystem orchestration across hyperscalers, SaaS, niche AI, and on-prem estates
• Commercial shifts to consumption and outcome-based pricing; monetized IP accelerators

Vendors that didn’t embed AI risk being abandoned by enterprises.

— Supply chain leader at a CPG firm

AI engagement models with providers are following a hybrid approach

Enterprises, on the other hand, are opting for a hybrid build approach while pivoting toward an AI-centric future. They’re keeping their strategy, IP, and sensitive data in-house and expecting providers to accelerate the development and scaling of managed services.

Exhibit 6: AI adoption archetypes

Source: Interviews with several enterprise leaders across the US and Asia

Buyer playbook for Services-as-Software and 18‑month roadmap

HITL remains a prudent approach, consistently endorsed by leaders as the durable near-term model. Here’s a prescribed 18-month roadmap based on our ongoing enterprise work and brainstorming with the leaders interviewed for this report:

100 days

• Stand up or refresh the AI council. Publish readiness criteria (accuracy thresholds, explainability, monitoring, rollback), and an HITL policy.
• Re‑baseline three active pilots to the KPI scoreboard (latency, forecast accuracy, conversion).
• Launch two SaS pilots with outcomes, automation‑curve reporting, and portability clauses.

12 months

• Kill or scale based on gate evidence. Create a reuse library of patterns (prompts, data products, agents).
• Close policy contradictions on personal vs enterprise AI use. Standardize approved tools and access.
• Shift one function (e.g., routing optimization or review summarization) from experiment to a standard operating model.

18 months

• Institutionalize the scoreboard in monthly business reviews; tie incentives to AI agility metrics.
• Migrate repeatable work from T&M to platform/outcome models where performance is stable.
• Expand education with demo days and internal academies to normalize Human + AI work.

Provider pressure test

• Show baselines and target deltas that matter (latency, accuracy, conversion).
• Price on outcomes with shared savings and a visible automation curve.
• Prove portability (schema ownership, artifact escrow, exit SLAs).
• Commit to explainability and auditability (logs, prompts, traces).

The Bottom Line: Build a comprehensive governance structure and fund AI projects that improve decision speed, resilience, and CX. SaS must be your standard operating model.

The journey toward an AI-empowered, agile enterprise is challenging but underway in organizations across sectors. Their experiences form a playbook that others can follow, adapting to their context but guided by common principles.

While still nascent, the shift to outcome-based, AI-infused service models is on the horizon. Prepare by rethinking vendor strategies and exploring pilots with progressive partners. Enterprises that internalize this change early will shape it, negotiating better terms and forging the ecosystems that suit their needs. As buyers, cultivate flexibility in contracts. As providers, invest in IP and platforms.

Enterprise contributors to this report

• Aleksandar Lazarevic, Vice President, Analytics – one of the largest meal-kit providers in the world
• Amit Agrawal, Director of Product Management, AI and Search – American retail company specializing in home improvement
• Director of Strategy and Transformation, Supply Chain, PepsiCo
• Jonathan Ozeran, Vice President, Gen AI, a health technology company
• Danielle Weis, Director of Digital Business, Nike
• Ronny Tan, CIO, one of the largest insurance companies in Indonesia
• Group CTO, general insurance provider in Southeast Asia
• Axel Ramm, Engineering Director, American multinational manufacturer and marketer of home appliances
• Ann Loung, Head of Artificial Intelligence & Data Science, Vietnam’s leading consumer-focused business group
• Senior Executive, a senior executive from a specialty retail brand