BENGALURU — Despite widespread enthusiasm for artificial intelligence, a majority of enterprise AI projects fail to move beyond pilot stages due to fragmented data systems, weak governance and shortage of talent, according to a joint study by HFS Research and Mindsprint released on Friday.
The 2025 Market Impact Report, titled “Crack the AI Scaling Wall and Redefine Business Success via Services-as-Software,” surveyed senior technology and business leaders across healthcare, retail, manufacturing and financial services in North America and Asia. The study found that only 10% to 15% of AI initiatives have achieved large-scale deployment, while the rest remain stalled in early experimentation phases.
Data and Cultural Barriers Impeding Scale
The report said most enterprises hit what it termed a “scaling wall” made up of four obstacles — technical debt, process drag, talent deficits and trust gaps. Many organisations, it added, continue to depend on legacy data systems not built for AI, resulting in fragmented data pools that make integration complex and costly.
“AI is no longer the innovation lab’s plaything. It’s the enterprise’s new nervous system,” said Ashish Chaturvedi, Executive Research Leader at HFS Research and author of the report, in a statement included in the publication.
The study said lack of clear ownership for AI initiatives across departments further hampers scaling. Where projects are driven only by IT teams or innovation units, they often lose momentum after initial trials. By contrast, companies with dedicated AI councils or C-suite-level oversight show greater success in moving pilots into production.
Governance Seen as the Key Accelerator
The report identifies governance as the single most critical enabler of enterprise AI maturity. It recommends that organisations establish formal readiness gates — such as “proof of concept,” “pilot ready” and “scale ready” — each requiring documented evidence of performance, explainability and monitoring before deployment.
HFS notes that many firms are now forming internal AI governance councils that include product, data and risk leaders. These groups define policies around responsible AI use, including bias detection, data privacy and the need for human oversight in automated decisions.
A CIO from the insurance sector quoted in the report said “trust in the system is the hardest challenge,” adding that compliance and risk teams were being brought in earlier in AI lifecycles to prevent misuse.
“Human + AI” Model Emerges as Standard
Rather than aiming for full automation, most organisations are adopting a “Human + AI” framework — pairing algorithmic systems with human oversight. This hybrid model, sometimes described as human-in-the-loop (HITL), was endorsed by almost all business leaders interviewed for the study.
The report said this approach has already improved efficiency in sectors such as healthcare, logistics and financial services. It cited one healthcare executive who reported cutting decision time from 30 minutes to five seconds after deploying AI-driven data query systems.
While automation remains a long-term goal for many companies, the report said the near-term priority is achieving explainability and control over AI-assisted workflows.
“Services-as-Software” Offers New Route to AI ROI
One of the report’s central findings is the emergence of “Services-as-Software” (SaS) — a model that blends the scalability of software with the adaptability of services. Under SaS, enterprises move away from time-and-materials service contracts toward outcome-based engagements, where payments are tied to measurable business results such as decision speed or forecast accuracy.
The report outlines a “SaS maturity ladder” with four rungs:
-
Tooling – standalone AI utilities;
-
Playbooks – templated workflows;
-
Platforms – integrated multi-tenant services; and
-
Outcomes – shared-risk contracts tied to business metrics.
This evolution, according to HFS, is shifting how enterprises evaluate technology providers. Buyers are prioritising partners who embed AI into delivery processes and agree to share accountability for business outcomes. One supply chain leader quoted in the report said, “Vendors that don’t embed AI risk being abandoned by enterprises.”
Hybrid Deployment Dominates
The study found that 75% of enterprises now use hybrid AI deployment models — keeping strategic data and intellectual property in-house while relying on external providers for speed and scaling. About 15% of firms prefer provider-led AI, typically for specialised projects, while 10% continue to build entirely in-house systems.
The hybrid approach reflects a shift toward shared innovation, where technology vendors act less as labour suppliers and more as co-innovation partners.
The report noted that firms are moving away from measuring AI purely in terms of cost savings. Instead, they are tracking operational and customer experience outcomes such as decision latency, forecast accuracy, cycle time and customer satisfaction.
In practical terms, this means companies are using AI to enhance decision-making speed and agility rather than simply automate repetitive tasks.
Institutionalising AI
To help organisations overcome the scaling wall, HFS proposed an 18-month roadmap that begins with setting up AI councils and readiness criteria in the first 100 days, followed by evaluating pilots and codifying best practices by month 12. By month 18, the report recommends linking AI agility metrics to executive incentives and institutionalising AI reviews in monthly business meetings.
It also suggests using education and internal “demo days” to normalise collaboration between humans and AI across the workforce.
According to the study, the United States remains the fastest-moving geography for enterprise AI adoption, while Asia — particularly India — is emerging as a strong hub for AI delivery talent.
Sectors such as healthcare, retail and insurance are leading in pilot volumes, but manufacturing and logistics show higher success rates in scaling due to greater data standardisation.
