Why most enterprise AI never makes it out of pilot and the three capabilities we built to fix it
The AI operating model problem —
and how OnStak is solving it.
The launch of the OnStak AI Portfolio reframes the AI conversation from model performance to operating-model readiness — and ships three capabilities to close the gap.
Just over 80% of enterprise AI projects fail to deliver intended business value — twice the failure rate of regular IT projects. That's the headline finding from RAND Corporation's 2025 analysis of more than 2,400 enterprise AI initiatives — and RAND notes that failure rates have barely moved in three years, after the industry has collectively spent $30–40 billion learning the lesson the hard way.
The diagnosis is unsparing, and at OnStak we think it's also correct: the gap isn't between organizations and the models. The gap is between pilots that work in a sandbox and the operating model an enterprise has to build around AI for production to absorb the work.
That diagnosis is what the OnStak AI Portfolio, launching today, is built to help close.
of enterprise AI projects fail to deliver intended business value — twice the failure rate of regular IT projects. The failure rate has barely moved in three years.
Most AI conversations still treat the model as the bottleneck. Spend more on the model, get more out of it. The data tells a different story. Pilots routinely demonstrate value; production routinely fails to capture it. The difference is rarely the model — it's whether the rest of the operating model (data, workflows, controls, evidence, audit) is ready to receive it.
Most enterprises don't have an AI model problem. They have an AI operating model problem. Pilots install AI well enough. Production is where the work is — and where most enterprises stall. The OnStak AI Portfolio underpins that transition.
Fabio Gori — Chief Product and Marketing Officer, OnStakRead Fabio's full perspective on the AI operating model shift →
The OnStak AI Portfolio delivers three capabilities, each built around the same underlying principle: get correlation right, and the operating-model gap closes.
Correlates across every layer of the stack and passes only what's service-relevant to the model. Typical token reduction of 15–20× per decision for AIOps use cases. AIOps is the first live use case.
In production today across healthcare, retail, and manufacturing. Packages spanning Safety & Security, Behavioral analytics, and custom analytics. No rip-and-replace.
Produces audit-grade evidence as a runtime byproduct — not a quarterly retrofit. In active POC with a healthcare design partner today.
AI Correlation Fabric — Live correlation graph tracing a critical path across 5 stack layers · Incident #4218 · 19 entities
The first customer of the AI Correlation Fabric is OnStak itself. Inside our Application Modernization practice, a 25-app estate is now moved in 5–6 months instead of nine, at 30–40% less effort than the interview-led programs that have defined this category for the last decade. Because the Fabric is left in place after go-live, AI is already running on the modernized estate the moment it's switched on. No throwaway tooling. No second migration to "AI-ready" the environment after the fact.
This matters for buyers asking the obvious question: do you actually use what you sell?
Delivery time compressed for a 25-app estate using AI Correlation Fabric
Reduction vs interview-led modernization programs — with AI live on day one
Per decision in AIOps use cases — fewer tokens, faster performance, less hallucination
We didn't design Correlation Fabric in a lab. We built it because our own modernization teams needed it to ship real work without re-platforming the environment twice. If it can carry our own delivery practice, it can carry yours.
Ben Haddox — Field CTO, OnStakVideo AI Analytics is in production today across three industries — all built entirely on existing camera infrastructure, with no rip-and-replace and no parallel system to maintain.
Reduction in severe fall injuries
US healthcare network · 250+ existing cameras
Operational cost reduction · 30% efficiency improvement
US retail chain · 500+ branches
Improvement in security & safety compliance
US manufacturer · 24-camera estate
The OnStak AI Portfolio will be running live at Cisco Live, Las Vegas, May 31 – June 4. AI Correlation Fabric end-to-end — both in IT operations and inside regulated AI workflows. The work happening in real time, not in a slide.
Can't make it to Cisco Live? Register for the AI Portfolio Overview webinar and see the Correlation Fabric in action from wherever you are.
Register for the webinar →OnStak builds the technology that moves AI from pilot to production — backed by Application & Data Modernization. The California-based company came into existence in 2013, founded by Muhammad Haq, Chief Executive Officer and Co-Founder, to cater to true digital disruption. For more than a decade, OnStak has been migrating the unmovable, modernizing the untouchable, and fixing the applications and data everyone gave up on.
75% of OnStak's workforce are AI architects and certified engineers. 30+ enterprise AI deployments and enterprise AI strategy consulting delivered across Healthcare, Finance, Public Sector, Manufacturing, and Retail. OnStak and its subsidiary Digitalstates partner with Cisco, NVIDIA, AWS, and Splunk. Cisco MINT partner. Headquartered in Milpitas, California, with operations across the United States, Canada, and Europe.
More information at www.onstak.com
Source: RAND Corporation · 2025 analysis of 2,400+ enterprise AI initiatives.
Briefings, interviews, or a Cisco Live meeting: Jessie Pengilly · jessie.pengilly@onstak.com · +1 855 811 2345
Ready to move from
pilot to production?
See every capability, every outcome, every integration — or talk to the team that has 30+ production deployments behind them.