Why Industrial AI Projects Stall Before Production
This article is sponsored by IFS and was written, edited, and printed in alignment with our Emerj sponsored content guidelines. Learn extra about our thought management and content material creation companies on our Emerj Media Services page.
A widening hole has emerged between industrial enterprises’ ambitions for AI and their means to operationalize it; this hole is manifesting in productiveness losses, stalled initiatives, and avoidable operational prices that immediately erode margins and competitiveness.
Research summarized by MIT Sloan, drawing on U.S. Census Bureau knowledge from tens of 1000’s of producing corporations, shows that organizations adopting AI expertise an preliminary 1.33‑proportion‑level decline in productiveness, pushed by misalignment between new AI methods and current processes and workflows.
Meanwhile, the operational vulnerabilities that AI is meant to handle stay pricey and protracted. Analyses from the National Institute of Standards and Technology (NIST) estimate that U.S. producers incur $18.1 billion in preventable downtime losses yearly as a part of a broader $119.1 billion in avoidable upkeep‑associated prices, pushed by reactive upkeep, defects, delays, and stock disruptions.
These are exactly the classes of operational instability that AI‑enabled predictive and prescriptive methods are supposed to cut back, but they continue to be largely unmitigated when AI can’t be successfully built-in into current workflows.
Emerj lately spoke with IFS executives to debate how asset-intensive industries can bridge the GenAI divide. Leaders featured embody Kriti Sharma, CEO of IFS Nexus Black, and Somya Kapoor, CEO of IFS Loops. During these conversations with Emerj Editorial Director Matthew DeMello and CEO Daniel Faggella, three key themes got here to the fore:
- Accelerating time-to-value by means of high-velocity deployment: Moving from multi-year implementations to three-week production-grade outcomes.
- Scaling digital workforces to protect institutional information: Preserving institutional information as senior technicians retire at document charges.
- Implementing governance by way of supervisor agent frameworks: Utilizing “Supervisor Agents” to observe autonomous workflows and mitigate hallucination dangers.
Accelerating Time-to-Value by means of High-Velocity Deployment
Episode: Solving Hard Industrial Problems with Fast AI Deployment – with Kriti Sharma of IFS Nexus Black
Guest: Kriti Sharma, CEO, IFS Nexus Black
Expertise: AI Engineering, Industrial Environments, Rapid Product Development
Brief Recognition: Kriti Sharma is the CEO of IFS Nexus Black. A pioneer in AI for ladies, she beforehand served as VP of Artificial Intelligence at Sage and based the AI for Good initiative to develop moral autonomous methods. Kriti holds a Master’s in Computer Science from University College London (UCL) and at the moment spearheads the strategic Anthropic partnership for IFS.
Kriti Sharma explains that many industrial groups have misplaced persistence with lengthy AI applications that by no means make it to manufacturing. Her strategy begins by putting engineers immediately within the operational atmosphere, the place the actual constraints dwell — last-minute orders, sudden absences, emergency repairs, and the fixed reshuffling that make manufacturing planning one of many hardest unsolved issues in trade.
Working on the ground permits groups to search out the information that really governs the method, SCADA alerts, P&ID diagrams, sensor streams, stress, temperature, vibration, and to construct instruments that replicate the true complexity of the system quite than a laboratory abstraction. This proximity additionally reveals the operational patterns that drive downtime and reactive work, enabling earlier intervention and elevated throughput.
Kriti describes the trail to autonomy as incremental quite than all‑or‑nothing:
“Production planning continues to be a tough drawback. You’ve obtained final‑minute orders, somebody goes off sick, and a restore throws the entire plan out of whack. You begin by fixing the actual drawback with extra intervention and managed deployments, and as you see success, you dial up the autonomy over time. Keep your head within the cloud and your toes within the mud, assume large, however take small steps ahead.”
— Kriti Sharma, CEO, IFS Next‑Black
Scaling Digital Workforces to Preserve Institutional Knowledge
Episode: How Digital Workers Are Changing Industrial Performance – with Somya Kapoor of IFS Loops
Guest: Somya Kapoor, CEO, IFS Loops
Expertise: Agentic AI, Supply Chain Optimization, Business Process Management
Brief Recognition: Somya Kapoor is the CEO of IFS Loops and spent over 15 years in “engine room” management roles at international software program giants SAP and ServiceNow, specializing in scaling options that bridge the hole between company software program and frontline actuality. Somya holds an MBA from Santa Clara University and a level in Computer Science Engineering and is a acknowledged authority on industrial AI adoption and the administration of hybrid human-machine workforces.
Somya Kapoor highlights a shift reshaping industrial AI: organizations are shifting from inflexible, structured knowledge silos to environments the place multimodal data — paperwork, sensor streams, SCADA alerts, and unstructured inputs — may be reasoned over in a single workflow. This shift permits the deployment of digital employees, brokers able to finishing up multi‑step operational duties quite than merely analyzing knowledge.
Kapoor emphasizes that the trail to worth doesn’t start with automating total departments. The most dependable place to begin is the again‑finish operational work that consumes human time however follows clear logic — duties like stock replenishment, provider order administration, or guarantee verification. Digital employees enter these workflows as interns, studying firm‑particular guidelines and exceptions earlier than steadily taking over extra autonomy because the group turns into comfy with agentic conduct.
This staged strategy preserves institutional information that will in any other case disappear by means of turnover. A discipline technician might spend 15 to twenty minutes digging by means of guarantee documentation; a digital employee can retrieve and interpret the identical data immediately and retain that functionality completely.
Kapoor frames the broader lesson immediately:
“You hear in regards to the 5% of AI tasks that succeed and the 95% that fail. The distinction isn’t the mannequin. It’s whether or not the work is agentic from the beginning. When you design the system to take motion quite than simply analyze, you keep away from ending up with pilots that by no means go anyplace. The tasks that succeed are those the place the group is able to let the AI act, not simply observe.”
– Somya Kapoor, CEO of IFS Loops
Implementing Governance by way of Supervisor Agent Frameworks
The transition to agentic methods introduces new operational dangers, from knowledge publicity to reasoning errors. Kapoor notes that whereas constructing brokers have turn out to be commoditized, sustaining and monitoring them at an industrial scale stays the first hurdle to ROI. Reasoning fashions nonetheless hallucinate on factual duties, and with out guardrails, autonomous workflows can drift shortly.
To safe these workflows, Kapoor describes a supervisor‑agent mannequin that assigns each digital employee a transparent identification and audit path. This ensures that actions stay grounded in verified enterprise directions and company knowledge quite than free‑type mannequin output.
She outlines a 3‑half verification framework for governing autonomous brokers:
- Audit Trails: Every motion taken by an agent should be logged for actual‑time and historic assessment, creating accountability and traceability.
- Cross‑Model Validation: Supervisor brokers consider the work of operational brokers, performing as a second layer of reasoning to catch errors and implement alignment with enterprise guidelines.
- Operational Guardrails: Trigger notifications to alert human topic‑matter specialists when an agent encounters inconsistencies, edge circumstances, or deviations from anticipated conduct.
As organizations strategy the 2030 retirement cliff, the flexibility to operationalize institutional information turns into a defining aggressive benefit. Leading corporations are adopting a purchase‑over‑construct bias, focusing inner engineering on integration quite than infrastructure.
This permits groups to focus on the small proportion of pilots that efficiently attain manufacturing by prioritizing AI‑prepared knowledge and narrowly outlined workflows. By combining predictive upkeep with safe agent governance, producers can stabilize operations in opposition to rising downtime prices.
