Planning AI reaches a quantified milestone as Cencora reports 10% forecast improvement

This report draws on 482 source articles published from 12 to 18 October 2026 (Europe/Paris), grouped into 395 distinct stories. New this week, Cencora reported a 10% forecasting improvement on a limited set of pharmaceutical SKUs, Waabi and Warrior Logistics reported live autonomous-trucking results in Texas, and South Korea approved new robot trials while Hyundai and Kia outlined a longer-term autonomous-factory roadmap.

This week in 60 seconds

• Planning gets a measured result: Cencora says early AI demand sensing improved forecasting by 10% on a small number of pharmaceutical SKUs.
• Transport shows contrasting evidence: Waabi and Warrior report on-time Texas truck operations with no safety incidents, while freight operators still face cost pressure and a gap between AI investment and demonstrable returns.
• Warehouse activity broadens: CEVA and Zalando are applying dual-arm robots to unfamiliar fashion returns, while other companies are upgrading warehouse systems and testing new robotics.
• AI capacity remains a supply-chain story: Chip, memory, power and data-centre constraints are prompting investment and localization, but demand, financing and project timelines remain uncertain.
• Skills and security stay operational: Research finds widespread workforce-readiness concerns, alongside growing attention to vendor, software-package and AI-agent risks.

10% forecasting improvement Cencora reports from early demand sensing on selected SKUs 87% less transport-management time Phleetto reports from operations outside the UK 40% of Australian organisations surveyed by BSI reported a supply-chain cyber incident in the previous year 83% of supply-chain leaders in ASCM–DSCI research say their workforce is not ready to use AI effectively

Planning and Inventory

This week’s planning stories range from a limited, quantified demand-sensing result to broader programmes that are still establishing connected data and decision processes. The contrast is useful: an improvement can be reported without yet demonstrating that it transfers across products, sites or disruption conditions.

Cencora reports better pharmaceutical forecasts Deployed

Cencora introduced an AI demand-sensing tool for pharmaceutical procurement. It compares historical forecasts with incoming order patterns, flags emerging changes and explains proposed adjustments. The company reports a 10% forecasting improvement, with early use limited to a small number of SKUs and buyers.

Cencora says improved forecasts could support availability or reduce excess inventory. It is also exploring knowledge graphs connecting manufacturer, FDA, geopolitical and financial information to identify shortage risks.

Morrisons replaces spreadsheet workload planning Deployed

Morrisons is rolling out a workload-planning platform from Logile across supermarkets, convenience stores and its logistics network. The system translates centrally defined tasks into local plans and performance dashboards, replacing spreadsheet-based processes.

The retailer says the change should reduce central administrative work and give store managers more reliable, locally tailored plans. The operational question is whether those plans improve execution across locations without adding new data or maintenance burdens.

US Air Force renews PTC inventory planning Deployed

PTC and the U.S. Air Force renewed their Servigistics relationship under the ESCAPE program, continuing work that began in 2017. The next phase expands forecasting and multi-echelon inventory optimisation to align service parts with aircraft and system readiness.

Exception-based management is intended to let planners focus on higher-value decisions. For an asset-intensive network, the central operating test is whether better parts positioning translates into readiness while avoiding unnecessary inventory.

Why it matters: Cencora offers a concrete early measure, but only within a narrow initial scope. Buyers should compare such results with broader programmes by tracking forecast quality alongside availability, inventory and the workload of managing exceptions.

Procurement and Suppliers

Procurement activity combines new sourcing tools and funding with a more practical push to connect supplier, spend and contract information. Critical-materials sourcing also shows how procurement objectives can blend commercial continuity with geopolitical diversification.

Maine Pointe embeds dSilo sourcing agents Announced

Maine Pointe expanded its partnership with dSilo, embedding agents in the sourcing module of its AI-powered TVO Control Tower. The software analyses contract, supplier, spend and pricing data to identify issues such as revenue leakage, non-compliance and purchase-price variance.

The companies point to a prior project that processed 350,000 supplier invoices to provide SKU-level pricing visibility. That volume suggests a data-intensive use case, but invoice processing alone does not establish realized savings or the effort needed to resolve flagged issues.

Scimplify raises $25 million for sourcing Funding

Scimplify raised $25 million, with Hitachi among the backers, for its AI-powered critical-materials sourcing platform. The company says its annualized revenue run rate has tripled to $200 million and plans to use the funding for U.S. expansion and model development.

Its stated proposition is to help reduce Western reliance on Chinese sources of rare earths, lithium and specialized alloys. For procurement teams, AI matching is only one part of the task: supplier qualification, material specifications, traceability and alternative capacity still determine whether a source is usable.

Ambio funds recycled-plastics procurement Funding

Ambio raised €5.1 million in seed funding to build an AI procurement platform for recycled plastics. Its approach combines supplier verification, structured material data, AI matching and logistics, serving customers in the Nordics and German-speaking Europe with materials sourced from recyclers across Europe.

The model targets a sourcing problem where material attributes and supplier reliability matter alongside price. Its supply-chain value will depend on whether structured information and verification help buyers secure suitable material consistently.

Why it matters: The strongest procurement use cases in this week’s notes connect AI to usable supplier and transaction records, rather than treating recommendations as a substitute for qualification. Decision-makers should check whether tools improve sourcing outcomes and compliance without obscuring how suppliers were assessed.

Warehousing and Robotics

Robotics is reaching different stages of warehouse work, from live returns handling to trials and long-range factory plans. In parallel, systems and workflow changes remain important: physical automation does not replace the need to coordinate inventory, tasks and people.

CEVA and Zalando automate fashion returns Deployed

CEVA Logistics is deploying Sereact dual-arm robots at sites in Greven, Germany, and Swiebodzin, Poland, to process returns for Zalando. The robots are designed to identify, grasp and sort fashion items without fixed SKU profiles or item-specific training.

The use case targets a labor-intensive reverse-logistics task in which returned products may vary from one item to the next. The deployment account does not provide measured productivity or cost results, so the business case remains difficult to compare with other returns operations.

Kesko selects Manhattan warehouse platform Announced

Kesko selected Manhattan Associates’ cloud-native warehouse platform to replace its grocery-trade legacy WMS. The planned rollout begins at five distribution centres, with a later greenfield site and terminal-hub operations also in scope.

The system includes Manhattan Active Agents and Agent Foundry. Kesko’s stated aim is greater real-time operational agility, but the rollout is a technology and process transition, not yet evidence of improved throughput or service.

South Korea approves mobile-robot trials Announced

South Korea approved a first quadruped-robot patrol trial on pedestrian walkways and a special exemption for 20 mobile humanoids at semiconductor, automotive, shipbuilding and logistics sites. Maum AI is to trial patrol robots in Cheonan, while Robros is to deploy bipedal robots for production tasks at four sites or companies.

The approvals create room for testing in settings where safety, movement around people and the reliability of physical tasks are central. The ministry plans manufacturing cybersecurity standards by early 2028 and technical standards addressing “AI washing” by year-end.

Hyundai and Kia outline factory roadmap Announced

Hyundai and Kia outlined plans for autonomous operating factories, including humanoids in new plants from 2032. Demonstrations included vehicle-exterior inspection, standardized logistics robots, virtual commissioning with Nvidia and Polaris, a manufacturing AI agent being piloted at Hyundai’s Ulsan EV plant.

The companies say automation is intended to shift workers toward safer, higher-value tasks rather than reduce headcount. These are long-range plans and company intentions; they should be distinguished from present-day operating results.

Why it matters: Returns sorting, warehouse management and factory robotics address distinct constraints and should not be evaluated with one automation metric. Buyers need task-specific measures—such as item handling, exception rates, uptime, safety and labour redeployment—before assuming a pilot or roadmap will scale.

Freight, Transport and Delivery

This week brings reported performance from an autonomous freight lane, process gains from routing and freight software, and new tools for shipment identification and delivery proof. At the same time, cost pressure and field conditions remain significant: a planned route or digital workflow must survive the realities of operations.

Waabi and Warrior report Texas operations Deployed

Waabi and Warrior Logistics announced autonomous-truck operations on the Dallas–Houston corridor, integrated with Warrior’s existing network. The companies report 100% on-time pickup and delivery and zero safety incidents since commercial operations began earlier in 2026.

The partners plan to identify additional lanes. The reported service and safety figures offer a live operating signal, but the notes do not establish the duration, volume or conditions behind them, limiting comparisons with other freight networks.

SimpliRoute reports routing gains Analysis

SimpliRoute says its routing platform is evolving into an AI-native TMS using agentic AI. The company reports results of 10% fewer vehicles, 30% lower vehicle operating costs and 25% more deliveries per truck for clients.

It aims for agentic AI to reach 80% of revenue and is expanding in Brazil and the United States. The cited results are company-reported; route density, service requirements and local operating conditions will affect how readily other fleets can reproduce them.

Carrier Logistics identifies missing-number shipments Deployed

Carrier Logistics and LSP44 launched a live AI-powered API that helps locate less-than-truckload shipments when a PRO tracking number is missing or incorrect. Users can search using company names, delivery locations, shipment descriptions and approximate dates, with likely matches ranked from about a week of records.

The tool targets manual customer-service research that the partners say can cost carriers upward of $10 per shipment. The participating-carrier and customer counts have not been disclosed, so its reach is not yet clear.

Phleetto enters UK freight-software market Pilot

Phleetto is recruiting UK pilot customers that manage more than 100 full-truckload movements a month. Its software targets freight coordination through spreadsheets, email and phone, offering tendering, carrier assignment, documentation and reporting alongside existing ERP systems through APIs and EDI.

The company cites results outside the UK of 87% less transport-management time and 15% lower spot costs. Those figures are not results from the UK pilot, whose purpose is to test the approach in a new market.

Why it matters: The freight evidence is promising but uneven in scope: a reported autonomous lane, company-reported routing gains and tools that target administrative friction. Operators should compare results with a baseline that includes service reliability, safety, route conditions, exception handling and total operating cost.

Visibility, Data and Agents

The new emphasis is on getting intelligence into operational work: supplier-risk data in everyday tools, connected planning across functions and agents that can assist with tasks. A live freight voice-agent expansion adds customer interactions to the mix, but system access and decision rights remain key design questions.

Resilinc brings supplier risk into Copilot Announced

Resilinc launched a Microsoft Copilot plugin giving procurement and supply-chain teams natural-language access to organisation-specific supplier, disruption and compliance intelligence. Built on the Model Context Protocol and Resilinc’s supply-chain graph, it is intended to help assess exposure to tariffs, sanctions, forced labour and rules including the EU Deforestation Regulation.

The operational promise is to reduce the distance between risk information and the people making decisions. Its usefulness will depend on whether underlying supplier records are current and whether users can trace how an answer was generated.

CMA CGM expands multilingual voice agents Deployed

CMA CGM and HappyRobot are expanding a multilingual AI voice-agent deployment after pilots in sales, customer service and finance. The sales agent Eva has contacted more than 4,000 customers in the UK and Senegal; approximately one-third reportedly moved into active discussions about shipping needs.

The group aims to deploy Eva in 20 countries by the end of 2026. A broader rollout will test how well the agent handles language and customer context across markets, as well as how staff manage handoffs and exceptions.

FourKites ties service to outcomes Announced

FourKites announced an “Outcome-as-a-Service” model linked to measures such as on-time-in-full performance, expedited-freight spend and cost to serve. The company says its network graph spans more than 1,600 brands, 500,000 carriers and 3.6 million facilities in 176 countries.

FourKites also set a goal of using AI agents to complete one billion hours of operational work for customers over the next decade. That is a company target rather than a verified result, and the measures attached to the service will matter more to buyers than the scale of the ambition alone.

Why it matters: Bringing information into an existing workflow can reduce friction, but it also makes data provenance, permissions and accountability part of the operating design. Buyers should define which actions agents may recommend, prepare or execute, and how exceptions are reviewed.

Security, Risk and Regulation

This week’s risk stories show supply-chain exposure at two levels: cyber incidents affecting trading relationships, and weaknesses in software and AI-agent ecosystems. Regulation is also moving toward clearer controls for physical AI, export-sensitive technology and systems used in public functions.

BSI finds cyber risk remains reactive Survey

A BSI study of more than 1,400 managers found that 40% of Australian organisations had suffered a supply-chain cyber incident in the previous year, while 60% managed risk reactively. Respondents cited gaps in supplier visibility and coordination, and two-thirds said AI would increase theft and cybercrime risk.

These findings point to a tension: AI can help assess exposure, but it may also expand the number of tools and connections that need oversight. Risk programmes that rely on periodic checks may struggle to keep pace with changing vendors and software dependencies.

Bloom flags npm package exposure Analysis

Bloom Security reported that more than 1,500 npm packages could be exposed to takeover through a flaw in how npm Trusted Publishing linked trust to a GitHub repository’s changeable name rather than its immutable identifier. The researchers said malicious package versions could be published under legitimate names and installed automatically by AI agents.

Bloom measured about 70 agent-installed packages per endpoint monthly. npm has remediated the issue; the case illustrates why AI agents’ software dependencies need the same provenance and update controls as other production components.

South Korea plans manufacturing standards Announced

Alongside approvals for mobile-robot trials, South Korea’s ministry plans manufacturing cybersecurity standards by early 2028 and technical standards addressing “AI washing” by year-end. The measures respond to growing use of AI and robots in industrial environments.

For logistics and manufacturing operators, the rules could make qualification and security requirements more explicit. Companies testing physical AI will need to follow how standards define system claims, access controls and responsibility for failures.

INEC limits AI in election logistics Announced

Nigeria’s Independent National Electoral Commission says it is developing an AI governance framework for electoral logistics, anomaly detection and election security. It has also described geospatial routing to 176,846 polling units and inventory forecasting, while saying AI will not decide election outcomes.

The commission says automated systems will be audited and full automation of results collation or accreditation will not be permitted. The example underscores how explicit boundaries around AI use can be part of operational governance, not just technical design.

Why it matters: Supply-chain cyber controls must extend to vendors, software packages and agents that can install or act on behalf of users. In regulated or high-consequence operations, documented limits and auditability help make that exposure governable.

AI Infrastructure and Capacity

AI-related demand continues to redirect investment toward chips, servers, power equipment and regional production. But capacity announcements sit alongside constraints in memory, power, financing and project timing, making the infrastructure build-out a source of both opportunity and supply risk.

ASML warns planned capacity may fall short Analysis

ASML reported record third-quarter 2026 net sales of €11.662 billion and net income of €4.96 billion, and raised the lower end of its 2026 sales forecast to €44 billion. The company warned that planned 2027–28 capacity may not meet demand and is reviewing whether to expand further.

ASML cautioned that customer forecasts are not firm orders, so the scale and timing of any capacity increase remain uncertain. For supply-chain planners, demand signals upstream do not automatically translate into equipment availability or dependable lead-time relief.

Taiwan weighs Texas AI supply-chain hub Announced

Taiwan is considering up to US$4 billion for an industrial park in Texas intended to support Taiwanese suppliers and U.S. partners, including Nvidia’s ecosystem. Reported aims include AI-server assembly, semiconductor manufacturing and a more resilient U.S.-based supply chain.

The project is part of a broader investment push and is not completed capacity. Its importance will depend on which suppliers participate, when production begins and how the site fits into existing component networks.

Hyosung doubles circuit-breaker capacity Deployed

Hyosung Heavy Industries opened an ultra-high-voltage circuit-breaker plant in Changwon, South Korea, doubling production capacity. The company says the expansion addresses power-grid equipment shortages linked to AI growth and supports a global supply-chain push involving local production or exports for India, the United States and Europe.

Power equipment is a less visible but essential input to data-centre expansion. Increased factory capacity is a tangible response, although the notes do not quantify how much it will ease customer lead times.

FICG expands Malaysian optical capacity Announced

Taiwan’s FICG plans to more than triple Malaysian manufacturing capacity for optical components used in AI data-centre networks by the end of 2028. The company describes the move as a China+1 response to requests from U.S. customers and rising demand for optical communications parts.

The expansion adds a regionalization dimension to AI infrastructure sourcing. The stated timeline means buyers will need to manage current supply arrangements while evaluating future capacity and qualification requirements.

Why it matters: AI infrastructure growth is creating new demand across the supply base, but announced capacity is not the same as available, qualified supply. Procurement and operations teams should map dependencies beyond chips to power, optical parts, memory and the timing of regional projects.

Adoption, Skills and Sustainability

Survey evidence points to a gap between organisations’ appetite for AI and the workforce’s readiness to use it effectively. Meanwhile, sustainability coverage is moving from reporting toward operational data and choices, while logistics programmes in Singapore pair AI adoption with worker reskilling.

ASCM–DSCI finds workforce-readiness gap Survey

A joint ASCM–DSCI report says 83% of surveyed supply-chain leaders consider their workforce unprepared to use AI effectively, while 91% of organisations plan to use AI to augment workers. Adoption across 13 supply-chain functions averaged 1.94 on a five-point autonomy scale.

The report identifies readiness, trust, cybersecurity gaps and uncertain return on investment as barriers. Organisations with AI-ready workforces reportedly deployed AI across more functions and had a 71% project success rate, linking workforce capability to the ability to implement projects.

NTUC and ASME support SME reskilling Announced

Singapore’s NTUC and ASME launched the SME Transformation Accelerator, expecting about 120 SMEs in retail, professional services and logistics to participate and at least 480 workers to benefit from wage increases or skills allowances. The programme combines AI-enabled solutions with job redesign and reskilling.

Qualifying costs can receive up to 70% co-funding through the NTUC Company Training Committee Grant. The design makes workforce transition part of the adoption plan rather than a separate activity.

MIT survey finds sustainability governance lags Survey

MIT’s 2026 sustainability report, based on 1,810 professionals across 91 countries, found that 83% consider sustainability important or very important to long-term business success—the highest share in the report’s seven-year history. Among companies with public goals, 51% had a formal cross-functional team and 33% had board-level oversight.

Respondents cited tariffs, geopolitical conflict, measurement gaps and AI-governance issues as obstacles to decarbonisation. Separately, a Scope 3 analysis argues that credible management needs activity data, governed emission factors, transparent allocation and data lineage so teams can identify operational levers such as modal shift, utilisation, routing and carrier selection.

Why it matters: AI adoption plans need to account for the people who use the systems and for the quality of the operational data they depend on. Sustainability data is most valuable when it can inform sourcing and transport decisions, not merely produce a disclosure.

Expert Analysis

This week’s most consequential pattern is that the evidence is becoming more specific in some operational pockets while remaining difficult to generalize. Cencora reports a 10% forecasting improvement on a limited number of pharmaceutical SKUs, and Waabi and Warrior report service and safety results from a live Texas corridor. Those signals add measurable detail to the recent movement of AI into operational work, but neither establishes repeatability across a broad network.

There is also a clear contrast between new capacity and persistent constraints. Taiwan’s potential Texas hub, FICG’s planned Malaysian expansion and Hyosung’s opened plant address localization and infrastructure needs, while ASML warns that planned capacity may still fall short and reports describe pressure on memory, power and data-centre projects. For supply-chain leaders, the risk is not simply a shortage of AI hardware: investment timing and the availability of supporting inputs can also affect production plans and supplier commitments.

Finally, the week’s adoption evidence reinforces the human and control dimensions of scaling. The ASCM–DSCI finding that 83% of leaders see workforce-readiness gaps sits alongside Singapore’s funded reskilling effort and the BSI survey’s cyber-risk concerns. In contrast to a purely technology-led story, these findings suggest that training, supplier oversight and defined authority are becoming practical determinants of whether AI can be used safely in daily operations.

What to watch next

• Whether Cencora extends demand sensing beyond its initial small set of SKUs and reports results tied to availability or excess inventory.
• Whether Waabi and Warrior add lanes and provide further detail on operating duration, volume, service conditions and safety performance.
• How CEVA and Zalando measure returns-robot productivity, exception handling and cost at Greven and Swiebodzin.
• Whether Taiwan’s proposed Texas hub and FICG’s Malaysian capacity plan reach defined project milestones toward their stated production goals.
• Whether South Korea’s planned manufacturing cybersecurity and AI-washing standards produce clear requirements ahead of their announced timelines.
• How Singapore’s SME accelerator translates its expected participation into logistics-worker training and job redesign.