EN

CUSTOMER CASES

Customer cases

Customers

Manufacturing

A planning assistantMany answers production planning needs lived only in spreadsheets or with experienced hands. Asgard organised each department's records into a company knowledge base. The planning assistant answers with scheduling and time estimation skills, and reads the ERP and shop floor systems read-only.An optical lens manufacturer

Agents for in-house systemsEvery in-house system this company added meant redesigning the process around it. Asgard loaded SAP and system data into a data lake, with semantic models above it for Mimir. Asgard wrote each system's APIs as skills: staff give one instruction, and each system's agent does the operating.A materials manufacturerDeck · 13

Distribution, services and business processes

Lookups and formsStaff signed in to systems from different vendors in turn, copying and pasting between them. Asgard loads each system's data into a data lake, and managers ask Mimir in one sentence about repair trends, stock and attendance. Agents operate those systems the way staff do, and stop for approval before anything is written or sent outside.A foodservice equipment distributorDeck · 7

A customer receives agent configuration and a knowledge base. Configuration, requirements and decision records stay in the customer's own Asgard workspace.

Demonstration projects

Simulated scenarios and figures

Manufacturing

Rush order assessmentWhen a customer adds a rush order, sales, production planning, purchasing and outsourcing each check the systems in turn before anyone can say yes or no. The agents read the customer's email and drawing, and query numbers in the ERP, MES and outsourcing systems through semantic models. The agents lay out an order plan with material, scheduling and heat treatment skills, and stop for the plant manager's and the sales manager's approval before anything is written back.Demonstration projectTraditional manufacturing

WAT drift root causeThe YMS weekly report shows the WAT margin of one product closing in lot after lot. The agent uses semantic models to join YMS, MES and EQP data and finds the furnace that every drifting lot went through. Before the lot holds, the early PM and the note to the customer are written back, the agent stops for approval.Demonstration projectSemiconductor

ECN impact assessmentA customer sends an ECN after hours, and the departments used to wait until Monday to sign off one by one. This demonstration project writes the rules and exceptions engineering CAM uses to judge a revision change into skills. The agents assess the change by these rules, Sindri gathers a decision pack, and nothing is written back without approval.Demonstration projectPCB manufacturing

Retail and e-commerce

Store stockout responseOn Sunday night the POS day-end close shows that a best seller at the flagship store will sell out tomorrow. The agent uses semantic models to check stock and promotions across stores, the distribution centre and e-commerce, and finds stores that can transfer stock and a restock option. Before transfer orders, the restock purchase order and the promotion change are written back, the agent stops for the operations manager's approval.Demonstration projectRetailDeck · 8

Tracing a food cost overrunAfter the month-end close, some stores' food cost rates pass the alert line, and the CFO needs the cause before the board meeting. The agents use semantic models on the finance, POS and SCM systems to compare cost rates, actual usage and supplier prices. The agents draft recipe corrections and a supplier renegotiation, and stop for approval before anything is written back.Demonstration projectRestaurant chain

Member marketing and service ticketsMember marketing needs the abandoned-cart customers it can win back, and the customer service supervisor needs the tickets at SLA risk. The agents use query APIs to check best sellers, channel and store KPIs, low-stock SKUs and parcels with delivery exceptions. The agents use write tools to create coupons, member segments and stock transfer orders, and to escalate service tickets.Demonstration projectRetail

Restaurant lookups and writesThis demonstration project has customer service, operations, procurement and inventory, finance, and R&D agents. With the restaurant chain semantic model and their own toolkits, the agents look up revenue, margins, stock, suppliers, the menu and bookable time slots. The agents can also create purchase orders and add marketing campaigns.Demonstration projectRestaurants

Distribution, services and business processes

Expense claims and manager sign-offAn employee needs to claim an expense, but does not want to work through the complex UOF web pages. The agent reads each of the company's forms, with its fields, required items and valid options, from skills. The agent fills in forms for employees and releases forms for managers, and repeats the content back for the person's confirmation before anything is sent.Demonstration projectE-approval

Peak-season repair dispatchIn midsummer, the air conditioning at a VIP chain store stops, and the contract sets a time limit to arrive and fix it. The agent reads the service contract, checks technician load and parts stock with semantic models, and lays out a technician and parts plan. Before the technician assignment, the parts transfer and the emergency purchase order are written back, the agent stops for the dispatch manager's approval.Demonstration projectAppliance service

Reply on a late shipmentA major customer calls to ask where a late urgent shipment is and when it will arrive. The agent finds the shipment status in dispatch and fleet records, and reads the contract terms to work out the delay penalty. The new arrival time and apology for the customer go out only after the customer service supervisor approves them.Demonstration projectTransport and logistics

Delivery recovery before UATA fixed-price project is a few weeks from UAT entry, defect fixes cannot keep up with new defects, and a key module is blocked. The agent reads the UAT entry criteria and the client's escalation email, checks defects, plans and staffing with semantic models, and lays out a recovery plan. Before the plan revision, the staff reassignment, the scope change addendum and the client reply are written back, the agent stops for the delivery director's approval.Demonstration projectIT consulting

Warranty waiver reviewThis demonstration project handles a disputed warranty repair: the warranty terms allow refusal because of an indicator, but the manufacturer's service bulletin allows a warranty waiver. The agent compares the bulletin's conditions, the warranty policy, the purchase date and the repair history one by one, and sets out the grounds for the waiver. Until the customer service manager approves, the waiver, the last part's allocation, the rush purchase and the manufacturer claim stay out of the systems.Demonstration projectConsumer electronics distribution and service

After-sales operating metricsThe service manager needs engineer load, work orders about to expire and common failure modes. This demonstration project builds the after-sales service semantic model on the service data. The queries are built as tools that only read data and do not write back to any system.Demonstration projectAfter-sales service

Finance and hospitality

Overbooking responseThe night audit finds a room type overbooked for tomorrow. Overnight, the agent checks room status, loyalty tiers and the group contract, and lays out a plan. Before upgrades, relocations and guest notices are written back, the agent stops for the reservations manager's approval.Demonstration projectHotel

Funding a large redemptionA large redemption must settle at T+2, and the investment, trading and risk desks used to rush their own calculations before the cut-off. In this demonstration project, the agent checks the redemption instruction, lays out where the cash comes from, estimates selling pressure and verifies the investment policy limits. The agent writes the block order and the client confirmation into the systems only after the chief investment officer and the head of risk approve them.Demonstration projectInvestment management

Get started

Book a two-hour workshop

AI feasibility review of a live workflow

Book a workshop About our services