Infor VISUAL + on-prem AI
AI for Infor VISUAL ERP: Answers and Agents for Job Shops
Short answer
Infor VISUAL runs on a SQL Server database with a documented schema and an API Toolkit, which makes it a reasonable target for grounded natural-language question answering and light automation without touching the core application. On-prem AI reads job, routing, estimate, and inventory tables directly, so a planner or estimator gets an answer in seconds instead of running a saved report or waiting on a scheduler. The right entry point is usually quoting turnaround and job status visibility, both of which are read-only and low-risk to pilot.
- ERP
- Infor VISUAL, VISUAL ERP, VISUAL Enterprise
- Industries
- Manufacturing, Job Shop, Make-to-Order, Contract Manufacturing
- Written for
- Operations Manager
If you run VISUAL for a job shop or make-to-order manufacturer, you already know its strength: a straightforward relational schema that a competent SQL developer can query in an afternoon. That same schema is what makes VISUAL a good fit for retrieval-augmented question answering, because you are not reverse-engineering a black box, you are joining PART, JOB, JOB_ROUTE, and ESTIMATE tables the way your reporting team already does.
The gap most operations managers describe is not a lack of data, it is a lack of time to get at it. An estimator re-quoting a part wants to know what you charged the last three times you ran it and what actually happened to margin, not just what the estimate said. A scheduler wants to know which jobs are behind because of a specific work center, not a generic late-job report. Those are natural-language questions today, run through Crystal Reports or ad hoc SQL by whoever has access.
VISUAL shops skew toward smaller IT teams, sometimes a single administrator who also owns the network and the phones. That changes the calculus for AI: you do not want a project that requires a dedicated data engineering team to maintain. A connector that reads the VISUAL database on a schedule, keeps a semantic layer in sync, and answers questions through a model running on a single GPU server in the server closet is a realistic scope for a shop this size.
Job shops also carry disproportionate tribal knowledge. The senior estimator who has quoted every recurring part for fifteen years is a single point of failure. Grounding a model on historical estimates, routings, and job cost variance does not replace that judgment, but it gives a newer estimator or a covering planner a documented starting point instead of a guess, and it gives the outgoing expert something worth capturing before they retire.
What usually gets in the way
The problems we hear most from operations manager teams running Infor VISUAL.
Quoting turnaround depends on one or two people
RFQs sit in a queue until the senior estimator has time, because pulling comparable historical jobs and routings from VISUAL means running a report or writing a query. A slow quote loses work to a shop that answers faster.
Shop floor status is a phone call away
Shop Floor Data Collection captures labor and move transactions, but turning that into 'is job 48213 going to ship on time' still means a planner opening several screens or asking the supervisor directly.
Engineering changes ripple quietly
An ECN against a PART_ROUTING or bill of material does not automatically tell you which open jobs, purchase commitments, or in-process work are affected. That impact analysis is manual and easy to miss under deadline pressure.
WIP and job cost variance are reviewed too late
Labor and material variance against estimate shows up on a job cost report after the job closes. By then the pattern that caused the overrun has usually repeated on two or three other jobs.
Tribal knowledge is not written down
Which vendors are reliable for which commodities, which work centers are the real bottleneck, which customers change specs mid-job: this lives in a handful of people's heads and leaves when they do.
Where AI earns its place in Infor VISUAL
Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.
Quote-history lookup for estimators
An estimator asks what a similar part quoted for, what the routing looked like, and how actual cost compared to the estimate on the last few runs.
Touches: PART, ESTIMATE, ESTIMATE_ROUTING, JOB, JOB_ROUTE, JOB_MATERIAL
Outcome: Cuts the research portion of a re-quote from a manual report pull to a direct answer, freeing time for the judgment calls that actually need an experienced estimator.
Job status and exception summary
A planner or customer service rep asks which jobs are behind schedule, what caused the delay, and which are at risk of missing a promise date.
Touches: JOB, JOB_ROUTE, OPERATION, JOB_TRANSACTION (Shop Floor Data Collection)
Outcome: Turns a status question that used to mean checking three screens into one grounded answer, with the source job records shown alongside it.
Scheduling and capacity exception triage
An agent flags work centers running behind, jobs stacked against the same bottleneck resource, and suggests which late jobs are candidates for overtime or outsourcing.
Touches: WORKCENTER, JOB_ROUTE, RESOURCE, Advanced Planning and Scheduling data
Outcome: Surfaces the two or three work centers actually driving lateness this week instead of a flat list of every late job, which is where scheduler time is usually wasted.
Engineering change impact assessment
When an ECN changes a bill of material or routing, an agent lists open jobs, purchase commitments, and WIP that reference the old revision.
Touches: PART, PART_ROUTING, BOM structure, PURCHASE_LINE, JOB_MATERIAL
Outcome: Reduces the chance a change ships against the wrong revision by giving engineering a checked impact list before the change goes live, not after a nonconformance.
Expedite and late-purchase follow-up
An agent drafts a status-check email to a supplier for purchase lines tied to jobs that are now at risk, referencing the PO number and required date.
Touches: PURCHASE_LINE, PURCHASE_ORDER, JOB_MATERIAL, VENDOR
Outcome: Cuts routine expedite follow-up from a buyer manually drafting each email to a reviewed, one-click send for the lines that actually matter.
Nonconformance and corrective action drafting
A quality technician describes a defect and the agent drafts a structured corrective action request, pulling the part, job, and prior similar events for context.
Touches: CORRECTIVE_ACTION_REQUEST, INSPECTION, PART, JOB
Outcome: Shortens the time to a usable draft CAR from twenty minutes of typing to a few minutes of review and edit, without changing the approval workflow.
Lot and serial traceability lookup
A quality or customer service person asks which jobs and shipments a specific lot or serial number touched, for a recall or customer complaint.
Touches: LOT, SERIAL, INVENTORY_TRANS, JOB_MATERIAL, SHIPPER
Outcome: Turns a traceability request that used to take an experienced planner twenty to thirty minutes into a query answered in under a minute, with the underlying records cited.
Reference architecture
The architecture reads VISUAL's SQL Server database (directly or from a replica), keeps a semantic layer of tables, views, and business terms in sync, and serves answers through a model running on hardware you control. No write-back happens without a person approving it, and every answer can be traced to the rows it came from.
- 1
VISUAL connectors
Read access to the VISUAL SQL Server database or a replica, plus the VISUAL API Toolkit for anything not cleanly reachable by SQL view, such as Shop Floor Data Collection transactions.
- 2
Data and semantic layer
A mapping between raw VISUAL table and column names and the business vocabulary your team actually uses, so 'late jobs' resolves to the right join across JOB and JOB_ROUTE, consistently.
- 3
Model serving
An open-weight model (Llama, Qwen, or similar class) served with vLLM or Ollama on a GPU server sized to your concurrent user count, running inside your network.
- 4
Retrieval and agents
Retrieval-augmented generation over the semantic layer for read questions, plus narrowly scoped agents (expedite emails, CAR drafts) that stop at a draft until a person approves.
- 5
Governance and audit
Role-based access that mirrors VISUAL security groups, a visible query behind every answer, and a log of every question asked and every draft action taken.
Integration notes for your ERP team
- VISUAL's SQL Server schema is well documented; most read use cases can be built on views over the core tables without touching the VISUAL API Toolkit.
- Shop Floor Data Collection transaction data (labor, move, and inspection events) is the richest source for real-time job status and usually needs the API Toolkit or a change-data-capture feed rather than a static nightly extract.
- Estimating data (ESTIMATE, ESTIMATE_ROUTING) is often the highest-value source for a first pilot because it is self-contained and low risk to expose.
- If you are on a hosted or CloudSuite variant of VISUAL, check what direct database access Infor permits; some hosted environments require going through the API layer instead of a database replica.
- Write-back (updating a job note, closing an operation) should go through the same APIs and validation VISUAL's own screens use, with a person approving the action first.
- Access control is enforced by re-checking the requesting user's VISUAL role and site/plant assignment before returning a result, not by trusting the model to filter correctly.
- For multi-site VISUAL installs, keep the semantic layer aware of which site a job belongs to so cross-site answers do not silently blend data that should stay separate.
Deployment options
Air-gapped on-prem
Shops running VISUAL for defense or aerospace-adjacent work, or anyone who simply does not want quote and cost data leaving the building.
Model, database connector, and application run entirely on hardware inside your network, with no outbound path required for normal operation.
Private or sovereign cloud
Shops that already run infrastructure in a private cloud tenancy or want GPU capacity without buying hardware.
Same architecture deployed in a dedicated tenancy under your control, with the ERP connector reaching back to your VISUAL database over a private network path.
Hybrid
Shops piloting on a single server before deciding on a permanent home, or running the model on-prem with a private cloud for backup and DR.
Model serving stays local for latency and control; less sensitive workloads such as document indexing can run off-site if that better fits your infrastructure.
Compliance and data control
How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.
CMMC 2.0 / DFARS 252.204-7012
Many VISUAL shops supply DoD primes. Keeping the model and data on infrastructure inside your CMMC boundary avoids sending CUI-adjacent job and cost data to a third-party API.
AS9100
For aerospace job shops, traceability, corrective action, and document control requirements are met by grounding answers directly in VISUAL records and keeping a log of every query, not by relying on model memory.
ITAR
Where jobs involve ITAR-controlled technical data, on-prem deployment avoids sending that data to a public model API, which can itself be a deemed export depending on where the model runs.
General data protection and access control
Access mirrors your existing VISUAL security groups, so a person can only get answers about jobs, customers, or cost data they could already see in the application.
Where Netray fits
ERPray
Natural-language question answering and dashboards over VISUAL job, estimate, and inventory data is exactly ERPray's job: read-only by default, shows the underlying query, and respects VISUAL role permissions.
Custom build
Shop-floor-specific agents such as expedite drafting or CAR generation are usually built as narrow, purpose-specific agents rather than general chat, and fit the custom build track.
How an engagement runs
Phase 1 . 2-3 weeks
Discovery
- -Review of your VISUAL schema, customizations, and current reporting stack
- -Access and hosting model decision (air-gapped, private cloud, or hybrid)
- -Shortlist of 2-3 pilot use cases with the highest ratio of value to integration effort
- -Data quality check on the tables the pilot will depend on
Phase 2 . 6-8 weeks
Pilot
- -Working connector and semantic layer for the pilot use cases
- -Model serving stood up on agreed hardware
- -Pilot group of estimators, planners, or schedulers using it against live data
- -Accuracy review against a set of known-answer questions
Phase 3 . 4-6 weeks
Production
- -Hardening of access control, logging, and audit trail
- -Rollout to the full estimating, planning, or quality team
- -Runbook for monitoring, updates, and incident response
- -Documented handoff of the semantic layer so your team can extend it
Phase 4 . Ongoing
Scale
- -Additional use cases added from the discovery backlog
- -Periodic model refresh evaluation as open-weight options improve
- -Usage review to prioritize what to automate next
Questions to ask any vendor, including us
A short list that separates real Infor VISUAL AI work from a chatbot demo.
- Does the AI read VISUAL directly, or does it require exporting data to a third-party SaaS platform first?
- Can I see the exact SQL or API call behind every answer, not just the generated text?
- Does access control mirror my existing VISUAL security groups, or is it a separate permission model I have to maintain?
- What happens to the model and the data if I cancel the contract? Can I keep running it myself?
- Who approves any action that writes back into VISUAL, and can that be turned off entirely for a read-only pilot?
- What hardware does this actually require, and what does it cost to run per month once it is in production?
- How is the semantic layer kept in sync as we add custom fields or modify VISUAL over time?
- Can you show me this working against a real (not simulated) VISUAL job history?
Frequently asked questions
Can AI work directly with Infor VISUAL's database without a middleware platform?
Yes. VISUAL's SQL Server schema is documented and query-friendly, so a retrieval layer can read job, estimate, and inventory tables (or a replica of them) directly. The VISUAL API Toolkit fills in gaps for transactional data like Shop Floor Data Collection events that are harder to query cleanly with raw SQL.
Is this only useful for large VISUAL shops?
No, the opposite is often true. Smaller job shops with a lean IT team benefit most from a system that answers questions directly instead of requiring someone to build and maintain a new report every time a question comes up. The infrastructure footprint (one GPU server) is modest.
Does this replace our estimator or scheduler?
No. It gives them faster access to historical data (past quotes, job cost variance, routing history) so they can make the judgment call faster and with better context. The estimating and scheduling decisions stay with the person; the AI removes the research bottleneck in front of that decision.
How does this stay compliant if we do defense-related job shop work?
Run the model and data pipeline on infrastructure inside your own network or CMMC boundary rather than a public API. That keeps job, cost, and technical data from leaving your control, and gives you an audit trail of every question asked, which is generally what auditors and primes are looking for.
What is the realistic first use case to pilot?
Quote-history lookup for estimators and job status Q&A for planners are the most common starting points: both are read-only, both are self-contained within a small set of VISUAL tables, and both save time every single day rather than solving an occasional problem.
Do we need to move off VISUAL onto CloudSuite to get AI?
No. AI built on direct database access or the API Toolkit works against the on-prem version of VISUAL you are already running. Moving to a hosted or CloudSuite version is a separate decision that affects which data access path is available, not a prerequisite for adding AI.
How long before an estimator or planner is actually using this day to day?
A focused pilot on one or two use cases typically takes six to eight weeks from kickoff to a working tool in front of a pilot group, assuming VISUAL database access and a clear use case are agreed on during discovery.
Related guides
AI for Infor SyteLine and CloudSuite Industrial
Add grounded AI to Infor SyteLine or CloudSuite Industrial: natural-language answers, agents over IDOs and ION, on-prem or CloudSuite deployment.
Infor XA on IBM i + on-prem AIAI for Infor XA on IBM i: On-Prem, No Replatforming Required
Add AI to Infor XA on IBM i (AS/400) without a migration project. Read DB2 for i directly, ground answers in live XA data, keep everything on-prem.
Job shop ERP + AIAI for ECI JobBOSS2 and E2 Shop System
Add AI to ECI JobBOSS2 or E2 Shop System for faster quoting, tighter scheduling, and shop-floor answers, without sending your job data to the cloud.
Shop floor + on-prem AIAI shop floor assistant for operators working inside your ERP
An on-prem AI copilot answers operator questions against your ERP work instructions, travelers, and routings in plain language, at the machine, without a screen full of menus.
Know what an ECO actually touchesAI for Engineering Change Impact on BOMs, Routings, and Open Orders
AI reads a proposed ECO against live BOM, routing, and open order data to show engineering exactly what a change touches before it gets released.
RFQ-to-quote + on-prem AIAI for quote-to-cash in manufacturing and job shop ERPs
AI reads incoming RFQs, drafts quotes from your ERP's own cost and pricing history, and flags parts your shop has never quoted before, keeping the estimator in control.
Plan it with numbers
Legacy ERP AI Modernization Assessment
Score your legacy SyteLine, LN, or Baan environment to find out whether AI can modernize it in place or whether platform upgrade work needs to come first.
Free ToolERP Integration Complexity Calculator
Estimate the hours, cost, and elapsed time of your ERP integration workstream from interface count, integration patterns, and middleware maturity.
Free ToolAI Project Cost Estimator
Turn project scope, integration count, data readiness, and team weeks into a defensible AI project budget with contingency built in.
GuideManufacturing ERP Selection Criteria: Decision Framework
Select the right manufacturing ERP with a structured decision framework. Functional requirements, TCO analysis, vendor evaluation, and risk assessment.
GuideAI Agents for ERP: The Complete Guide
Everything you need to know about AI agents for ERP systems. How they work, ROI expectations, implementation approaches, and real-world results.
GuidePreparing ERP Data for AI: A Practical Guide
Prepare ERP data for AI use: extraction patterns, schema documentation, and the data quality checks that determine whether your copilot is trustworthy.
Talk it through with an engineer who knows Infor VISUAL
Bring one real question your team cannot answer from the ERP today. We will map the data path, the model, and where it runs, and tell you honestly if AI is the wrong tool for it.