The data is there. The answers aren't.
You bought the ERP to see the business. Now every question turns into a report request, a CSV export, and a week of waiting. The information exists — the path from question to answer is what's broken.
AI agents · Acumatica ERP · Manufacturing IT
AI that understands your ERP, your shop floor, and how the work really runs.
Situational Business Development builds working AI agents on top of the systems you already run — Acumatica, Dynamics GP, your MES, your SQL estate — and consults on the IT that keeps a manufacturer moving. Forty years of enterprise engineering behind every recommendation. No pilot theater, no hand-off to juniors.
Read-only by default·Every action logged·Human approves the write
0+
Years building business systems, from batch mainframe to autonomous agents
0
Platform eras delivered in production — not studied, shipped
0
ERPs known end to end: Acumatica and Microsoft Dynamics GP
0
Senior engineer on your project — the one you talked to
The situation
Most companies do not have an AI problem. They have a situation.
The ERP holds the truth but nobody can get at it. The plant runs on spreadsheets that live on one person's laptop. A 20-year-old system still prints the labels and nobody dares touch it. And somewhere above all of it, someone is being told to "do something with AI." We start by reading the situation honestly — then we build only what changes it.
The plant knows things the system doesn't.
Real lead times, which vendor actually delivers, which routing is fiction. That knowledge sits in people's heads and side spreadsheets. When they retire, it leaves with them.
AI pilots that never become systems.
A chatbot demo impresses the board and then dies, because nobody wired it to the ERP, gave it rules, or decided who approves what. An agent without permissions, logging and an owner is a science project.
Capabilities
Four practices. One engineer who has done all four.
Most firms sell you one of these and subcontract the rest. Here they are the same person, which is why the AI work actually lands inside the ERP and the ERP work survives contact with the shop floor.
AI agents built for a company that already has systems
Not a chatbot bolted to your website. Agents that hold a job: watch something, reason about it with your rules, draft the work, and put a person in front of anything consequential.
What changesThe exception list gets worked before it becomes a shortage — and the person
working it starts from a drafted answer instead of a blank screen.
- Agent design & build — scoping a real job an agent can own, then building it on Claude and today's leading models
- Tooling into your systems — Acumatica REST, SQL, file drops, EDI, email, MCP servers so the agent can actually act
- Retrieval over your documents — SOPs, specs, routings, contracts, customer history, answered with citations
- Guardrails & approvals — least-privilege credentials, write policies, human-in-the-loop, full audit trail
- Evaluation before trust — replay against your real history and measure it before anyone relies on it
- Run & improve — monitoring, cost control, and turning every human correction into a durable rule
Getting more out of the ERP you already run
I don't sell implementations. You already have the system — the work is making it fit the way you actually operate, from someone who understands both the accounting and the production side of the same transaction.
What changesThe system stops being something people work around and becomes the place
the answer comes from.
- Extensions & customization — screens, fields, workflows and business logic the standard product doesn't cover
- Integrations — CRM, EDI, e-commerce, shipping, quality, payroll, bank feeds, custom endpoints
- Reporting that gets used — GI/inquiry design, dashboards, Report Designer, KPI ownership
- Configuration & data cleanup — item and BOM structures, duplicate masters, and the setup decisions quietly costing you money
- Microsoft Dynamics GP — keeping it healthy, reporting out of it, and connecting it to what you are building now
- Ongoing support — a retained engineer who already knows your configuration and your people
IT consulting for manufacturers and mid-market organizations
The unglamorous work that decides whether anything else succeeds: systems strategy, process, integration, and someone senior who can sit with both the plant manager and the controller.
What changesDecisions get made from one set of numbers, and the plant stops running a
parallel operation in Excel.
- Systems assessment — what you own, what it costs, what it is actually doing for you
- Fractional IT leadership — roadmap, budget, vendor management, an adult in the room for technology decisions
- Production & inventory process — BOMs, routings, work orders, scrap, cycle counting, costing that ties out
- Shop floor to office — MES, scanning, labeling, scheduling and quality data connected to the ERP
- Vendor and ISV selection — evaluations run by someone with no reseller commission at stake
- Disaster recovery & continuity — what happens the morning the server, or the ransomware note, arrives
Legacy modernization, data liberation and integration
Old systems are rarely stupid — they encode decades of business rules. The job is to extract the rules and the data without breaking the operation that depends on them.
What changesThe system nobody wants to touch becomes a documented, integrated part of
your stack — or retires cleanly, with its data intact.
- Legacy application archaeology — COBOL batch, UNIX C, PowerBuilder client/server, ancient SQL: read, documented, mapped
- Data migration — extraction, cleansing, reconciliation, and a defensible answer to "does it tie out?"
- Interim integration — keeping the old system fed while the new one takes over, instead of a big-bang gamble
- Rewrite vs. wrap decisions — an honest recommendation, including the cases where you should leave it alone
- Reporting continuity — the report the CFO has run every month since 1998, preserved to the penny
- Knowledge capture — before the person who wrote it retires
AI agents
Agents with a job description.
Every agent below is defined the same way: what it watches, what it does, and where it stops. That last one is the part most AI projects skip — and the reason they never get past the demo.
Order to cash
Order promise guard
WatchesReleased sales orders, on-hand and on-order supply, vendor lead times, capacity
DoesFlags promise dates that supply cannot support, drafts the purchase order or reschedule that fixes it
Stops atAny write to the ERP. The planner approves; the agent posts.
Planning
MRP exception worker
WatchesThe MRP exception list nobody has time to read all the way through
DoesTriages by real dollar impact, groups related exceptions, recommends the action for each cluster
Stops atExpedite fees and vendor commitments — those go to a buyer
Quoting
Quote & configuration assist
WatchesIncoming RFQs, drawings, prior similar jobs and their actual costs
DoesBuilds a first-pass BOM and routing, prices it against real history, shows its reasoning line by line
Stops atThe number you send the customer. An estimator signs it.
Finance
Close & reconciliation agent
WatchesSub-ledgers, inventory valuation, WIP, unbilled and uninvoiced, intercompany
DoesRuns the tie-outs before the controller does and explains each variance in plain language
Stops atJournal entries. It proposes; finance posts.
Knowledge
Institutional memory agent
WatchesYour SOPs, work instructions, specs, past tickets, contracts and email history
DoesAnswers "how do we do this here" with a citation, in seconds, for anyone on any shift
Stops atAnything it cannot source. It says so rather than inventing.
Data
Migration & integrity agent
WatchesLegacy extracts, master data, duplicate customers and items, dirty history
DoesProfiles, matches, proposes merges, and produces the reconciliation packet for sign-off
Stops atDestructive changes. Nothing is merged without an approved list.
How it is wired
Your systems stay the system of record. The agent is a very well-briefed colleague.
Agents get scoped credentials, a mapped view of your data in business language, your written rules, and tools that can only do what policy allows. They run on a schedule or on an event, they log everything they saw and did, and the default posture is read, analyze, draft — then wait.
- Least-privilege service accounts, never a shared admin login
- Write policies defined per action, not per agent
- Your data is not used to train anyone's model
- Cloud, private cloud or on-premises — the architecture follows your risk posture
Acumatica & Microsoft GP
ERP work that starts the day after go-live.
I am not an implementation shop and will not bid against one. You already have the system. My work is making it fit how you actually run — extending it, connecting it, reporting out of it, and building the AI layer on top of it.
01
Extend it
The screens, fields, workflows and business logic the standard product doesn't cover — built on supported extension points so your upgrades stay boring.
2–6 weeks typical
02
Connect it
CRM, EDI, shipping, quality, e-commerce, bank feeds, shop floor. Two systems that each believe they hold the truth, reconciled into one.
2–8 weeks typical
03
See into it
Generic Inquiries, dashboards and Report Designer, with one agreed definition per metric so two departments stop arguing about whose number is right.
1–4 weeks typical
04
Keep it healthy
A retained engineer who already knows your configuration, your customizations and your people by name — instead of a ticket queue that starts from zero every time.
Ongoing
Still running Microsoft Dynamics GP?
Plenty of companies still run GP perfectly well, and I am not here to sell you off it. What I do is keep it useful — and stop it from being the reason you cannot build anything new.
- Reporting and data extraction that doesn't depend on the one person who knows the SQL
- Integrations between GP and the systems that grew up around it
- An honest read on where GP is genuinely holding you back — and where it plainly is not
- If you do decide to move someday, advice from someone with no commission riding on the answer
Acumatica, then AI on top of it
This is the combination almost nobody offers. An ERP consultant who builds agents, or an AI shop that actually knows what a GI, a sub-item, or an inventory allocation is — you rarely get both, and the gap between them is where projects die.
- Agents built directly against Acumatica's REST and inquiry layer
- Automation designed with the upgrade path in mind — no unsupportable customizations
- Reporting and agent logic sharing one definition of a metric
- Start with support, add agents where the pain is measurable
Manufacturing & operations
Built for companies that make things.
Discrete and mixed-mode manufacturers, job shops, distributors with assembly, and the mid-market organizations that support them. The vocabulary is already familiar — you will not spend the first month teaching it.
Production & planning
Work orders, routings, capacity, scheduling reality vs. the schedule on the wall, and MRP output someone actually acts on.
Inventory & costing
Accuracy programs, cycle counting, lot and serial traceability, standard vs. actual cost, and margins that survive audit.
Shop floor connectivity
Scanning, labeling, machine and labor data collection, MES and quality systems talking to the ERP instead of past it.
EDI & customer demand
850s, 856s, releases, portals and forecast files translated into orders your planners can trust — without re-keying.
Quality & compliance
Nonconformance, CAPA, certifications, customer audits, and the document trail that has to exist before you need it.
Operating metrics
On-time delivery, OEE, scrap, true labor and machine utilization — defined once, reported the same way everywhere.
Experience
Five platform eras. The same discipline every time.
Every generation of technology was going to make the last one irrelevant. None of them did — they stacked. Knowing what still runs underneath is exactly what makes the AI layer on top trustworthy.
ERA 01 · CORPORATE MAINFRAME
Batch systems that were not allowed to fail
Corporate IT where the nightly cycle ran the company. You learned data integrity, restart-and-recovery, change control and the discipline of testing before deploying — because there was no "just push a fix."
COBOLJCLCICSDB2VSAMBatch scheduling
What carries forward
Data lineage and reconciliation instincts. When an AI agent proposes an adjustment, the first question is still
the mainframe question: where did this number come from, and what happens to everything downstream?
ERA 02 · MIDRANGE & UNIX
Server-side development, multi-user for real
UNIX server development and administration: writing the services, tuning the database, and owning the box the business ran on. Concurrency, performance and "who has that record locked" stopped being theory.
UNIX / LinuxCShellOracleInformixRDBMS designCron & daemons
What carries forward
Everything an agent runs on is still a server with limits. Scheduling, logging, retries, back-pressure and failure
modes are old problems wearing new names.
ERA 03 · CLIENT / SERVER
PowerBuilder and genuinely rapid development
Rapid business application development with PowerBuilder against Sybase and SQL Server — sitting with users, building the screen they described, showing it the same week. The original agile, before it was branded.
PowerBuilderDataWindowPowerScriptSybaseSQL ServerStored procedures
What carries forward
The habit of building with the user in the room. It is why prototypes here show up in days and why the
first version of an agent is something you can react to, not a document.
ERA 04 · ENTERPRISE ERP
Microsoft Dynamics GP, then Acumatica
Implementation, integration and support of the systems that run finance, inventory and production together — where a single mis-set posting rule shows up three departments away, two months later.
AcumaticaDynamics GPGeneric InquiriesREST / APIData conversionEDIReporting
What carries forward
Cross-functional literacy. Understanding how a work order becomes a cost, a receivable and a tax position is the
prerequisite for letting any AI near it.
ERA 05 · AI & AGENTS · CURRENT
Building agents on Claude and modern LLMs
Current, hands-on work: designing and shipping AI agents around Acumatica and manufacturing processes — tool use, retrieval, evaluation, guardrails and the operational plumbing that keeps them honest in production.
ClaudeAgent architectureMCP toolsRAGEvalsPrompt & context engineeringPython
What carries forward
Nothing about this era changes the fundamentals. An agent is a new kind of program with an old set of obligations:
be correct, be auditable, be recoverable.
“The technology changes every decade. The obligation never does: know where the number came from, know who is accountable for it, and be able to put it back the way it was.”
— Situational Business Development, operating principle
How we work
Small, fast, and paid for by the result.
No twelve-week discovery phase that produces a binder. The first engagement is deliberately short, so you find out what it is like to work together before anything big is at stake.
STEP 01
Read the situation
A working session with the people who actually do the job. We leave with the three things that cost you the most and a blunt read on which are fixable now.
Days, not weeks
STEP 02
Blueprint
A short written plan: what gets built, what it touches, who approves what, how we will know it worked, and what it costs. Yours to keep either way.
1–2 weeks
STEP 03
Build the first useful thing
One agent, one integration, one report path — running against your real data, in front of your real users, early enough to change direction.
2–6 weeks
STEP 04
Run, measure, extend
Monitoring, corrections folded back in as rules, and the next thing built only when the current one has earned its keep.
Ongoing
Engagement models
Pick the shape that fits the problem.
Start here
Situation assessment
A fixed-scope look at your systems, data and processes, ending in a written plan with priorities and costs.
Best when you know something is wrong but not what to do first.
Build
Agent sprint
One agent, scoped and delivered end to end: tooling, guardrails, evaluation, rollout and the runbook.
Best when there is one painful, repetitive job with clear rules.
Extend
ERP enhancement build
One defined piece of ERP work, delivered end to end: an extension, an integration, a reporting layer, or the automation that removes a manual step.
Best when the ERP is in place but won't do the one thing you need.
Retain
Fractional IT leadership
A senior technologist on call: roadmap, vendors, budget, escalations, and continuous improvement.
Best when you need the judgment more than the hours.
Straight answers
The questions that come up first.
Is our data safe if we let AI touch the ERP?
It is safe if it is engineered to be. Agents here run under least-privilege service accounts scoped to
exactly the data they need, default to read-only, and require an approved policy for any write. Your data is not used
to train models. Where your risk posture demands it, the whole stack can run in your own cloud tenant or on-premises.
Everything an agent reads and does is logged, so an auditor can reconstruct any decision.
We tried an AI pilot and nothing came of it. Why would this be different?
Because most pilots are scoped as demos rather than jobs. If an agent has no owner, no permissions, no
measurable target and no path into the system of record, it cannot graduate. Every engagement here starts by naming
the job, the approver and the number that has to move — and the first build runs against your real data, not a sample.
Do you replace our IT staff or work with them?
Work with them, almost always. Internal teams usually know exactly what is wrong; what they lack is
uninterrupted senior time and cover to make an unpopular recommendation. In smaller organizations with no IT
department, the fractional leadership model fills that seat directly.
We are still on Microsoft Dynamics GP. Do we have to move?
Not from me, you don't. Running ERP migrations is not what I do, and GP still runs plenty of companies
well. What I can do is make GP more useful where you are — reporting, integrations, getting data out of it — and give
you a straight read on whether it is actually the constraint. If the answer really is "move," you will get that
opinion from someone with no commission riding on it.
Will custom automation break our Acumatica upgrades?
Not if it is built with the upgrade path in mind. The preference is always published APIs, inquiries and
supported extension points over anything that has to be re-applied every release. Where a genuine customization is
the right answer, it gets documented and version-controlled so the next upgrade is a test, not an excavation.
Can you work with our existing MES, CAD, EDI provider or ISVs?
Yes — that is most of the work. If it has an API, a database, a file drop or even a scheduled report,
it can be integrated. Four decades of interfacing systems that were never meant to talk to each other is, honestly,
the through-line of this whole practice.
How do engagements start, and what do they cost?
They start with a conversation and then a short, fixed-scope assessment so both sides can judge the fit
before committing to a build. Pricing is fixed-scope for assessments and defined projects, retainer for ongoing support
and fractional leadership. You will get a number in writing before work begins — never an open meter.
How small is too small?
If you have an ERP or a manufacturing operation and a problem that costs real money every month, you are
the right size. Some of the highest-return work has been a single agent or a single integration for a company with
forty employees.
Start here
Tell me the situation.
Not a lead form that disappears into a CRM. It comes to me, and you get a reply from the person who would do the work — usually within one business day.
- Emailsecarter@sitbizdev.com
- PhoneBy request — ask in your first message and we'll set a time.
- CoverageRemote-first across North America. On-site for cutovers, plant walk-throughs and anything that needs eyes on the floor.
- ResponseWithin one business day, from a human who read it.
Useful things to mention: the ERP you run, roughly how many people use it, what you make or do, and the one process that costs you the most right now. Three sentences is plenty.