MANUFACTURING · MARGIN, BOOKED AGAINST KEPT
Rebuild margin after freight, rebates and returns
The ERP margin report is accurate and it stops at gross margin. Freight sits in a carrier bill, rebates accrue in a spreadsheet and returns post to a ledger account, so none of the three ever reaches the customer who caused them. OutcomeCatalyst allocates all of it back and ranks customers on what you keep rather than what you book.
Margin intelligence, in plain terms
A manufacturer with three plants runs three ERPs, and the same physical part carries a different part number in each. That single fact is why nobody can answer what a part costs, what a customer is really worth, or which line is holding back the whole plant. Margin intelligence joins them and answers all three.
From three ERPs to one number, in four steps
No migration, no consolidation project, and no change to how any plant runs day to day.
Why this matters in mid-market manufacturing
Multi-plant manufacturers are usually the product of acquisition, and each acquisition arrived with its own ERP, its own part numbering and its own supplier relationships. Consolidating those systems is a multi-year project that rarely finishes and never pays back on the schedule promised, so most operators live with the fragmentation permanently.
The cost of living with it is specific. Procurement negotiates plant by plant against suppliers who see the whole account, which means the supplier has better information than the buyer. Two of your own plants bidding separately for the same part is not a rare edge case, it is the default outcome of the structure.
Margin reporting has the same blind spot in a different place. Gross margin is calculated at the invoice, but freight, rebates, returns and expedites land afterwards in different ledgers. A customer can be your largest by revenue and your worst by contribution, and the standard margin report will never show it.
On the floor, capacity is misunderstood in a way that costs capital. Plants buy another line because throughput is short, when the constraint is a single cell running at under half its potential. The downtime reasons that would prove it are written in shift notes rather than captured as data.
Agentic AI in manufacturing, without the hand-waving
Three words get used interchangeably by vendors and they are not the same thing. The difference decides whether you get a dashboard or a decision.
Most agentic AI pilots on a plant floor fail for a reason that has nothing to do with the model. An agent asked what a part really costs needs to know that 4471-B in Ohio, 4471B in Texas and SKU 88213 at the contract manufacturer are the same item. Three ERPs, three numbering schemes, no shared key. The agent has no path to walk, so it guesses, and a confident guess on a consolidation decision is expensive.
The agents here are deliberately narrow. Each has one job, a defined set of sources it may read, a written standard to check against, and a person who approves before anything leaves the building. That is what makes them safe to run against real money, and it is why they survive an audit.
The data layer agentic AI actually needs
Every workflow above runs on one layer. Building it is most of the work, and it is the part nobody demos.
This is the part most vendors skip, because a layer does not demo well. It is also the reason one piece of infrastructure supports SKU margin, direct spend and working capital at once, instead of three separate tools each rebuilding the same context badly.
It also compounds. The second workflow is faster to stand up than the first and the fifth is faster still, because the entities, the ontology and the connectors already exist. Most of what a new workflow needs is already sitting in the layer.
What else runs on the same layer
Once the layer exists these are weeks of work rather than months, because they read the same resolved entities.
Built for the stack a plant actually runs
These are the systems referenced in the workflow above. Anything with an API, a database or an export can be connected, including controls and historians on the floor.
Questions plant owners and operators ask first
Do we have to consolidate our ERPs first?
No, and that is the point. Consolidation is the project most manufacturers cannot finish. This reads each plant where it stands and reconciles parts and costs on top, so you get the joined view without the migration.
How is this different from our BI or margin reports?
A margin report is accurate about what the invoice said. It cannot include the freight bill that arrived two weeks later in a different ledger, or the rebate accrued at quarter end, or the credit memo for a return. Those are what separate booked margin from kept margin.
How do you match parts across plants without a common code?
By specification rather than by code. Descriptions, dimensions, supplier part numbers and unit-of-measure are reconciled, and low-confidence matches are surfaced for a human to confirm rather than assumed. Your engineers see the match before it is used.
Will this disrupt production or touch the controls?
No. Machine data is read from the historian or MES, not written to. Nothing on the floor is controlled, adjusted or interrupted, and the plant runs exactly as it did before.
Our third plant runs a system from the 1990s. Is that a problem?
It is common and usually workable. Older systems almost always expose a database, a scheduled report or a flat-file export, and any of those is enough. It is often the plant with the most trapped value precisely because nobody has looked.
How long before we see a number we trust?
Most engagements are live on a first workflow in four to six weeks. Direct-spend price spreads are the usual starting point because they are verifiable against invoices you already have, which makes the result easy to check independently.
Who owns the decision to reprice or consolidate?
Your team, always. The system produces the comparison, the draft supplier request and the supporting documents. Procurement and commercial leadership decide what to send and what to accept.
Does this replace our ERP or our planning system?
No. It reads from them and writes findings back, so buyers, planners and controllers keep working in the systems they know. There is no new interface to adopt and no data leaves your control if you deploy in your own environment.
AI customer margin and cost to serve: common questions
What is cost to serve?
Everything a customer costs you that does not appear in cost of goods. Freight driven by how they order, rebates earned against contract tiers, returns and allowances, and the handling that goes with small frequent drops. Allocated properly, it routinely moves an account by ten points or more.
Where does the data come from?
The ERP for revenue and cost of goods, carrier bills for freight at shipment level, the rebate accrual workbook, and credit memos for returns. The work is allocating each of those back to the customer that caused it, which is exactly the step no standard margin report performs.
What do you do with a customer who loses money?
Usually not fire them. The loss is normally caused by order profile and returns rather than by price, and a minimum drop size or a tightened returns policy can move an account to positive without touching the headline rate. The point of allocating it properly is that the conversation can be about how they order instead of about a price increase.
