What are the Key Takeaways from this Executive Summary?
Quick answer
- A model is only as current as what it can read. Connected to your order and stock records, it answers from today. Without that, it guesses.
- Model Context Protocol (MCP): an open standard for connecting a model to tools and databases, so it can fetch live order, stock and tracking data.
- Openrouteservice and OpenStreetMap: open routing and map data you can run on your own servers, with no per-call licence.
- The thing to check: can you see which records an answer was based on? If not, you cannot audit it.
Dropshipping means coordinating parts you do not own. Several suppliers. Stock levels that move hourly. Shipping routes. Customers asking where their parcel is.
Manual processes fall behind in predictable places. Stock counts go stale. A supplier slips and nobody notices until a customer writes in. This post covers what automation reads, what it can decide, and what should stay with a person.
What does a fine-tuned model change?
Quick answer
The useful version is narrow. A model that has seen your order history, your carrier terms and your returns policy can read a situation in your own terms. A general one cannot.
Three jobs fit this well:
- Check before promising. Is the stock there, at that supplier, today?
- Notice a slip early. A supplier whose dispatch times are drifting shows up in the records before it shows up in complaints.
- Draft the message. A delay note a person approves is faster than one a person writes.
What does not fit: letting it re-route spend or change a supplier allocation without a named person approving. Those are commercial decisions.
How do the open-source pieces fit?
Quick answer
Two pieces do most of the work.
-
Model Context Protocol (MCP) is an open standard for plugging a model into tools and data sources. Think of it as one common socket instead of a custom cable per system. The model can then fetch an order, a stock level or a tracking status when it needs one. The practical gain is narrow and real: the answer comes from your records, and you can log which records it read. The practical cost is access control — a socket into your systems needs the same care as any other.
-
Openrouteservice (ORS) and OpenStreetMap cover maps and routing. ORS is open route planning software that runs on OpenStreetMap data. You can host it yourself, which means no per-call licence and no limit set by someone else’s pricing page. It handles different vehicle profiles and constraints, such as roads to avoid. It will also calculate a distance matrix, which is what you need to compare delivery options across a set of stops. Because the data and the code are open, you can add what you know: your own depot locations, your own access rules. The trade is that you maintain it, and map data quality varies by region — check yours before you rely on it.
-
Reading live records. The point of both pieces is the same. An answer about an order should come from the order record, not from a model’s memory. When it does, you can check it. When it does not, you are trusting a sentence.
On routing: a planned route is a model output, not a measurement. It is worth comparing planned against actual for a month before you let it set appointments.
How does this compare with a rules-based system?
Quick answer
Four differences matter in practice.
-
Reacting to change. If a supplier stocks out mid-morning, a fixed rule keeps promising. A system reading live stock can stop, or switch to a second supplier if you have allowed that. Who approves the switch is a policy question, and you should answer it before go-live.
-
Looking at more than one thing. Cost, distance, stock and customer location can be weighed together rather than in sequence. Useful, and it makes the result harder to explain — so keep the inputs to a decision visible.
-
Telling people. Most of the work in a delay is the telling. A drafted note to the customer, plus an alert to the warehouse, removes the part that gets forgotten. Whether it sends on its own is your call.
-
Swapping parts out. With an open standard between the model and your systems, replacing the model or adding a data source does not mean rewriting every connection. That is the argument for open standards, and it is about switching cost, not performance.
What does a day look like?
Quick answer
Walk through a morning.
Supplier A is behind on dispatch. The dispatch records show it before any customer does. The affected orders are listed by name.
Supplier B has the same items. The switch is proposed, not made, because it changes what you pay. A person approves it.
Options are compared for the orders that are now late. One courier still meets the promised date at a known cost. The drafted customer note explains the change and the new date.
The warehouse gets the same list. Nobody reconstructs the story at 4pm.
That is the whole claim for this kind of automation: the reading and the drafting happen early, and the decisions that cost money still belong to a person. Before buying any of it, check one thing in your own operation — how you currently find out that a supplier is running late, and how long that takes.
