Make-to-order manufacturing

Give your customer a delivery date you can defend.

I calculate lead times from the history of your orders in ERP and MES. Sales can quote a date without calling the shop floor, and you know what it is based on.

Book 30 minutes about your lead times

Free. I will check whether your data is enough to calculate the date.

See how it works in practice
Computer visualisation, example data

Order 0412 · received 24 Nov

When will we ship?

15 DecemberP80

P80: 8 of 10 finished within 21 days

Lead time of 10 similar past orders, in days

Bottleneck: work centre C, 4 days of queue

Illustration on sample data

Łukasz Kamiński, production engineer since 2005, systems integrator since 2021. I am preparing a doctoral thesis on how small and medium-sized companies adopt AI.

Where a delivery date goes wrong

The date a sales rep gives is only as good as what stands behind it. This is where I started at Kocher+Beck Polska.

The forecast in the source system ignored queues. It differed from the actual routings, because it did not take the queue in front of each work centre into account. The sales rep had no date to rely on.

One figure, several definitions. On-time delivery calculated in three ways gave three different results: leaving out overdue orders that were never shipped, counting the first issue instead of the closing one, and measuring against a postponed date. Without a written methodology, each of those numbers could be defended and challenged.

A figure that cannot be traced to a document. When a sales figure on the dashboard is challenged and cannot be traced to documents, nobody can settle which version to believe. The argument about the number replaces the decision.

See how the date is calculated →

How the date is calculated

The panel does not trust the forecast from the source system. It takes the date from the lead-time history of similar orders and adds the queue in front of the work centres.

1. Read the history, lock nothing

Orders and routings come from the MES, read without locks. Invoices come from SAP Business One, read only.

2. P80 from similar orders

The date follows P80 of similar orders: the time within which 8 of 10 similar orders from the history were completed.

3. Add the queue in front of each work centre

The bottleneck is the work centre with the longest wait, not the sum of the operation times.

4. Count unfinished orders too

Orders still in progress are treated as incomplete observations (the Kaplan–Meier method). Without that, only finished orders would count and the date would come out too optimistic. A B2B customer sees only “+X days”; the sales rep sees the breakdown.

A production and deadline panel at Kocher+Beck Polska

A Polish manufacturing company from the Kocher+Beck group, making to order. I designed and built the panel myself. It reads data from the MES and from SAP. It writes nothing to SAP, and only the sales module writes to the MES, through my own procedures in a separate database schema.

Deployment status

Business panel
The work is completed and the panel is in use at the company. The first modules have run on the server since September 2026.
Autodesk Inventor add-in
Completed. From order parameters it creates 3D models and 2D documentation for the design department.
Apps for HAUSBUD
In use: a quoting app and, since 31 October 2026, a construction-site management app (one user; page in Polish). A renovation company, a different industry.
Not there yet
A before-and-after measurement of the effect. That is why I quote no savings and no improvement in on-time delivery.
  • Delivery date at 80% confidence builtThe date comes from the lead-time history of similar orders: the time within which 8 of 10 similar orders from the history were completed. This is a method, not a measured accuracy. Orders in progress count as incomplete observations, and the bottleneck is the work centre with the longest wait, not the sum of times. A B2B customer sees only “+X days”.
  • Shop-floor queue and alerts builtThe order of jobs, stalled orders, a list of orders that can still wait, and a “stage has slowed down” alert. The main goal: fewer reschedules.
  • On-time delivery in two variants builtMeasured against the original promise and after rescheduling, with a written method, so that the figure can be defended in a meeting.
  • Source systems unchangedI read the MES without locks and write only through my own procedures. For SAP I use read access only. The interface works in Polish, German and English.

Read the case study Book 30 minutes about your lead times →

The summary is in English. The full case study is in Polish.

  • over 1,700automated tests of the panel (September 2026)
  • Same as SAPthe sales report by customer matches SAP to the grosz in six report settings, checked in October 2026
  • 3 languagesof the interface: Polish, German and English

Source: the case study. The tests were counted in September 2026 and the SAP comparison was checked in October 2026. These figures describe the software, not a business result. Not claimed: any before-and-after effect. I have no measurement of the time to prepare a management report, of the number of questions to sales about a date, or of on-time delivery. I propose starting with the first two, with the method agreed with the company. No screenshots and no figures from the company's data.

Your systems stay, and the panel must not break what already works

I replace nothing. I read data from your systems and write only through agreed procedures. The limits of each system are set before the first connection, and every change is tried on a copy first.

  • SAP Business One read access on the company's server Read through Service Layer, data only through GET requests (apart from the session login): invoices, notes and corrections in the Polish localisation, and advance payments. The sales report by customer in the panel matches SAP to the grosz in six report settings, checked in October 2026.
  • MES system (SQL database) on the company's server SQL read without locks, with short queries. Writes only through my own procedures; the tables of the vendor's system stay untouched.
  • Autodesk Inventor completed An add-in for the design department. From the order parameters it generates 3D models and a full set of 2D drawings in the department's standard, using the frame and title block from your company's drawings. The whole set updates with one click.
  • SAP S/4HANA, ECC and other ERP systems offer, no deployments yet I have no deployment on these that I could show. For any other system I first check how it exposes data: through an API, views, a file import or a procedure. Until I have checked, I promise no integration. I start with a proof of concept on a copy or in a test environment.
Read without locks
Short queries of limited scope, a time limit, and heavy diagnostic queries outside working hours. The connection pool is limited and sessions carry the application's name, so they are visible in the server's monitoring.
Write only through agreed procedures
Documents reach the MES only through my own procedures in a separate schema. I do not change the vendor system's tables. Every data change goes in as a small batch, first on a test copy and after explicit approval.
SAP: read only
I send SAP only read requests, with a shared limit on parallel requests, so that the panel does not load the financial system.
Permissions and isolation
Each role sees its own screens. A B2B customer sees only its own data: the filter comes from the customer's identity in a signed cookie, and automated tests check that queries for document lines always filter by customer. Passwords are stored as salted hashes, and logins have an attempt limit and a log.
AI only on aggregates
The model has no access to the database and receives aggregated figures. The assistant window warns users not to type in customer names, order numbers or personal data, but it does not filter what they type. AI functions are off in the base configuration.
Where the data is
The panel runs on the company's server. The interface loads no external scripts or content delivery networks and works on a network without internet. AI functions need a model provider. I agree where the data is processed in the data processing agreement before the start, and I do not promise a region in advance.

SAP and SAP Business One are trademarks or registered trademarks of SAP SE or its affiliates in Germany and other countries. Autodesk and Inventor are trademarks or registered trademarks of Autodesk, Inc. and/or its subsidiaries or affiliates. I am not a partner, certified contractor or authorised representative of SAP or Autodesk. I name the systems only descriptively, to show what I connect data to.

Who it is for: where the queue in front of a work centre decides the date

This English page is for foreign groups that run plants in Poland. An example is Kocher+Beck Polska, a Polish manufacturing company from the German Kocher+Beck group, where I built the panel. I answer enquiries in English. The case study and the legal documents are in Polish.

Outside Kocher+Beck Polska and HAUSBUD I do not name clients. Below are the types of make-to-order manufacturers in which the problem recurs and in which this approach makes sense.

Where I would start, and with what

  • Packaging and labels. Short runs in many variants, with a date that depends on the queue at several work centres. Start: a date from lead-time history.
  • CNC machining. One-off and small-batch jobs share the same machines, so the date follows the queue in front of the machine, not the sum of operation times. Start: the shop-floor queue and stalled orders.
  • Steel structures. Every order needs shop drawings and a full set of documentation. Where the design can be described by parameters and rules, the documentation can be generated from them. Start: checking whether the design can be described by parameters. I have the Inventor add-in for one design department; for other industries it is an offer.
  • Tooling and moulds. Every order is a different design and a long path through several departments. The question about the date goes to the production manager, because the sales rep cannot justify it. Start: a date with confidence and a status view for the customer.
  • Plastics and injection moulding. A queue of machines, changeovers, and dates promised to sales. When the plan changes, on-time delivery has to be measured against the original promise and after rescheduling. Start: on-time delivery in two variants.
  • Made-to-order joinery and furniture. Orders with the customer's dimensions arrive by email and in spreadsheets, and production runs through several stations. Start: the order agent and the delivery date.

When it is not worth starting

  • If no system records orders and routings, the panel has nothing to calculate from. I then start by putting the data in order.
  • If the date always follows from one operation, a simpler spreadsheet is enough.

Four steps, always starting from one process

01 · Call and diagnosis

First a free 30-minute call: we settle on one number you cannot defend today, or a date nobody can justify. If the data is enough, the next step is the diagnosis: 2 weeks, from 4,900 PLN net plus VAT, credited against the implementation fee. In it I check what access to MES, ERP or SAP is possible.

You get: the scope of the pilot and a measure to compare before and after.

02 · Pilot on one process

I build the first view with read access only, on a test copy or with read-only permissions. Nothing is written to your systems.

You get: a working version to check with your people.

03 · Implementation

I release in small phases, each after the agreement of your IT and the process owner: accounts, server, permissions. Writing to systems only through agreed procedures, first on a copy.

You get: a version on the company's server and a release checklist.

04 · Development

Short iterations, reviews with the management and a request log. I compare the result with the starting point and decide with you on the next module.

You get: a decision on the next stage. Support after go-live is scoped separately.

You see scope and price before you start.

Before the start you get a description of the work and a schedule, and you decide on the implementation separately.

  • The first conversation, 30 minutes: free.
  • Diagnosis: 2 weeks, from 4,900 PLN net plus VAT, credited against the implementation fee.
  • The first module of the panel: from 12,000 PLN net plus VAT.
  • A separate integration project: indicatively from 12,000 PLN net plus VAT.

All prices are net amounts in Polish zloty (PLN), and VAT is added on top. The order agent pilot in shadow mode runs for 30 days free of charge.

Order agent: a module in a shadow pilot

No deployments · looking for the first pilot companies

The agent reads an order from an email, a spreadsheet or a PDF and prepares a document for a sales rep to approve. It reaches your ERP only after approval, and the lines it cannot identify go to a separate queue instead of being guessed.

The pilot runs in shadow mode for 30 days, with no write to the ERP and no setup fee. At the end you get a report built from your own orders. The agent has no deployment yet, which is why it starts in shadow mode.

Work out what manual order entry costs you → Ask about the pilot See the demo (Polish)

Why me

I have been a production engineer since 2005: I worked in manufacturing, planning and logistics. Since 2021 I have been a systems integrator. I hold an Executive MBA. On the call you talk to one person who knows the shop floor and writes the code himself. I designed and built the Kocher+Beck Polska panel myself.

I am preparing a doctoral thesis in management and quality sciences on how small and medium-sized companies adopt AI, and how to tell that an implementation delivered something. For you this means that on every project I first agree with you what to measure before and after.

Beyond Kocher+Beck Polska, HAUSBUD (a renovation company, a different industry; page in Polish) uses my quoting app and, since 31 October 2026, a construction-site management app (one user). The reduction of a quote from about 30 to about 5 minutes is a design assumption, not a measurement; 82 automated tests check the calculation logic and do not measure quoting time.

Profile and experience on LinkedIn ↗ · Book 30 minutes about your lead times →

Questions production managers ask about systems, data and responsibility

I settle these at the start, before the first view is built.

Where do we start, and what does it cost?
With a free 30-minute call about one process, for example a delivery date you cannot justify today. If your data is enough, the first paid step is the diagnosis: 2 weeks, from 4,900 PLN net plus VAT, credited against the implementation fee. The first module of the panel costs from 12,000 PLN net plus VAT. Before the start you get a description of the work and a schedule, and you decide on the implementation separately. The order agent pilot in shadow mode runs for 30 days free of charge.
Do I have to replace my MES, ERP or SAP?
No. The panel stands next to the systems that already work and reads data from them. Writing to the vendor's system goes only through agreed procedures, and I do not change its tables.
Which systems do you work with?
I have worked with read access to SAP Business One through Service Layer, with read access to the database of an MES system, and with an add-in for Autodesk Inventor. For other systems I first check how they expose data: through an API, views, a file import or a procedure. Until I have checked, I promise no integration.
How do you calculate the delivery date?
The panel does not trust the forecast of the source system. It calculates the date from the lead-time history of similar orders using P80, the time within which 8 of 10 similar orders from the history were completed, and it adds the queue at the work centres. Orders that are not yet finished count as incomplete observations (the Kaplan–Meier method), so the date does not come out too optimistic. This module is completed in the Kocher+Beck Polska panel, but I do not yet have a measurement of how much it improves delivery dates.
What results have you measured so far?
None. At Kocher+Beck Polska I have not yet made a before-and-after measurement, so I give no savings and no improvement in on-time delivery. The work there is completed and the business panel is in use at the company, but that is not a measurement. I propose starting with two measures: the time to prepare one management report, and the number of questions to sales about a date. The method has to be agreed with the company.
Are the panel and the AI safe for our data?
The panel reads data without locks and writes only through procedures. The report commentary and the AI assistant get aggregated figures from the tools, with no personal data and no customer names; text a user types in goes to the model provider, and the window only warns about it. The assistant is off in the base configuration, and whether to turn it on, and with which model provider, we agree before the start. For AI functions, the processing region is agreed in the data processing agreement before the start; I do not promise one in advance.
Do you have deployments at clients?
At Kocher+Beck Polska the work is completed and the business panel is in use at the company; the first modules have run on the server since September 2026. There is no before-and-after measurement yet. The Autodesk Inventor add-in is completed. Housebud Wyceny is an app used by the company HAUSBUD; the reduction of a quote from about 30 to about 5 minutes is a design assumption, not a measurement; 82 automated tests check the calculation logic. The order agent has no deployment yet: I am looking for the first companies to pilot it in shadow mode.
Who checks the output of the AI?
I base critical calculations on explicit rules. AI can help read a document or find information, but it can also make mistakes. In each process I set the validation, the handling of exceptions, and the points where a person approves an offer, a date or a write to a system.
What does maintenance after implementation cover?
Before the start I agree the scope of monitoring, how failures are reported, and who is responsible for changes on the integration side. Care and further development have a separate scope and cost, depending on the number of processes and their importance to the company.
What if it does not pay off?
That is why I start from the cost of the current process and a small scope. The ROI model is an estimate, and a pilot lets you check the assumptions. If the result does not justify further work, I recommend stopping or changing the scope.
Are you a partner of SAP or Autodesk?
No. I am not a partner, certified contractor or authorised representative of SAP or Autodesk. I name SAP Business One and Autodesk Inventor descriptively, to show what I connect data to. SAP, Autodesk and their products are trademarks of their respective owners.

Book 30 minutes about your lead times

If you prefer, pick a slot in my calendar (Google Meet, Monday to Friday, 8:00 to 16:00 Polish time). Or write what you calculate by hand today, or what you cannot promise a customer: I usually reply within 24 hours on working days and agree a time for a free 30-minute call.

I usually reply within 24 hours on working days. You do not need a ready specification. Prefer to talk now? +48 667 206 420