ButterflyCopilotButterflyCopilot
← Blog

How fetra Automated Order Entry with AI Agents

The order was already in the inbox, in more variants than the desk could key when a peak hit. A parental leave made it urgent. Order entry then took 80 percent less time, and the inside-sales roster stayed as it was.

ButterflyCopilot

Dark title card with a small line drawing of a platform trolley. A small line reads order entry, variants, then 80 percent in large type. The line under it says time saved in order entry. The parental-leave seat was not restaffed. No new inside-sales hire. Four rows: the seat, a calculated savings potential of more than 50,000 euro a year because the seat stayed empty; the quarter, more than 40 working days saved in the second quarter of 2026; the peaks, apprentices stay in their own departments and the inbox still gets entered; the variants, color, tyres, a modification, a complete set, matched to master data and then into Navision. A stamp reads reported for fetra, not your forecast. A line under the stamp: trolleys and carts, Borgholzhausen, more than 1,000 products, no published headcount. Source line names Workist, 11 November 2026. The 80 percent is time saved in order entry. The euro figure is a calculated potential.

0106

The order had already arrived. Someone was still choosing the variant, and then typing it.

fetra Fechtel Transportgeräte builds trolleys, platform carts, and hand carts in Borgholzhausen. Steel, aluminum, stainless steel. A family company in the third generation, manufacturing in Germany. The range runs past 1,000 products, from an office trolley to running gear rated for 24 tonnes, and the carts sell across Europe. Light trolleys, transport carts, special carts, hand carts, wheels and castors, lifting equipment. A cart that can be had in that many materials is not one product. It is a color, a set of tyres, a modification, sometimes a complete set. The catalogue is also an inbox.

Orders come in by email, every day, in the layout that customer used this month. For years, inside sales read the mail and keyed the header and the lines into Microsoft Dynamics Navision. It worked while the team was whole. It tightened as soon as someone was out, or as soon as the volume rose. Several times a year a peak arrived and the desk could not keep up. The answer, for years, was to pull apprentices off other departments. They left their own work. Sales trained people who only knew the process on the surface. The error rate did not thank them for it. A mistyped item number and a variant mixed up under time pressure were the shape of the mistake.

The hole became urgent when a long-standing inside-sales colleague went on parental leave. Years of knowing which cart was which went with her. The ordinary next step was a new hire and a long teaching-in. fetra did not take it.

The mail goes to the agent

They put Workist's order agent on the inbox. The sales team was in the room early, so the worries were on the table before the switch, not after it. From then on, an incoming order goes straight to the agent. The agent reads the header and the line items. It matches the customer and the product against master data. It transfers the finished order into Navision.

An order email lands
        ↓
The agent reads the header and the lines
        ↓
The customer and the variant are matched to master data
        ↓
A clean, exact order can become an unreleased draft
        ↓
A near miss, a tight date, or a shaky field waits for a person
        ↓
A peak is a batch, not a reason to borrow an apprentice

The boxes are the instruction file. They follow the shape the story describes. They are not a reconstruction of fetra's screens, and they are not Workist's project plan.

The published account of the result is specific, and the sentences are not the same kind of sentence. Order entry took 80 percent less time. In the second quarter of 2026 the company saved more than 40 working days. No new inside-sales hire. Apprentices were no longer pulled off their own departments, and the peaks were handled. The parental-leave seat was not restaffed, and that is where the calculated savings potential of more than €50,000 a year comes from. Calculated potential. Not cash counted out of a drawer.

The error rate in order entry dropped. The old mistakes were a mistyped item number and a variant mixed up when the desk was rushed. The account of the new desk is that the transfer into Navision no longer makes them.

Jan Völlering, Head of Marketing, put the staffing part in one place.

Thanks to Workist, we were able to keep our staffing structure unchanged while the business grew. That alone made the solution worthwhile for us. On top of that, our error rate in order entry is significantly lower, and we can scale our processes efficiently as order volumes rise.

And the part about what the people were for:

For us, it was never about replacing people. It was about freeing them from work nobody enjoys, so they can focus on the things that really make a difference for our customers.

The same point, shorter, from the same person: "Our employees can now focus again on the things that really make a difference for our customers."

A tidy variant is not an order

Speed is how a confident agent creates a sales order because someone else's year hit 80 percent. Speed is how "blue" becomes the nearest blue in the table when the customer wrote a word the table does not contain. Speed is how a peak is handled by pulling the same apprentices, only faster, because the queue got deep and a person wanted it gone.

The file below keeps the shape and drops the plant. One mailbox. The customers, the ship-tos, the colors, the tyres, the modifications, the allowed combinations, and the sets are the company's. The model may copy a string the document already printed. It may not translate it, and it may not choose the customer, the variant, or the create. A combination that is not on the table is an exception, not a nearest color. A fuzzy match is shown to the reviewer with the candidate filled in, and it does not skip them, including after the tests pass. A field the model is not sure of is a blank. A field it is almost sure of still waits. A date inside the tyre's lead time waits, because the date is real and the promise might not be. A whole order waits if one line is wrong.

On the fixture, which is ours, a made-up platform trolley in blue, on a grey tyre, with no modification, becomes a draft at the list price. The same trolley with a pneumatic tyre is not a row, and it waits. A doubled letter in the customer's name is a review, not a silent match and not a rejection. A set expands into the two components the table already lists, at the sum of their list prices. A pair becomes two pieces when the pair price matches. A price that does not match waits, and the list does not move. A quantity the model is not sure of is thrown away, and the exception does not contain the digit. Those rows exist so a builder can test the combination and the refusals. They are not a fetra catalogue, and they are not an order anyone should ship.

Avant Tecno is the other customer-order desk on this site. That one matches a finished item code, keeps an expert on every create, and has a second module for supplier confirmations. This desk does not confirm a purchase order and does not write to a supplier. Its hard problem is the variant. Its busy day is the peak.

The useful caution is the table. A result like the one in the story sits on master data that already knows which color, which tyre, and which modification are a real product. An empty variant table will not get that result in the first month. The file spends that month getting the combinations out of the company's head and onto a sheet a rule can refuse.

What a smaller manufacturer should steal

If orders still arrive as email, and the catalogue is really a set of choices, the first month is one mailbox and the combinations you are willing to get wrong in private.

The useful part of fetra's story is the split, not the clock. Let the model read the mail. Let the table decide the variant. Keep a person on anything that is not an exact match, including after the tests pass. Give a peak a batch size and a holding file, and refuse the old answer, which was to borrow apprentices until the pile was gone. Write the ERP an unreleased draft, in Navision if that is the one, in Business Central if the company has moved, or as a file if the ERP can wait. A second mailbox waits.

One mailbox
        ↓
Thirty days of proposals, the variant table, and a stopwatch on the orders a person already keys
        ↓
Thirty days where the reviewer works the queue, and nothing is created
        ↓
Thirty days where a signed draft may be written, unreleased
        ↓
Exact orders can skip the token only after a run of drafts nobody had to edit
        ↓
A fuzzy variant stays on the queue

Time the stretch you already have. Minutes from the mail in hand to a draft a person will sign. How the pile splits across an exact ready order, a fuzzy customer, a fuzzy variant, an invalid combination, a price miss, a date inside the lead time, a field the model was not sure of, and a duplicate purchase-order number. How many ready drafts the reviewer accepts with no edit. How many review candidates were the wrong variant. How many of those drafts you later reverse. How many orders this agent released or posted. How many apprentices it pulled. Those last two stay at zero.

fetra's scoreboard is not your forecast. 80 percent of the order-entry time, more than 40 working days in one quarter, a calculated savings potential of more than €50,000 a year, no new inside-sales hire, and peaks handled without borrowed apprentices are this company's published story. The error-free transfer is the account on that page, not a measured zero you can paste over. No headcount is published with the story. Do not invent one. Measure your own minutes. Do not paste their 80 into a confidence threshold, and do not put €50,000 on a proposal as a promise.

What the instruction file will make an agent do

The file below is the brief. It is not Workist, and it is not fetra's Navision company. None of that schema is public. The blanks are yours. The same draft works as a JSON file a reviewer can open, and later as an unreleased sales order in Navision or Business Central, using the payload in the file.

It will make the agent:

  • Refuse to run while a REPLACE line is empty, while an EXAMPLE row is still in a live folder, while an apprentice is on the roster, or while anyone has typed 80 percent, €50,000, or 40 days into the config.
  • Take one mailbox. Read a proposal that copies the document. A confidence below the line the company typed blanks the field. On the fixture, a quantity of 9 at 0.40 is quantity_missing, and the digit is gone.
  • Match the customer by account id, or by the whole legal name, or by the whole alias, and only then by a string score with a gap to the next customer. North does not match North Depot GmbH. A near miss is a review, with the candidate shown.
  • Resolve a line to one item number from the allowed combination of family, color, tyre, and modification. A code that is present and wrong is not rescued by the description. A missing combination is variant_invalid. A comma in the modification is not two modifications.
  • Expand a complete set only from the set table. Price the line from the list, including a pair that converts to two pieces and a set that must equal the sum of its components. On the fixture, 189.00 stays 18900 cents. 199.00 waits. One pair at 378.00 becomes two pieces at the list price. One set at 285.00 becomes the two components.
  • Hold the whole order when any line fails. A date inside the tyre's lead time is a review, not a silent promise. A credit hold does not become a draft the reviewer can wave through.
  • Leave the ERP unwritten until the gates pass. After that, a reviewer token can write an unreleased draft. An exact order can skip the token only when a later switch is on. A fuzzy order never can. Nothing in the file releases or posts.
  • Drain a peak in one batch and stop. Further doubtful orders wait in a holding file. pull_apprentices cannot be turned on.
  • Write a morning file from this company's own counts: ready, review, exceptions, touchless writes, reversals, minutes. No comparison with the case study. No percent.

The tests are in the file. The hashes of the ready draft, the set, the pair, the price exception, the discarded 9, the fuzzy customer, the morning file, the audit, the payload, and the reviewer's approval are locked. An ERP file on that run fails the test. If the agent clears any of those by inventing a variant, releasing an order, emailing anyone, or staffing a peak with an apprentice, the test fails.

The instruction file

The brief has the role, the variant tables a spreadsheet can hold, the proposal schema, the extraction prompt, the confidence rule, the fuzzy match, the allowed combination, the set expansion, the price and the unit, the lead time, the review queue, the peak batch, the reviewer's token, the Navision and Business Central payload, the audit, the three metrics, the ninety days, the cost caps, and the tests that decide when a file may be treated as a draft the ERP has been offered.

Replace every REPLACE line. Delete every EXAMPLE row. Let it sit beside one mailbox for a month a person still keys by hand, before anyone treats a file as an order the warehouse has already been told to pick.

Sources

The company, the agent, the quotes, and the figures are from the pages below. The address, the third-generation family company, and the catalogue categories are from fetra's own site. The figures are as those pages report them. They are not an independent audit, and they are not a forecast for another manufacturer. No headcount is published on these pages. The euro figure is a calculated savings potential from not restaffing one seat. The line about a transfer without errors is the success story's account of the new desk.

  • Workist, "How fetra Automates Order Entry with Workist," Markus Bimüller, 11 November 2026. fetra Fechtel Transportgeräte. Trolleys, platform carts, and hand carts in steel, aluminum, and stainless steel, ordered in countless variants: color, tyres, a modification, a complete set. Orders by email, keyed into Microsoft Dynamics Navision by hand for years. Peaks several times a year. Apprentices pulled from other departments and trained for a process they only knew superficially, which did little for the error rate. A long-standing inside-sales employee on parental leave. The sales team involved early. The agent reads header and line-item data, matches customers and products, and transfers the order into Navision. Calculated savings potential of more than €50,000 a year because the seat was not restaffed. 80 percent time saved in order entry. More than 40 working days saved in the second quarter of 2026. No new inside-sales hire. Apprentices no longer pulled. Peaks handled. The error rate significantly lower, and the page's account that the agent now transfers the data without errors. More than 1,000 products, from office trolleys to running gear rated for 24 tonnes, sold throughout Europe. Jan Völlering, Head of Marketing, fetra Transportgeräte, in the two quotations above, and: "Our employees can now focus again on the things that really make a difference for our customers."
  • Workist, "Wie fetra die Auftragserfassung mit Workist automatisiert," the same story in German, same date, same figures, same quotations. The colleague on parental leave is given in the feminine. This page uses that, and it does not add a fact the German page does not share with the English one.
  • fetra, fetra.de. fetra Fechtel Transportgeräte GmbH, Industriestraße 17-21, 33829 Borgholzhausen. A family company in the third generation, manufacturing in Germany. Catalogue categories include light trolleys and rollers, transport carts, special carts, hand carts, wheels and castors, and lifting equipment. This page uses the site for the address, the family company, and the categories. It does not use the site for a headcount, a revenue, or the agent results, and it does not use the prices printed on the homepage.

Stay in the loop

Product updates and enterprise AI notes. No spam.