Say a thirty-person haulage and warehousing firm in Gebze receives shipment requests by email all day long. One customer attaches a PDF order form, another an Excel sheet, a third just writes “two pallets to Manisa tomorrow morning”. Two dispatchers read every message and type pickup address, delivery point, pallet count and date into the transport management system. On Mondays the queue is still there at noon, and several times a month a truck heads to the wrong district. What this firm needs is not “an AI”. It needs a filled-in record, placed next to the original request, that a dispatcher can check and approve in seconds. That is the target of this service. We pick one process and put numbers on how it performs today, then try models on your real emails and documents. Next the model runs in the shadow, alongside the dispatchers, and only once the results hold up do we wire it into your existing software. From day one it is written down which contract covers customer data and in which country it is processed. Meetings take place by video and the set-up is done over remote access.
Integration is much more than choosing a model. These pieces make up the body of the project, and how much weight each carries depends on the process.
Who receives the request, which fields get read, where it waits, which step produces most mistakes? We sketch the flow step by step with the people who do the work and note how many minutes each step takes now. Any gain is later calculated against that baseline.
The model pulls fields out of the email or attachment, then your own rules take over: is the district on the official list, is the pallet count plausible, does the customer code exist in your accounts? Any field that breaks a rule is flagged. Simple checks like these catch most model slips.
Business-grade services that large providers run in EU regions are tested side by side with open models that could run in your own account, on the same examples. We look beyond accuracy at cost per case and at where the data ends up.
The dispatcher sees the original request and the record the model filled in, next to each other. Uncertain fields are colour-coded and fixed with one click. Corrections are not lost; they flow back into the measurement data.
Results land where the team already works: a draft trip in the transport system, a pre-entry in Logo or Mikro, a reply draft in Outlook or a tagged ticket in the helpdesk.
A processor agreement with the provider, the country where data is processed and the terms for transfer abroad, plus updates to your personal data inventory and, if required, your privacy notice. Your lawyer or KVKK adviser gives the legal view; we supply complete technical facts.
A brief monthly report covering accuracy, corrected fields and cost per case. When a customer changes their document layout or the provider ships a new model version, a dip in quality does not slip past unnoticed.
The calendar depends on the process and on the state of your data. After each phase we review the figures together and you choose whether to go further.
For two weeks we log the number of requests, handling time per request and the kinds of errors. The target is written from those figures.
Several models and settings are tried on past requests; you get a table of accuracy, expected time saved and cost.
The model processes live requests in parallel while dispatchers carry on as before. Results are compared without touching a single record.
One customer group or request type first, then the rest. The review screen stays mandatory until quality has been demonstrated.
Put a number on success before you begin. “Take some load off the dispatchers” is a hope. “Three in four requests turn into bookings with no edits, and the rest cost a dispatcher under a minute each” is a target. If nobody fixes the figure in advance, each person reads the trial differently at the end, and the project either never finishes or dies quietly. Agree the number with management, and accept stopping if the prototype falls short.
In our experience the job does not vanish, it changes. Instead of typing, people check, handle exceptions and talk to customers. Bringing the team in early, especially when mapping the process and designing the review screen, makes acceptance far easier.
It hinges on the contract behind the service. Most business-grade offerings exclude customer inputs from training by default; free personal accounts can work differently. We go through the terms with you before any work starts and put our reading in writing.
We avoid locking the connection to any single vendor. Because you keep the measurement set, a new model can be tested on the same examples within days, and the switch decision rests on figures.
Baseline and prototype usually wrap up within a few weeks, and the prototype table already tells you whether carrying on is worthwhile. The shadow run and rollout take longer or shorter depending on volume and on the system being connected.
Either as a project with a defined scope or at €55 per hour plus VAT with an estimate first. Model usage fees are forecast per case and per month before we start, and you pay them straight to the provider. Eligibility for digitalisation support from KOSGEB, TÜBİTAK or a regional development agency is something to confirm with that institution; we neither apply on your behalf nor promise any grant.
Describe the task, how many times a month it comes up and which software it runs in today. Expect a first assessment from us within one working day.
We have your enquiry
A reply will reach you by the next working day at the latest. If your message says work has come to a halt, it goes to the top of the pile.
No such city in our list. Try another spelling, or just pick the closest big city: we work entirely over remote connections, so nothing about the service changes from one province to the next.
The only cookies here are the essential ones: they keep the site running and remember the city you picked. Nothing is used for advertising or tracking. See our privacy notice for more.
