Chances are AI is already at work in your company, just off the books. A sales rep has quotation emails written by a chatbot signed into with a private Gmail address; someone in accounts drops a bank statement into the same kind of tool and asks for the highlights. Whether the technology impresses is no longer the point. What matters are three down-to-earth questions: which chores does it genuinely take off your team, what is the monthly bill for that, and where do the documents you hand it actually travel? Apply helps SMEs, accounting and law practices, clinics, hotels and online sellers in Turkey answer those questions with evidence. We pick the tasks worth automating, trial them in a small pilot, run models and documents in a controlled environment and teach staff to treat AI output with healthy scepticism. Nobody has to travel; the whole engagement runs on screen-sharing sessions and secured remote connections.
Join at whichever step suits you. If you are starting from nothing, though, this sequence wastes the least money: establish what is worth doing, prove it on real work, build a safe place for it to run, and finally train the people who will use it daily.
Groundwork for a board-level decision: an honest inventory of the AI services employees have quietly adopted (the list tends to surprise directors), which ideas would pay for themselves and where KVKK draws the line. The result is a staged roadmap you can approve as it stands or trim.
One work step is automated, with timings taken before and after. The AI lives where your team already spends its day rather than in yet another window: Outlook, Teams, Logo or Mikro, a WooCommerce or ikas store. If the pilot misses the saving agreed in advance, it is switched off.
A sealed-off environment in your company's own account, in EU regions of Azure or AWS or with European hosts such as Hetzner and OVHcloud. GPU time only when needed, search across your internal documents, and spending tracked continuously. We neither sell hardware nor turn up to install servers.
Online workshops run department by department, built around examples from each group's actual workload. Everyone leaves with two habits: double-checking whatever a model produces, and recognising the kinds of information that stay out of any assistant for good. Sales and finance get separate scenarios.
A strong candidate recurs several times a week, a single mistake is cheap, and a person checks the output before it reaches a customer or the books. If any of those three conditions is missing, you will hear it from us before a proposal is written.
Staff review only the fields the model is unsure of. e-Fatura already arrives as structured data; the real burden is receipt photos clients send on WhatsApp and scanned expense slips. Supplier, date, net amount, KDV rate and total are extracted and passed to Logo or Mikro as a proposed journal entry.
No reply leaves without a human click. The shopper's message is matched to the order, Yurtiçi Kargo or Aras Kargo tracking is queried, and the draft states the true delivery status. Similar questions from the Trendyol and Hepsiburada Q&A boxes feed into the same flow.
Each answer points to its source. Assistants ask about the sterilisation routine or what to do when a patient cancels at short notice, and the tool quotes the relevant paragraph of the clinic's own manual. It never diagnoses and never touches patient records.
Reception has the final word. Enquiries written in German, Russian, English or Dutch get triaged and compared with free rooms and board packages, and a proposed answer is drafted in whatever language the guest used.
Prices and delivery terms never come from the model. Take a hypothetical exporter in Gaziantep: incoming English or Arabic enquiries are summarised, suitable catalogue items are suggested, and the sales rep types the figures and sends the reply personally.
Customer data does not belong in a free chatbot on someone's personal account. Before any trial we agree together which information may go to an outside provider and which must stay inside your Microsoft 365 tenant or an environment you own. When personal data is involved, KVKK calls for a data processor agreement, an entry in your personal data inventory and an update to the privacy notice. If the provider's servers sit outside Turkey, the KVKK rules on cross-border transfer come into play as well. The Personal Data Protection Authority's recommendations on generative AI make a sensible starting point. We keep everything in writing so that, as data controller, you can justify your choices whenever asked.
First we test at a desk using genuine historical examples, then live with a handful of staff; opening the tool to the whole company comes last. Meetings, configuration and reviews take place on Teams or Google Meet, Monday to Friday, 09:00-18:00 Turkey time.
We choose a single job and gather around fifty real examples of it from the past: inbound emails, scanned receipts, guest messages. We also record how much time a person currently spends on them.
Before any employee sees the tool, the model is run against those samples and its output compared with what your staff actually did at the time. The pass mark is set in advance, say “amount and KDV rate correct on nine documents out of ten”.
If it passes, a couple of volunteers add it to their normal routine for a few weeks. Every draft approved, corrected or thrown away is counted, and the real saving is worked out from those counts.
Good numbers mean the tool reaches everyone else with a one-page usage rule and the KVKK records in place. Borderline numbers earn one more round of tuning. Poor numbers end the project, which costs far less than a yearly subscription to software nobody opens.
At present there is no standalone AI statute in Turkey, although regulatory work in this field is worth keeping an eye on. For now the framework is largely set by KVKK: where personal data is processed, the rules on legal basis, transparency, security and cross-border transfer all apply. Regulated sectors such as finance may add their own requirements. Our part is the practical side, meaning which tool is used with which data and for what purpose. Legal assessment remains with your lawyer or KVKK adviser.
It may. The regulation can reach beyond companies established in the EU to those placing AI-enabled products or services on the EU market, and to systems whose output is used there. How far it touches you depends on what the system does and on your role. If a European customer raises it in a supplier questionnaire, a tidy register of the tools and data you use is half the answer, and we can compile that register with you.
Yes, with open-weight models. The model runs inside your company's cloud account on a GPU machine in an EU region, or with a European host such as Hetzner or OVHcloud, and your text never passes to a third-party AI vendor. The trade-offs are real: GPU hours are expensive, patching and upgrades fall to you, and these models may trail the largest commercial ones in capability. For most SMEs the sensible middle ground is a hosted service whose contract fixes the processing location and keeps your content out of training. Those terms shift over time, so we reread them whenever a project starts.
A flat ban rarely ends the habit; it just drives it underground, and people carry on from their phones. What works better is offering one sanctioned tool, managed under a company account with known data terms, plus a single page of rules: no customer names, national ID numbers, health details, price lists or contracts in any outside assistant. Once everyone knows which tool is approved, the unofficial use fades quickly.
Who can open what. Copilot draws on every file and message the signed-in person has access to, so SharePoint folders shared with “everyone” years ago suddenly become very visible. If payroll or board papers are lying around, we tighten permissions and sort out sensitivity labels first. Then, rather than licensing the whole company, buy five or ten seats and review after a month who is really using it and for what. For work that depends on Logo, your ERP or the online store, Copilot alone rarely suffices.
Three items. First, our time: a fixed fee or €55 per hour plus VAT, with a written ceiling either way. Second, model usage, which the provider invoices to you directly by volume of text processed; we forecast it at the outset and track it daily during the trial. Third, if needed, the cloud environment itself. For continuity after the pilot, AI work can be added to a Start, Business or Premium plan, and a dedicated team is quoted on scope.
Describe the job your team repeats every day and the software it happens in. We will look at the data available and propose a pilot with a measurable goal.
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.
