Gains:
- Being able to divide managerial tasks into three layers: mechanical, analytical and responsible decision-making, and distinguish which role artificial intelligence can safely undertake in each layer.
- Ability to apply a discipline that verifies each artificial intelligence output through the steps of connecting it to the source, recalculating it and passing it through administrative filtering.
- Anonymizing company and personal data within the scope of KVKK and gaining the habit of choosing a safe vehicle
Running a business means spending most of your day surrounded by numbers, reports, emails, meeting minutes and decisions. Whether you are a general manager, operations manager, business development specialist, human resources manager, purchasing officer or a small business owner; The essence of your job is the same: turning dispersed information into a meaningful decision. This is where artificial intelligence (in short, AI - a computer system that understands human writing and produces text, summaries and analysis) comes into play. But let's start from the beginning with a very clear sentence: AI does not make decisions, it gives time and drafts to the manager who will make the decision. In this unit, we will learn where AI produces real value in business management, where it is dangerous, how to verify each output and how to protect company data.
What exactly does AI accelerate in business?
Let's divide a manager's job into three layers. The first layer is repetitive and mechanical tasks: compiling the monthly report, summarizing the meeting note, drafting the email, making the spreadsheet readable. At this layer, AI is an almost perfect assistant, reducing your hours to minutes. The second layer is analysis and interpretation: finding trends in sales data, questioning why a KPI (Key Performance Indicator — the number that measures the health of the business) is falling, coming up with scenarios. Here AI is a powerful thinking partner, but what it says must be verified. The third layer is the responsible decisions: who to fire, which supplier to contract with, how to divide the budget. This layer belongs to the human; AI only makes options visible.
Distinguishing these three layers is the most important skill of this module. Because not using AI at the first layer is laziness; Using it blindly in the third layer is irresponsible.
Tip: Before outsourcing a task to AI, ask yourself: “What will be the outcome if this output is wrong?” If the result is a typo, use it with peace of mind. If the outcome affects a person's job or the company's money, be sure to use the AI only as a draft generator and verify.
Verification discipline: the danger of unsigned decisions
An experienced manager reads a document before signing it. The AI output is also an unsigned document. AI sometimes produces hallucinations — that is, it invents a number, source, or fact that does not exist in highly convincing language. So filter each AI output through a three-step filter:
- Connect it to the source. Does the number given by the AI come from the data you give? Or could it have been made up? Do not accept any numbers that are not in your data.
- Recalculate. If it gives a percentage, a total, or an average, check it manually or with a table. AI is also wrong in arithmetic.
- Administrative filter. Does the output match your industry knowledge, the reality on the ground, and the reality of the company? “Sounds right” is not enough; "I know it's true" is required.
This discipline may seem tiring, but with practice it becomes automatic and saves you ten times your time. An AI report used without verification may one day cause you to present a wrong number to the board of directors and lose credibility.
Data privacy: do not write company secret in chat box
Business data is precious and often confidential: employee salaries, customer lists, sales figures, supplier prices, strategies yet to be revealed. Sticking this information into a public AI tool is like telling a secret to someone you don't know. KVKK (Personal Data Protection Law - the law regulating how personal data will be processed in Türkiye) requires you to protect employee and customer personal data.
So the rule of thumb: de-identify (anonymize) data before giving it to the AI. Change names to codes like "Employee A", "Supplier 1", "Customer X", etc. Use rate or band instead of actual salary. If possible, choose a corporate AI tool approved by your company that does not use data for training. The checklist at the end of the unit will turn this into a habit.
Attention: Pasting the real customer list or payroll into a public tool by saying "it's a small table anyway" is a risk of both KVKK violation and loss of trade secrets. If in doubt, always encode the data.
three mini cases
Case 1 — Correct use that saves time. The operations manager of a logistics company spent 90 minutes every Monday compiling the weekly operations report. He began having AI prepare a summary and interpretation of anonymous weekly data; The time was reduced to 20 minutes. He devoted the 70 minutes he earned to the site visit. The AI wrote the report, but the director checked every number.
Case 2 — The cost of skipping verification. A business development specialist of a retail chain asked AI "last year's growth rate" and put the 34 percent figure into the presentation without verifying it. At the meeting, the finance manager asked about the figure; real growth was 19 percent. The AI had made up a prediction with the missing data given by the expert. The expert lost confidence. Lesson: the number that is not linked to the source does not enter the presentation.
Case 3 — Return from privacy breach. An HR manager was about to paste a table full of all employees' names and salaries into a free AI tool. One of the team warned. Instead, they turned names into "Employee 1-40" codes and salaries into bands. The analysis was done with the same quality, no personal data was leaked.
Four copyable templates
1) Question about placing the job on the layer:
Your role: an experienced business consultant. I'll describe the job as follows: [job description]. Tell me if this is (a) mechanical/repetitive, (b) analysis/interpretation, (c) responsible decision-making. Briefly explain which role is safe for the AI to take on in this job, which part must remain human.
2) Verification check prompt:
I will examine the AI output below. List me every numerical claim in it, item by item. For each "does this number come from user-provided data or is it an inference/guess?" mark. Collect those of unknown origin under a separate 'VERIFY' heading.
3) Anonymization assistant:
I'll give you a chart in a moment. First, list me the fields in this table that may be personal/confidential (name, ID, salary, customername, supplier price) and suggest how they can be coded for each (e.g. name -> Employee A). Don't analyze yet; only the privacy map is output.
4) Administrative filter request:
I will evaluate the following AI suggestion: [suggestion].Test me this suggestion with 5 critical questions a manager should ask (realism, cost, risk, feasibility, ethics). Show the weak points of the proposal honestly; Don't try to get it approved.
Weak prompt / Strong prompt
Weak prompt:
Write me a business report.
This request is context-free: it is not clear which company, which period, which data, to whom it will be presented. AI produces a generic, useless text.
Powerful prompt:
Your role: assistant analyst to the operations manager. Write a 1-page executive summary using the attached anonymous weekly data (number of shipments, delay rate, cost). Structure: (1) 3 critical findings of the week, (2) change compared to last week, (3) 2 risks that need attention. Just use the numbers I give you, don't make up any numbers; Mark "no data" where you are unsure.
Size
poor approach
Strong approach
Role description
None
Net (analyst)
Data
uncertain
Anonymous and attached
Output structure
free
substantial and limited
fitting protection
None
"Don't make up the numbers" rule
availability
low
high
Common mistakes
- Giving all work to AI or none of the work to AI. The correct thing is to separate the work according to its layer.
- Skipping verification. Output that seems “convincing” may not be accurate; check each number.
- Pasting real personal/confidential data into the open tool. Anonymize first.
- Getting AI to sign a decision. AI produces suggestions; Responsibility and signature belong to the manager.
- Writing prompts without context. A request without a role, data and output structure is useless.
Tip: For the first week, use AI only on first-tier work (summary, outline, table editing). Once trust and habit are established, move on to the second layer (analysis). Always keep the third layer (decision) within yourself.
In summary
In business management, AI is a powerful assistant that speeds up mechanical work, a good thinking partner in analysis, but never a decision maker. Separate jobs into three layers: mechanical, analytical, and responsible. Validate each output with source linking, recalculation, and administrative filtering steps. Make sure to anonymize company and personal data; Don't forget KVKK and trade secret liability. An unverified AI output is as risky as an unsigned decision.
Application task
Write down 6 tasks from your own business that you did this week in a list. Label each “mechanical/analytical/decision.” Choose one of the mechanical ones and have the AI do it with the above prompt “Placing the job on the layer” followed by an appropriate outline. Run the output through the three-step verification filter and note the time you save.
checklist
- [ ] Have I classified the task as mechanical / analytical / decision?
- [ ] Did I anonymize the data before giving it to the AI?
- [ ] Have I sourced and recalculated every number in the output?
- [ ] Have I compared the output to my industry knowledge and the field?
- [ ] Have I been able to maintain that I have the final decision and signature?