AI Use Map for Your Company — Mohamad Charafeddine

Why start with a map, not a tool?

Most attempts to bring AI into a company start with the tool: a subscription is bought, then someone looks for something to do with it. The usual result is enthusiasm in week one and silence by week four. The safer route runs the other way. Start with the tasks your team actually does, find out how long they take, what kind of data goes into them and who checks the result. Only then ask where a tool could help, and where it should not go near.

This map is a simple inventory spreadsheet. It needs no technical background and can be filled in one or two sessions with the team. What you get at the end is not a "digital transformation plan". It is a short list of two or three tasks that suit a small, controlled pilot, and another list of tasks you have deliberately decided to keep in human hands.

The rule behind the whole map: skill before tool, and task before skill. Tools change every few months; what you know about your own tasks and data is what lasts.

What the spreadsheet contains

The attached file has two sheets: "Map", which you fill in, and "Value guide", which explains the allowed options for each column. The Map sheet has six rows filled with an illustrative example, fifteen empty rows for your tasks and a totals row at the bottom. The columns are:

Column	What to write	Note
A Task	The task with a clear verb: "Reply to…", "Summarise…", "Write…"	One task per row, not "marketing" as a whole
B Owner	The role or person who does it today	Prefer roles over names if the file will be shared
C Frequency per month	How many times it happens each month	An honest approximate number is enough
D Baseline minutes	Minutes one instance takes today, without any tool	Time it two or three times instead of guessing
E Monthly minutes	Calculated: frequency × minutes	Formula, do not type here
F Monthly hours	Calculated: monthly minutes ÷ 60	Formula, do not type here
G Input sensitivity	low / medium / high	See definitions below
H AI suitability	high / medium / low / not suitable	A first estimate, corrected by the pilot
I Human reviewer	Who checks the output before it reaches anyone	No name means the task is not ready
J Pilot decision	Pilot now / Later / No pilot	Your decision, not the formula's
K Suggested flag	Calculated from sensitivity, suitability and hours	A prompt for discussion, not a verdict
L Notes	Conditions, risks, what is needed before a pilot	Write down what would otherwise be forgotten

How the suggested flag (K) works: if sensitivity is "high", it warns against piloting before the data is anonymised or an enterprise tool with a clear data-processing agreement is used. If suitability is "not suitable", it shows "No pilot". If suitability is "high" and the task takes four or more hours a month, it shows "Strong pilot candidate". Everything else shows "Discuss". You can change the four-hour threshold inside the formula to fit the size of your company.

How to fill the map: six steps

Collect tasks from people, not from memory. Ask every team member to write five tasks they repeat weekly or monthly that take longer than they should. Five minutes from each person gives a truer list than an hour of the manager working alone.
Break big tasks down. "Managing social media" is a job, not a task. Break it into writing the post copy, replying to comments and preparing the monthly report. AI helps with specific steps, not with whole jobs.
Measure current time. In the week before filling the map, each person records the real time of two or three tasks. This is your baseline. Without it you will never know whether the pilot saved anything.
Classify input sensitivity strictly. The question is not "is the task important?" but "what will be pasted into the tool?". If the task needs customer names, phone numbers, health data or salaries, sensitivity is high however simple the task looks.
Estimate suitability with three questions. Is the output text, a summary, a classification or a draft? Can its quality be judged quickly? Would a mistake be caught before it does harm? Three yeses mean high suitability; one or none means low or not suitable.
Decide together and record the reason. Spend half an hour on the sheet as a team and choose only two or three tasks for "Pilot now". Write in the notes why the others were set aside; that reason saves you from reopening the debate every month.

Definitions for columns G and H

Input sensitivity

Low: public or already-published information: product specifications, ad copy, the published returns policy.
Medium: internal but non-personal information, or content with names that can easily be removed first: meeting notes, a quote with the client's name taken out.
High: personal data about customers, staff, patients or students, detailed financial data, contracts, trade secrets, passwords. This does not go into a public tool. If it must, anonymise first or use an enterprise tool with a contract that defines how the data is processed.

AI suitability

High: text drafts, summaries, rewording, extracting tasks from text, classification, ideas. The output is quick to check.
Medium: specialised translation, data analysis, content that needs precise terminology. Possible, but review is heavier.
Low: exact calculations, decisions that affect people, tasks that depend on a relationship of trust.
Not suitable: anything that needs a professional signature, a legal or medical decision, or where a mistake cannot be undone.

Blank template (text version)

If you prefer paper or a text document to the spreadsheet, use this card for each task:

Task card for the AI use map

Task: [verb + subject, e.g. summarise the weekly meeting notes]
Owner today: [role]
Frequency per month: [number]
Minutes per instance (measured): [number]
Monthly hours: [frequency × minutes ÷ 60]
What will be pasted into the tool? [describe the input]
Input sensitivity: [low / medium / high]
Is the output text, a summary or a draft? [yes / no]
Can its quality be judged quickly? [yes / no]
Would a mistake be caught before it does harm? [yes / no]
AI suitability: [high / medium / low / not suitable]
Human reviewer: [role or name]
Pilot decision: [Pilot now / Later / No pilot]
Reason and conditions: [one or two sentences]

A complete example: a small distributor in Tripoli

Illustrative example

"Al-Marfa Distribution" is a fictional company of eight people that supplies groceries to shops in Tripoli and Koura. The manager sat with the team for an hour and filled six tasks. These are the filled rows in the spreadsheet:

Task	Freq.	Min.	Hours/month	Sensitivity	Suitability	Decision
Replying to repeat questions	120	6	12	medium	high	Pilot now
Catalogue product descriptions	30	20	10	low	high	Pilot now
Weekly meeting summary	8	45	6	medium	high	Pilot now
Payroll sheets	1	240	4	high	low	No pilot
Translating price quotes	10	30	5	medium	medium	Later
Sensitive complaints	6	25	2.5	high	low	No pilot

What did the team learn from the sheet?

The six tasks add up to about 39.5 hours a month, but more than two thirds of that sits in just three tasks, all of them suitable. That is where the pilot starts.
Payroll takes four hours and the temptation to speed it up is strong, but sensitivity is high and suitability low. Decision: no pilot; perhaps improve the spreadsheet template itself instead of using AI.
Translation is partly suitable but needs an approved glossary before any trial, so it is "Later" with a written condition.
Sensitive complaints take little time but carry a high risk to the customer relationship. The team decided a person writes them, with the option to polish the style after every detail has been removed.

The team left with three tasks to pilot, a clear human reviewer for each, and one written rule: no customer name or phone number is pasted into any tool.

After the map: what to do with it

Take the "Pilot now" tasks into the 30-day AI pilot plan (resource 15) and give each one a measure and a review date.
Use the frequency and baseline minutes in the time and value calculator (resource 16) to estimate what the pilot might save, including review and rework time.
If you need a new tool, compare candidates with the AI tool selection scorecard (resource 14), and turn the data sensitivity you found here into a knock-out condition.
Review the map every three months: tasks disappear, new ones appear, and tools improve or change their terms.

Common mistakes

Mistake: listing whole functions instead of tasks ("accounting", "marketing"). Fix: break them into steps whose time you can measure and whose output you can describe.
Mistake: guessing current minutes from memory. Fix: record real time for one week; guesses inflate the expected savings later.
Mistake: rating sensitivity by how important the task is rather than by what will be pasted into the tool. Fix: always ask what exact text or file will go into the tool.
Mistake: leaving the human reviewer column empty. Fix: a task without a reviewer is not ready for a pilot; assign someone or postpone it.
Mistake: choosing ten tasks to pilot at once. Fix: two or three are enough to learn from; expand based on results.
Mistake: treating the suggested flag as a decision. Fix: the formula only raises a flag; the team makes the call in column J and writes down why.

Completion checklist

I collected tasks from team members themselves, not only from my own memory.
Each row holds one specific task with a clear verb.
Baseline minutes are actually measured for at least two or three tasks.
Sensitivity is rated by what will be pasted into the tool, and anything with personal data is rated "high".
Every candidate task has a known human reviewer, by name or role.
We chose only two or three tasks for "Pilot now".
We wrote the reason for setting other tasks aside in the notes column.
We set a date to review the map in three months.