How to pick the first two tasks to hand to AI in your New York business
A plain way to list what repeats in your week, judge which tasks are safe to start with, and check whether the change actually helped.
By the end of this guide you will have picked the first two tasks worth handing to AI in your own week. You need a sheet of paper or a notes app, about twenty minutes, and an honest memory of your last few workdays. No tools, subscriptions, or technical setup are required to choose.
The method comes from how a good first conversation about AI should work: look at two tasks that repeat in the owner's week, and pick a first result worth having. Do not start with the biggest problem in the business. Start with work that repeats and that you can check.
Step one: list what repeats
Think back over your last few workdays and write down anything you did more than once. Do not organize it yet. The goal is a raw list of ten to fifteen items, which you will cut down. Most owners find the repeating tasks hide inside email, phone messages, quotes, intake forms, and paperwork.
- 01Open a blank note
Write today's date at the top and title it something simple, like 'AI first two'.
- 02Walk your week backward
Start with today and move back three or four workdays, writing each task you did more than once, such as answering the same kind of email twice.
- 03Write tasks, not projects
Each line should be one small piece of work, such as 'reply to catering inquiries' rather than 'grow catering'.
- 04Mark how often each repeats
Next to each item, write roughly how many times it showed up this week, like 'x4'.
- 05Circle the ones you dread
Mark the circled ones that also eat time or sit unfinished. Those feelings point at good candidates.
- 06Cut to five, then two
Cross off anything that happens rarely, needs a physical hand, or cannot be described in a sentence. Of the rest, pick two.
- 07Write the result you want
Under each of the two tasks, write one sentence naming the result, such as 'catering inquiries answered within a day in my voice'.
Judge which tasks are safe to start with
Not every repeating task belongs in the first round. A safe first task is routine, mostly text, and low-risk if a draft comes out wrong. You stay in the loop either way. You read what AI drafted before it goes out under your name.
Use three tests. Does it repeat the same way each time? Is it words, such as an email, a description, or a summary, rather than a judgment call? If AI got it wrong, could you catch it in thirty seconds before sending? Two yes answers make a reasonable first task. Three is ideal.
Here is what that looks like in businesses a New York reader will recognize. These are examples, not clients. A Brooklyn restaurant answering catering inquiries: the inquiry email asks the same questions every time, so the reply pattern repeats, and the owner can check a draft against the real menu and dates before sending. A Long Island contractor following up on sent quotes: the follow-up is a short message, the facts live in the quote itself, and the owner signs off before it goes. An Albany accounting office chasing documents from clients: the request is nearly identical each time, and a draft reminder is easy to review before it is sent.
The other side of the line matters too. Leave anything with legal or tax consequences, anything with a price you have not set, and anything that depends on a relationship only you hold for later rounds. AI can help draft those eventually. It should not carry the first two tasks you hand over.
Measure what changed
Before you hand over the first task, write down what it costs you now. Two numbers are enough: roughly how many minutes the task takes per round, and how many rounds you did this week. Multiply them. That is your baseline. Without it, 'it feels faster' is the only thing you will ever be able to say.
- 01Write the baseline
Before the first try, note minutes per round and rounds per week for each of the two tasks.
- 02Set a review date
Pick a date two or three weeks out and put it in the same note.
- 03Log each round
Each time you use AI on the task, jot how long you spent reviewing and fixing the draft.
- 04Compare at the review date
On the date, add up your time now versus the baseline, and count anything that improved on the side, such as faster replies or fewer loose ends.
- 05Keep or swap
If a task got worse or stayed flat, say so in the note and pick a different one. The note is for you, not for a report.
Some wins do not fit in minutes. Replies go out the same day instead of sitting for three. Follow-ups stop falling through. Those count, and they are worth writing down in the same sentence as the time. What matters is that you can say what changed, in your own words, from your own note.
Honesty runs both directions. New York AI Lab does not guarantee savings or revenue. It picks useful work, measures what changed, and reports honestly. Your own note should hold itself to the same standard.
When you want a person in the room
If you would rather not do this alone, this is where a service like ours fits the picture. New York AI Lab gives practical AI help to New York business owners and their teams. On a free 25-minute fit call on Zoom, we look at two tasks that repeat in your week, tell you honestly whether AI Concierge, training, or neither is the right fit, and pick a first result worth having. Replies come within one business day.
AI Concierge is a real person from our team who learns your business, holds two 1-hour working calls a month on real tasks, and answers voice messages and email between calls with no limit on questions. Everything we build lives in your own accounts and is yours to keep. Bigger builds, such as API connections, CRM integrations, hosted agents, or dashboards, are scoped and quoted separately before anything starts. If what you want is a shared skill across your people, AI Training for Business is hands-on, built on your real work, runs in sessions of 90 minutes and up, and ends with take-home routines, safe-use ground rules, and a follow-up check-in. Owners usually need one AI subscription, such as ChatGPT or Claude, paid directly, and any cost is named before anything is bought.
Where people get stuck
The list comes out too big. Every task feels important when you write it down. Force a rank: keep only what repeats weekly, then keep only what you can describe in one sentence, then keep the two you would hand to a competent new hire on day one.
The first task goes badly, so the whole idea feels wrong. One bad draft usually means the task was vague, not that AI cannot help. Rewrite the task as an instruction you would give a person, including the facts the reader needs, and try once more before swapping in your second task.
The measurement gets skipped because the first try felt fine. A feeling is not a number. Go back and estimate the baseline from memory if you must. A rough count you wrote down beats a precise one you never wrote.