Short answer
Setting up an AI employee takes about a month to do well: prepare your rules and tool access in week one, go live with two or three jobs and approvals on in week two, review conversations and tighten rules in week three, then expand and measure in week four. The setup itself is a few focused hours, not a project.
What you are really setting up
An AI employee needs three things, the same as a person: access to the tools where work happens, knowledge of how you want things done, and a clear list of what it is responsible for. Setup is giving it those three. The technology matters less than your clarity about the jobs.
With Oraculo, an engineer joins four 1-hour Zoom sessions and connects your tools with you, so you are not configuring integrations alone. Your job is to bring decisions, not code.
Before day one: the preparation list
- Pick two or three jobs. Not twelve. For example: answer the business line after hours, follow up on quotes, send invoices. Use first 10 tasks to give your AI employee to choose.
- List your tools. Email, calendar, CRM, accounting, website platform, phone provider, social accounts. If a tool has an API or a login, Oraculo can usually work with it.
- Collect your rules. Prices, hours, service area, what you refuse, discounts you allow, how you greet customers.
- Decide approvals. Which actions need your OK: refunds, discounts, public posts, messages to unhappy customers.
- Write down your baseline. Missed calls per week, average reply time to a lead, invoices over 30 days late. You will measure against it.
- Pick an owner. One person (you, or a trusted manager) answers the agent’s questions during the first month.
Week 1: connect and teach
Use the setup sessions to connect tools and walk through your jobs. Expect to make account owners available to grant access, so schedule sessions when the person who controls your phone line or domain is reachable.
| Session | Typical focus | What you bring |
|---|---|---|
| 1 | Goals, channels and the first connections (email, calendar) | Your list of jobs and baseline numbers |
| 2 | Phone, messaging and CRM | Account access; scripts or examples of how you answer |
| 3 | Payments, invoicing, website or social | Price lists, logos, past posts, templates |
| 4 | Testing and approvals | Questions and edge cases that worry you |
Your engineer will adapt the order to your business. The point is that every connection is tested while you watch, not discovered later.
Teach it your way of working in plain language, and write the rules down. A short document is better than a long one; see how to write SOPs your AI employee can follow.
Week 2: go live with training wheels
Test as a customer before real customers do. Call your own line. Text from a second phone. Send a photo of a damaged item. Ask for something you do not offer. Be rude once. Write down every answer that makes you wince.
Then go live with approvals on. In practice that means customer-facing quotes, discounts and refunds wait for you; routine answers and bookings go out on their own. Give it a few days of real traffic, and read every conversation in the first week. A message like this is typical for week two:
Week 3: read, correct, tighten
Review a sample of conversations, around twenty. For each miss, decide whether the problem was a missing rule, an unclear rule or a judgment call that should stay with you. Fix the rule, and it remembers.
- Loosen approvals where it has been right every time, such as standard appointment confirmations.
- Keep strict approvals where mistakes cost money or trust.
- Add the three questions that came up that you did not anticipate.
- Check handoffs: did the right things reach you, with enough context?
Common mistakes and their fixes are covered in AI employee mistakes, and the security side in AI employee security checklist.
Week 4: measure and expand
Compare your baseline with the new numbers. Did missed calls fall? Did reply time to leads drop? Are invoices getting paid sooner? How to measure your AI employee’s performance gives a longer list of measures. If the answer is yes, add the next job: follow-up on quotes, review requests, weekly KPI reports. If it is no, find out why before adding anything.
This is also the point to decide on a plan. Starter is $12 per hour of work with a 6-hour weekly minimum (about $312 a month). Business is $8 per hour with a 43-hour weekly minimum (about $1,500 a month) and a dedicated Slack account manager. Setup is $499 once. The 30-day money-back guarantee gives you a clean exit if the numbers did not move.
Common setup traps
- Too many jobs at once. You cannot tell what works.
- Vague instructions. “Be professional” is not a rule. “Quote from this list; never discount” is.
- No named approver. Questions go unanswered, and work stalls.
- Skipping the baseline. You cannot prove it worked.
- Expecting perfection on day one. Plan for a month of tuning.
To walk through your own list of jobs and tools with an engineer, book a 20-minute demo.
Key takeaway
A good setup is small and measured: two or three jobs, clear rules, approvals on, and a month of review. Expand only after your baseline numbers move.
Frequently asked questions
How long does it take to set up an AI employee?
Setup with Oraculo is four 1-hour Zoom sessions with an engineer, and the employee can start working after that. Plan on about 30 days to tune rules, review conversations and expand to more jobs.
Do I need technical skills?
No. You need to know your prices, rules and tools, and to give access to accounts. The engineer handles the connections. You describe work in plain language, the way you would to a new hire.
What does setup cost?
A one-time $499 fee, including the four 1-hour Zoom sessions. Monthly plans begin at about $312 for Starter and about $1,500 for Business, with phone calls at $0.19 per minute and a 30-day money-back guarantee.
Which jobs should I start with?
Pick two or three repetitive, rules-based jobs where delays cost you money: answering calls after hours, following up on quotes and leads, or sending invoices. Expand after your first month of results.
What if it gets something wrong in week one?
Expect a few misses. Keep approvals on for money and public messages, read the conversations, and correct the rule behind each miss. Corrections are remembered, so the same mistake should not repeat.