The owner of a small business hears about AI every week. The sales lead wants it for follow-ups. The office manager wants it for invoices. Somebody suggests hiring an AI consultant. The hard part is telling a consultant who will change how the week runs from one who will deliver a slide deck and leave.
What should an AI consultant for small business actually do?
Some AI consultants stop at a plan. A useful AI consultant for small business goes further: it learns how your team works, picks one job for AI, sets it up in the apps you already use, and stays until your team uses it. The deliverable is a working setup in your own accounts, not a report.
Most small firms are still at the start. The U.S. Census Bureau found that less than 20% of firms with four or fewer employees reported using AI, against 37% of firms with at least 250 employees. One likely reason is capacity. Large firms often have people whose job is to connect new tools to old systems. A small firm usually has an owner, an ops lead and a few team leads, all busy with the real work.
So a consultant for a small or mid-size business has a different job from one who advises a large enterprise. Strategy matters less. Setup, testing and handover matter more. Judge the consultant by the jobs that AI does for your team three months later.
How do you pick the first job for AI?
Pick a job that repeats every week, lives in apps you already use, and is easy to check. A good first job saves hours, and a person can see in one glance if the AI got it right.
Use three tests:
- It repeats. The same steps happen every day or every week, such as copying data from one app to another.
- It lives in your apps. The facts the AI needs are already in your CRM (the app that holds your customer records), your accounting software, your support desk or your HR app.
- It is easy to check. A person can review the result quickly before it goes anywhere important.
Good first jobs look different by team. Sales can get a short brief before each call, built from the CRM and past emails. Support can get a draft reply that a person checks before it is sent. Finance can match invoices to orders and flag what is overdue. Operations can track orders and suppliers across apps and flag late ones. HR can answer common questions from the HR app and set up new hires across tools.
The industry changes the apps, not the method. A distributor might start with dealer questions about stock. A law or accounting firm might start with time tracking and client updates. An online store might start with returns and order questions. Leave payroll changes and payments for later, when your team trusts the setup.
What should you prepare before the first call?
Prepare a short list of the jobs that eat the week, the apps each job touches, and the names of two people. One person does the work every day. The other is an admin who can connect your apps and approve a plan.
For each job, write down three things. How often does it happen? How long does it take? What goes wrong when it is late or wrong? Rough numbers are fine. They help the consultant rank the jobs, and they give you a baseline to measure against later.
List every app the job touches, including spreadsheets and shared inboxes. Note who has admin access to each app. Many projects slow down in week one because nobody can approve an app connection.
Also note the AI assistant your company already pays for, if any. A setup that ignores it means a second subscription and a second place for people to work.
What does a good first engagement look like?
A good first engagement has five steps: learn the work, write a one-page plan, set it up in your real apps, test it with the people who do the work, and hand over with a written guide. Each step has something you can see and approve.
- Learn the work. The consultant talks to the people who do the job, not only the owner. They ask what takes the most time and what an error costs.
- Write a one-page plan. The plan says which jobs the AI takes over, which apps it needs, who can use it, and how you will know it works. You approve it before any setup starts.
- Set it up in your real apps. The consultant connects your actual CRM, support desk or accounting software, not a copy or a demo account. The AI gets only the access the plan names.
- Test with the people who do the work. Your team runs real tasks and reports what is wrong. The consultant fixes it and tests again.
- Hand over with a written guide. The guide covers how to add people and apps, change who can do what, and check what the AI did.
The plan in step 2 protects you. It turns a vague goal into a scope you can check. If a consultant wants to skip it, the scope will drift and so will the cost.
Which questions should you ask an AI consultant?
Ask four questions, and expect clear written answers. Who owns what is built? Where is our data stored? Does it work with the AI assistant we already pay for? What happens after handover?
Who owns what is built? The instructions, the app connections and the results should live in accounts you control. If they live in the consultant's own tools, you rent your own process.
Where is our data stored, and who can see it? Ask which region holds your data and which companies process it for the consultant. When a consultant handles personal data for you, it is usually a processor under data protection law. The UK regulator's guidance lists what a controller and processor contract must include, such as acting only on your documented instructions and deleting or returning the data when the contract ends. Ask for those terms in writing.
Does it work with the AI assistant we already pay for? Many AI assistants can connect to other apps. MCP (Model Context Protocol) is the standard way an AI assistant calls tools in other apps. A setup built on that open standard keeps working if you change assistants later.
What happens after handover? Your admins should be able to add a person, remove a person and check the activity log without a call to the consultant. Ask to see the written guide from a past project.
What are the red flags in an AI consultant?
The main red flags are a plan with no named job, a demo built on sample data, and a setup that only the consultant can run. Any one of these means the AI is unlikely to reach daily use.
Watch for these signs:
- The proposal talks about "AI strategy" for pages and names no job, no app and no person.
- The demo uses made-up data, and the consultant avoids a test in your real apps.
- The AI gets admin access to every app, when the job needs only one or two.
- Nobody can tell you what the AI did last Tuesday, because nothing is logged.
- The work lives in the consultant's accounts, and leaving means starting over.
- The handover is a meeting, not a document.
Unapproved AI use is a related risk. A Federal Reserve note says "shadow AI", meaning use without approval, may still be a substantive problem for organizations. A consultant who sets up approved access with clear limits gives your team a safe path instead of a workaround.
When do you not need an AI consultant?
You may not need an AI consultant if the job is small, one person does it, and that person is comfortable setting up tools. Many AI assistants can already draft emails or summarize documents with no setup at all.
A consultant also fits poorly if you only want a strategy report. In that case, a strategy consultant is the better choice, and you can decide on setup later. The comparison of AI consultants and automation agencies covers which option fits which need.
Bring in help when the job crosses several apps, touches customer or employee data, or needs to keep running after the person who set it up moves on. That is where most do-it-yourself pilots stall.
Where does Elaichi Services fit?
Elaichi Services is AI consulting, automation and integration delivered by Elaichi's forward deployed engineers, who work directly with your team instead of from a distance. They follow the same steps and hand over a setup that you own.
The engineers connect AI to the apps you already use, from a catalog of 700+ apps covering CRM, accounting, support, HR and project tools. If you use an app Elaichi does not support, they add it. The setup works with any AI assistant that supports MCP, so you keep the assistant you already pay for. Elaichi never sees your conversations with the AI.
The AI can only use the apps and actions you allow for each person. The actions it runs are recorded in your activity log. When you sign up, you pick EU, US or APAC for your company's Elaichi data, and for EU and US the data store stays in that region, and the security page explains what is stored where. The setup, the instructions written for your AI, and everything it produces live in your own Elaichi account and apps. After handover, you remove the engineers' access.
To see example jobs by team before a call, browse the use cases or check your apps in the connector catalog. The full plan is on the page about AI consulting and automation services, and you can read why AI pilots stall after the demo.
