The hard part of implementing AI in your business is rarely the technology. It stalls when a company starts too wide: a working group, a list of tools, a pilot nobody finishes. Start small, with one process that comes back every week, and you have something that works sooner and you know exactly what the next step is. This plan is written for businesses without their own IT department.
What implementing AI is, and what it is not
Implementing AI means that part of the work in your business is from now on done by software that understands text and makes a proposal itself. Think of reading and answering an enquiry, assessing a lead or processing an invoice. It is something else than giving everyone a subscription to a chat program. That helps people type faster, but the work stays with the same people.
It is not a large IT project either. You usually do not have to replace your software. An AI agent works with the systems you already use, such as your mailbox, your CRM and your accounting.
Step 1: pick one process, not ten
For one week, write down which work keeps coming back and who does it. Then pick the process that meets three conditions:
- It happens often, preferably every day.
- It usually runs the same way.
- A delay costs you something: a missed enquiry, a late invoice, a customer who has to wait.
For many small businesses that is following up enquiries or answering customer questions. If you are torn between several processes, it helps to put them side by side on how much time they cost and how much goes wrong when they are left waiting. That comparison is worked out in which process to automate first.
Step 2: define what good looks like
An AI agent does what you teach it. So before anything is built, it has to be on paper how the process should run:
- What is a good answer to the ten questions you get most often?
- When is an enquiry serious, and who should get it?
- What may the agent handle by itself, and what has to go past a person first?
This is the work that gets skipped most often, and it is the reason pilots stall. You do not need technology for it, you need someone who knows the process. Usually that is the person who does the work now.
Step 3: connect to the systems you already have
The agent has to be able to read where the work comes in and write where the result belongs. So make a list of the systems in the chosen process: where do enquiries arrive, where is the customer kept, where is the appointment or the invoice made. For each system, ask whether it has an integration. Many common packages do.
Pay attention to your data in this step. Ask where it is stored and who can access it.
Step 4: start with approval first
Do not let a new agent work on its own straight away. Have it make proposals that you approve first: the draft reply, the assessment of the lead, the draft invoice. That way you see in the first weeks what it does well and where it needs correcting, without a customer noticing anything.
With us, important actions go through you first. You see the proposal in Telegram and choose: send, edit or skip. Once an action proves itself again and again, it can go automatic. You decide when.
Step 5: measure, and only then expand
Agree beforehand what you will look at. Keep it to two or three things you can already count today:
- How fast does an enquiry get a first response?
- How many enquiries are left waiting?
- How many hours a week does this process cost your people?
Measure the same things before the start and after a month. If it is better, you pick the next process. If it is not, you know that before you have put more into it. Examples of work that often follows second and third are in nine tasks an AI agent can carry by itself.
Where it goes wrong
- Starting too wide. Five processes at once means doing step 2 five times, and then it gets done halfway five times.
- Nobody owns it. An agent needs someone who reads its proposals and says what should be better, certainly at the start.
- Looking only at the tool. Choosing the program is the smallest part of the work. Defining what good looks like and keeping it up afterwards is the largest part.
- No measurement beforehand. Without a number from before the start, you do not know after three months whether it delivered anything.
Do it yourself or have it done
You can implement AI yourself if you have someone in house who can build integrations and has the time to keep maintaining them. That last part is underestimated: systems change, and an integration that works today can stop in six months without anyone noticing.
If you do not have that person, there are two kinds of help. An adviser helps you choose and makes a plan; building and maintaining stay with you. A party that builds and manages takes over that part as well. The difference is worked out in advice only versus build and run.
Automatevia is the second kind. We build an AI team on our own platform, connect it to your systems and keep managing it, with one fixed point of contact. You can cancel monthly. What it costs is on the page about AI automation services for small businesses.
What it looks like in practice
At a real estate lender, a team of four AI agents processes the leads from several channels, puts them in the CRM and assigns each to the right colleague. At a real estate company, leads from five channels arrive structured, get a qualification score and are followed up automatically. In both cases it is the same kind of work: following up leads.
Frequently asked questions about implementing AI
How do you implement AI in a business?
Pick one process that comes back often, define how it should go when it goes well, connect the AI agent to the systems you already use, have it make proposals that you approve first, and measure after a month whether it is going better. Only then expand to a next process.
Which process do you start with?
With work that happens often, usually runs the same way and where a delay costs you something. For many businesses that is following up enquiries or answering customer questions.
Do I need to replace my software to implement AI?
Usually not. An AI agent works with the systems you already have, such as your mailbox, your CRM and your accounting, as long as they have an integration. In step 3 you check that per system.
How long does an AI implementation take?
That depends on two things: how well the process is already described and how many systems have to be connected. A process that is on paper and runs in one or two systems goes fastest. After the free scan you know which work a system can take over and where to start.
What does implementing AI cost?
That depends on the number of processes and on who does the management. If you build it yourself, you mainly pay in hours of your own people, also after the start. If you outsource it, you pay a fixed amount per month. Our price is on the page about AI automation services for small businesses.
Start with the free scan
Want to know where to start? Send us your weekly tasks and we tell you what an AI team can take over. To the free scan.
