What AI can do today in the management of an SME

In an SME you have to decide in which tasks generative artificial intelligence saves time and how to stop it from giving generic or, worse, wrong answers. These are the uses that work today in day-to-day management.

Which tasks AI helps with in an SME

It helps above all with text-heavy tasks and repeated queries: answering questions from customers and from the team itself, finding information in manuals and procedures, extracting data from documents and summarising what the ERP figures say. Today those tasks take hours and depend on a few people.

If a task consists of searching, reading, summarising or answering with information that already exists in the company, an AI can do a good part of it. If a decision has to be made with business judgement, the AI can prepare the data, and a person makes the decision.

AI-powered support

An AI support assistant can read each query, look for the answer in the company’s knowledge and reply by email. If it is also connected to the ERP, it can log the incident, follow it up and pass whatever it cannot solve to the right person.

A classic chatbot replies with ready-made phrases. A well-implemented assistant looks for the solution to the specific problem and records every answer, so the team always knows what has been answered and to whom.

An AI wiki for internal knowledge

In many companies, what is known is spread across loose documents and the heads of a few people. An AI wiki brings together manuals, procedures and Word or PDF documents, and answers the team’s questions with the specific information and its context. That way you do not always have to ask the veterans.

It is most useful when someone new joins, in teams spread across office, factory and branches, and in the areas where the same question comes up again and again.

Reading documents and data without typing them

AI also reduces data entry work: it can read delivery notes, orders or incoming invoices and propose the data for the management program, or summarise in plain language what sales, margins or stock show. At first, a person has to validate what it proposes until it is clear that it gets it right.

When mandatory e-invoicing between companies arrives, a good part of invoices will already come structured. AI will still be needed for orders that arrive by email, paper delivery notes and documents from small suppliers.

Why many AI projects fail

A common failure is using a generic model that does not know the company and answers vaguely. Other projects fail because AI is expected to work on its own from day one or because nobody is assigned to review its answers. It needs the company’s knowledge and a supervision period with a person in charge.

Another frequent mistake is to start with the most ambitious case. It works better to choose a well-defined task, measure how much time it saves and expand afterwards.

An AI agent reproduces the way a company works: if a process is badly designed, it repeats it faster. That is why UserIT works in alliance with Laecel, a consultancy for SMEs in the Valencia area with an area devoted to artificial intelligence applied to business: Laecel reviews the processes and UserIT implements the W4 agents.

How to implement AI in phases

Implementing AI in phases means not letting it work alone until the team has reviewed and corrected its answers for a while. First the company’s knowledge is gathered, then comes supervision and, when the AI’s judgement matches the company’s, it starts answering on its own.

In detail:

1. Index documents, manuals and procedures, in whatever format they are in.
2. Review every answer with the team until its judgement matches the company’s.
3. Let the AI answer on its own and pass on what it should not resolve.
4. Keep adding corrections and new documents, which improve the answers.

Data, privacy and control

Before starting, you have to decide where the data are stored, who can add information, which queries the AI can answer and which it must always pass to a person. These decisions are made by the company, and it is worth putting them in writing together with the data protection rules you already apply.

With those rules in writing, the AI works with permissions and limits, like any other program in the company.

W4: UserIT’s AI for support, knowledge and documents

At USER Informática we work this way with the W4 line. W4SUPPORT answers customers using the company’s documentation and, connected to the ERP, logs and follows up each incident. W4MIND makes that documentation available for the team to look up on the spot. W4INVOICE reads invoices and posts them to the accounts, and W4ORDER records orders.

The W4SUPPORT, W4MIND, W4INVOICE and W4ORDER profiles are on the services page.