What to clarify before hiring a consultant

Before hiring an AI consultant, identify the decision you need help making. It might involve choosing a process, assessing a tool or understanding whether your team can maintain a solution. A useful discussion ends with understandable priorities, identified uncertainties and a next step that someone can take responsibility for delivering.

Bringing a list of tools can help the discussion, but I first want to hear what happens in your business. Who receives requests? Where is information stored? Which work keeps returning to the same person because nobody knows how to decide? The answers reveal where to investigate and help prevent a project losing its purpose after starting with technology.

If you have already experimented with AI, bring what went wrong too. A summary that omitted a commercial commitment, a classification that confused services or an application needing constant correction all provide useful information. They reveal decisions that require context and the care needed before increasing the volume of work.

In my 2022 book, I advocated planning before execution and choosing tools according to need. Today I apply that discipline to AI and business automation. The intended outcome should be explicit: better responses, less repetitive work or a better informed decision. Each objective requires a different evaluation.

How to choose the first process to work on

The first process should involve a problem recognised by the team, inputs you can observe and an outcome you can check. Consider frequency, current effort, dependencies and the impact of failure. A small, understood process lets you learn before making larger commitments that affect the day-to-day operation of the business.

Imagine a service business where every request arrives through a different channel. Some team members copy messages into a spreadsheet, others use email as an archive and nobody checks whether an open contact already exists. Before asking AI to qualify opportunities, it is worth organising those inputs and deciding who follows up.

A rule can check mandatory fields and prevent duplicates. A language model can help interpret a request and suggest a category. The responsible person needs to see the original message and correct the suggestion. That sequence needs designing, including what should happen when information is missing.

In Sextas Ímpares #151, about the traffic manager’s work in the AI era, I demonstrated a useful distinction: I used AI to build a tool whose execution follows defined rules. Every step does not need to consult a model. I explore other applications in my article about using AI in business beyond chatbot questions.

What information to prepare for a serious assessment

To assess an AI opportunity, prepare examples of the current work, a description of the tools involved and the criteria the team uses. Identify who can authorise changes and which data need protection. This information should explain the process without unnecessarily exposing customer contacts, credentials or confidential documents during the discussion.

A carefully constructed fictional example may be enough to explain the initial flow. Show the incoming fields, the expected decision and where the result goes. If there are several exceptions, choose some representing real difficulties. This helps establish whether the challenge involves text interpretation, data quality or the definition of the service itself.

For reports, check the period, currency, column definitions and whether total rows are present. In lesson #142, about using Claude to analyse advertising, I showed how an export may include only the visible columns. A convincing analysis still depends on what was actually supplied to the model.

It also matters who maintains each tool and how changes are recorded. A process depending on a personal account without an identified replacement creates difficulties when someone leaves or is absent. Bring that information even if you do not yet have every answer: locating these dependencies is part of the assessment.

How to evaluate an implementation proposal

An implementation proposal should explain scope, responsibilities, acceptance criteria and operating costs. Ask for examples of what the solution will do and which situations will require human intervention. The decision becomes clearer when you can connect each deliverable to an identified need in your business and understand who will use it.

Start with a complete journey. If the solution classifies requests, where do they arrive, where are they stored and who receives them? If it sends messages, what authorisation allows that action? If it prepares a report, how is data freshness checked? These questions help compare proposals using similar language for very different work.

In lesson #145, about building an application with Claude Code, I explained preparation before generating code. The project involved users, scoring rules, deadlines and integrations. The demonstration is concrete; the lesson I carry into business projects is to clarify behaviour and access before accelerating development.

Include the handover to the team in your assessment. Who receives documentation? Who can correct a record? How is a failure monitored after delivery? A project may work in a demonstration yet remain difficult to operate. My article on planning an application with Claude Code develops the questions that help prepare that work.

How to establish whether the change was worthwhile

To understand whether implementation was worthwhile, compare the previous work with the new process across comparable periods and situations. Count execution, review, correction and usage costs. Also observe delivery quality and the tasks that still depend on the team, so the assessment represents the complete operation rather than one isolated step.

If a task used to take twenty minutes and the system now prepares a result in two, review time still needs measuring. Simple cases may be resolved quickly while exceptions require substantial work. Separate those groups to understand where improvement occurred and where the solution still needs adjustment.

Decide in advance what would make you continue, change or stop the trial. An incorrect classification may be corrected before a sales conversation; a message sent to the wrong recipient has different consequences. The criteria should reflect that difference and be understood by the people using the process every day.

Also follow up on the time released. Was it used to respond to more requests, improve follow-up or reduce overload? That discussion connects implementation with the business. I develop this concern in growth, margins, teams and AI, drawing on decisions I have made as a business owner.

What kind of support suits your business

The right support depends on whether you want an implementation team, training to develop internal skills or mentoring to discuss decisions. SpartAds provides the route to AI and automation services. At Laboratório da IA, I review projects in two mentoring sessions a month, while corporate training requests are assessed individually.

If you need implementation, explain the process and what you have already tried to the SpartAds AI and automation team. The initial contact should clarify the need. The scope and terms of any work depend on that assessment; this page does not advertise a universal package.

If the main difficulty is preparing people to work with these tools, read how to plan AI training for businesses. For ongoing guidance and discussion of your own projects, explore Laboratório da IA and the available training. These are different forms of support with different responsibilities for execution.