I’m Egils Boitmanis. As an agency owner, consultant and coach, I have worked with thousands of clients across the world. I have hired other consultants and held CEO, CMO and COO roles in various startups. I have experienced consulting from both sides: being asked for answers and paying someone else to help me find them.
That experience shapes how I would choose an AI implementation partner today: identify the business process that needs improvement, then find someone who understands that work and can use AI to improve it. Their technical ability matters. So does their judgement about what is worth building.
Keep experienced people close to your business
An owner cannot see every blind spot from inside their own company. My advice is to build relationships with trusted advisers early. An hourly conversation or a clearly scoped retainer can give you access to experience you do not need to hire full-time.
I would start with finance. A good product and strong marketing cannot keep a company operating indefinitely if it runs out of cash. Someone who understands your numbers can help you see a problem while you still have choices.
Next come sales and marketing: people who understand customers, demand and how the business earns revenue. Culture, leadership and the owner’s personal growth also deserve experienced support. The right mix depends on your stage and constraints.
I want advisers who challenge my assumptions, understand the business and leave us better able to make decisions. I apply the same level of care, expertise, and commitment to every AI consulting project I undertake.
What research tells us about consulting and AI
Four studies help explain why outside expertise can be useful and what to look for when putting it to work. The hiring advice below is my interpretation of their findings, applied to AI implementation.
1. Consulting can strengthen specific business capabilities
Bruhn, Karlan and Schoar’s study of 432 small and medium enterprises in Mexico found that access to a year of management consulting improved productivity and return on assets. Changes varied across firms, with marketing, financial accounting and long-term planning among the prominent areas. [1] My takeaway: ask a consultant which part of your business they can improve and how they would approach it.
2. Better processes can matter more than another tool
Research conducted by Bloom and his colleagues showed that Indian textile factories improved their production efficiency when they adopted professional management practices under expert supervision. Improvements came from reducing defects, organising production more efficiently and maintaining better control over stock. [2] Many of these methods were already widely recognised, but the factories had not applied them consistently. This finding is relevant to AI implementation because technology works best when existing workflows are clearly defined. Introducing automation into a poorly organised process may reinforce existing problems rather than resolve them.
3. Improvements need someone to maintain them
A follow-up to the Indian experiment found lasting differences between the groups, although about half of the practices originally adopted had been dropped. Frequent changes in management and insufficient availability of directors were commonly identified as key contributing factors. [3] I read this as a handover lesson: decide who will maintain the workflow, train colleagues and update the instructions after the consultant leaves.
4. AI capability depends on the task
Dell’Acqua and colleagues studied 758 Boston Consulting Group consultants. AI assistance improved performance on tasks within the model’s capabilities, but reduced the likelihood of a correct solution on a task chosen to fall outside them. [4] A polished demonstration is therefore only a starting point. I would ask a partner to test the actual work, including difficult inputs, and show how a knowledgeable person checks the result.
These experiments studied particular programmes, firms and tasks. They do not establish that every consultant or AI project will produce the same gains. They do give us useful reasons to look for diagnosis, practical implementation, task-specific testing and continuity.
Choose experience in your business problem, alongside AI skills
My view is that understanding AI tools alone is becoming a less useful way to define a consulting offer. As more people learn to use them, the valuable question becomes: what does this person understand about the work we need to improve?
Computers offer a useful comparison. We still need technical specialists for demanding work, but using a computer is now part of ordinary business practice. I expect AI capability to become embedded in many professions in a similar way. This is my view of the direction, rather than a claim that demand for AI specialists has already disappeared.
If customer relationships are the bottleneck, look for experience in customer service, account management or retention, alongside practical AI knowledge. If the problem is sales follow-up, look for someone who understands sales processes and can implement the right workflow. A purely technical builder may keep adding features while the business still struggles with the original problem. Equally, a domain expert needs enough technical understanding to make realistic choices.
Read their CV and LinkedIn experience, then go deeper. What responsibility did they actually hold? Which comparable process have they improved? Can they explain what they built, what did not work and what the client maintained afterwards? Ask for references where available, and inspect practical work rather than relying only on titles or certificates.
Start the relationship with one useful piece of work
I would write a short brief before choosing a partner: the process, the current difficulty, the information available, the people involved and the result we want. Then ask the candidates to explain their first step.
- Define a small deliverable. Clearly define the expected deliverables and assign responsibility for checking the completed work.
- Test the whole job. Include preparation, errors, checking and correction when assessing time or quality.
- Agree access and ownership. Specify which data and accounts are involved, what the system may change and what the company keeps.
- Plan the handover and cost. Identify a responsible employee, maintenance needs, subscriptions and ongoing support.
For example, a customer-service pilot might prepare replies from approved product information and a customer’s question. A colleague checks accuracy and tone before sending. The partner should understand the service process, the technical setup and the consequences of a wrong answer. This is an illustrative starting point, not a measured client case.
My vision: a business-owned AI operating system
The ownership question is one reason I created the Arete AIOS approach. My vision is a company-controlled AI working environment on a capable computer the business owns, holding its important knowledge, instructions and current work. It would use leading language models when useful, sending only the information needed for a particular task.
I want that environment to minimise sensitive data leaving the company and remove or mask identifying information before selected context is sent to an external model. This requires carefully tested data handling and access controls; keeping files locally alone does not secure every connection.
The longer-term vision includes voice interaction and practical access for employees. People could ask the system to help with their work, inspect the result and contribute corrections. Those corrections could improve maintained knowledge and instructions, while the AI helps employees learn and work more effectively. The aim is an organisation that keeps learning from its own experience.
What I have already put into practice
I have developed Arete AIOS through daily use in my own work and through teaching AI business brains at the University of Latvia and in cohorts co-taught with Gianluca Giuliani.
The current method organises five layers:
- Constitution for the owner’s rules
- Memory for business knowledge
- Skills for repeatable instructions
- Tools for connections and actions
- Focus for current priorities
We begin with one useful AI team member or workflow. The owner reviews the result, saves corrections and decides whether there is a reason to build the next one. Hands-on sessions and a community of entrepreneurs provide a place to ask questions and share what people are learning.
That practical work is the foundation. The integrated local environment, voice interface and automated data preparation described above are the direction I am building towards, rather than a promise that every part is already available as a finished product.
For owners who want to try the first step themselves, the free Business Brain build guide offers a practical place to begin. Whether you build yourself or hire a partner, start with a business responsibility you understand, one result you can inspect and a system your team can maintain.
