How to implement artificial intelligence in a business: start with the process, not the tool
Key Takeaways
- To implement AI in a business, start with a specific process that costs you time or money today, not with choosing a tool. The tool comes last.
- Almost 9 out of 10 organizations already use AI in at least one function, but only 6% turn it into significant financial impact (McKinsey, The State of AI, August 2026).
- The difference lies in redesigning the work: about 3 out of 4 high-performing companies redesigned their workflows around AI, compared with one in four among the rest.
- In Colombia, 82% of surveyed employees regularly or occasionally use personal AI tools for work, without the company ever deciding it (EY, Work Reimagined, 2026). Adoption has already started, even if nobody is managing it.
- A well-chosen first project should show measurable results in 60 to 90 days. If after three months nobody can say what changed, the problem wasn't the technology.
Most companies are implementing AI backwards.
The owner sees a demo, someone on the team recommends a platform, licenses get paid for, and there's a half-day training session. Three months later, two people use the tool to draft emails and everyone else keeps working exactly as before. This isn't an isolated case: MIT NANDA's The GenAI Divide report found in 2025 that 95% of organizations were getting no measurable return on their generative AI investments, despite an estimated investment of 30 to 40 billion dollars.
This guide explains why that happens and how to do it differently in a Latin American company that has no data department and no budget to get it wrong twice.
What does it mean to implement artificial intelligence in a business?
Implementing artificial intelligence in a business means using models that read, write, classify, or predict to do part of the work a person does today, and reorganizing the process around that change. The second half of the definition is the one almost everyone skips.
Having ChatGPT open on the sales manager's computer is not implementing AI in the company. It's one person using a tool. The company starts using AI when an entire process (answering quote requests, reconciling payments, reviewing contracts, following up on receivables) works differently because a system does part of the work and someone is accountable for making that happen every day.
In sectors like construction, agribusiness, retail and distribution, or professional services, the uses that show results fastest tend to be unglamorous: reading invoices and purchase orders, classifying customer requests, preparing proposal drafts, summarizing technical reports, or matching inventory against orders.
Why don't most companies see results from AI?
Most don't see results because they drop AI into processes that don't change. McKinsey puts numbers on it in its 2026 State of AI survey: 37% of organizations attribute some profit impact to AI, a figure that barely moved from the previous year, and only 6% get AI to account for 5% or more of their EBIT.
What sets that 6% apart isn't budget or the model they use. According to the same study, about three out of four high-performing companies redesigned their workflows around AI, while only one in four of the rest did. Some changed the way they work, while the others added an assistant to the same old routine.
In the companies we work with in Colombia and Central America, we see three paths that keep repeating and almost never work. The first is buying the tool before knowing what it's for, usually after a trade show or a video. The second is generic training, where everyone learns to write prompts but nobody knows which process to use them in the next day. The third is the endless pilot, which works well in the presentation and never reaches operations because nobody is responsible for taking it there.
| Common path | What happens in practice | Approach that works |
|---|---|---|
| Buying licenses first | Low, scattered usage, fixed cost with no clear return | Choose the process, then the tool |
| General prompt training | One week of enthusiasm, no change in operations | Training applied to a real team process |
| Pilot without an owner | Stays a demo | One owner with one metric and one deadline |
| Measuring "tool usage" | Numbers the business doesn't care about | Measure hours, errors, response time, or sales |
Is AI already entering your company even if you didn't decide it?
Yes, and probably through the back door. EY's Work Reimagined study found in 2026 that 92% of Colombian workers surveyed use AI tools at work and that 82% regularly or occasionally use personal AI tools, what is known as shadow AI. MIT reported something similar globally: in more than 90% of companies employees use personal AI, even though only 40% have bought an official subscription.
There are two ways to read this. The worrying one is that customer information, prices, and contracts are flowing through tools the company doesn't control. The good one is that your team has already taught itself to use AI to search for information, draft emails, and summarize documents, which are precisely the three most common uses EY reports. The manager's job isn't to start from scratch, but to bring order to what's already happening and point it toward the processes that move the business. When that goes well, many of the best solutions come from employees themselves, as we explain in what is employee-driven digital innovation.
How to implement artificial intelligence in a business step by step
The order we recommend is process, person, test, and measurement. At Suricata Labs we organize it with the OAT framework (Optimize, Accelerate, Transform), which helps decide what to do first and what to leave for later.
1. List the processes that hurt operations the most
Sit down with your team and write down where their time goes and at which moments customers are left waiting. Don't ask "where can we use AI?", because that question produces catalog answers. Ask which task they would make disappear if they could.
A distributor might discover its team spends hours typing purchase orders from PDFs into its system. A construction company might find that nobody reads the site supervision reports all the way through. Those are good starting points because the pain is already identified and anyone can see whether it improved. If you first want to know whether your company has the data, people, and governance to do it, see how to assess if you're ready for AI.
2. Classify each process with OAT
Optimize means using AI to do something you already do faster or with fewer errors: sorting emails, extracting invoice data, preparing drafts. This is where almost every company's first project lies, because the risk is low and results are measured in weeks.
Accelerate means redesigning a workflow so AI does a large part of the work and the person focuses on reviewing and deciding. For example, the system prepares the complete quote with prices and inventory, and the sales rep just adjusts and sends it.
Transform means creating something you couldn't offer before, like a new service or a business model that depends on AI. It's worth it, but it should almost never be the first project.
3. Pick a single process and give it an owner
The first project must meet three conditions: it repeats many times a month, it has a result that's easy to measure, and there is a person willing to be accountable for it. Without an owner, the project dies even if the technology works.
4. Set the baseline before touching anything
Measure how the process works today, for example how many hours the team spends on it each week and how long the customer waits for a response. Without that data, three months from now you'll have opinions instead of results.
5. Now choose the tool
With the process clear, the tool almost chooses itself. Often what you already pay for is enough, because Microsoft 365, Google Workspace, and several ERPs already include AI features. Other times you need an agent connected to your systems. What matters is that the decision comes from the process and not from the vendor's advertising.
6. Test with real data for 30 to 60 days
Work with real company information and with the people who do the work every day. The most useful adjustments come from them, not from the vendor. Define from the start which information can go through the tool and which can't.
7. Measure, decide, and repeat
At the end of the test, compare against the baseline and make a decision: scale, adjust, or discard. Discarding is also a good outcome if it was done quickly and cheaply. Then go back to the list from step 1 and choose the next process.
What to measure to know if AI is working in your company
Measure what matters to the business, not tool usage. The number of people who opened the platform says nothing about whether the company earns more or works better.
| Process type | Useful metric | Example 90-day target |
|---|---|---|
| Customer service and sales | Response time to requests or quotes | From two days to a few hours |
| Administrative and finance | Weekly hours spent on data entry or reconciliation | Cut in half |
| Operations and quality | Errors or rework per month | A visible drop against the baseline |
| Management | Time to have a report ready for decision-making | From a week to a day |
The values in the third column are references to help set your own target. Each company should set it based on its own baseline.
Real case: 18 companies in Ciudad Juárez that started with the process
In August 2026 we worked with eighteen companies in Ciudad Juárez, Mexico, as part of Innova Lab V3.0: cement producers, maquilas, metalworking shops, and retailers. None had a technology department or a software budget.
Before opening any tool, each company wrote down its baseline: when this happens, today we do that, it takes us this long and costs us this much. In three days they left with working prototypes. A metalworking plant built a purchase order analyzer for the traceability it had been tracking by hand. A cement company built one that assigns each order to the nearest distribution point. A tire distributor built an inventory analyzer that replaced hours of work in Excel.
None of them started with the tool. All of them started with a process that was already costing them money. Read how the program went in Ciudad Juárez.
Frequently asked questions about implementing AI in a business
How much does it cost to implement artificial intelligence in a business?
A first optimization project can cost little more than the licenses you already pay for, because many office suites and management systems already include AI features. The real cost is the team's time to define the process, test, and adjust. Projects that integrate agents with several systems or create new products require more investment and are best done once the first ones have produced results.
How long does it take to see results?
For repetitive, well-defined processes, 60 to 90 days is a reasonable timeframe to see measurable changes. If the project shows nothing in that time, the most likely cause is that the process was poorly chosen or has no owner.
Do I need a data team or engineers to use AI?
Not to get started. The first projects can be done with commercial tools and with the people who know the process. In El Salvador we trained 117 business advisors in AI, one in two starting at a basic digital level, and they ended up building products that used to require hiring an outside vendor. What you do need is someone with the authority to change how work gets done and the judgment to decide what information can be used with these tools.
Which AI examples work best at the start?
The ones that show results fastest are extracting data from invoices and purchase orders, classifying customer requests, drafting quotes and proposals, summarizing reports and contracts, and following up on receivables. What they have in common is that they repeat often and the output can be reviewed.
What are the risks of using AI in a business?
The most common risk today is employees uploading confidential information to personal tools the company doesn't control. Next come errors nobody reviews, because AI writes with great confidence even when it's wrong, and dependence on a vendor chosen in a hurry. These risks drop considerably with a clear policy on what data can be shared and with human review at the steps where a mistake is expensive.
Where to start this week
If you only do one thing after reading this, make it the list from step 1. Get three or four people who know operations well in a room and ask them which task they would eliminate if they could. That's your first AI project, and it almost never matches what the vendor shows in the demo.
If you want to do that exercise with support and come out with a prioritized plan, at Suricata Labs we run AI adoption assessments for companies in Latin America.
Explore our AI services | Schedule a conversation
Read also: Digital transformation for businesses: what it is, what it isn't, and how to do it right
Last updated: September 16, 2026
About the author
Other resources on this topic
AI for business: how to assess if you're ready
An AI readiness guide for businesses. What to evaluate before investing in AI, why ChatGPT isn't a strategy, and how to start with method.
What is employee-driven digital innovation?
Companies collect ideas from their employees and never implement them. What changes when generative AI lets the person living the problem build the solution.