AI as an operating model decision, not a software decision

There is a distinction that separates companies extracting real returns from artificial intelligence from those accumulating software subscriptions without moving the needle: the former treat AI as an organizational architecture decision. The latter treat it as a new tool plugged into an old process.

A recent report from Boston Consulting Group (BCG) makes this point non-negotiable. The study examines what so-called AI-first tech companies are doing differently — and the answer is not "using more powerful models." It is that they are reorganizing entire structures around AI agents, flattening management layers, and redesigning workflows end to end. This is not an overlay. It is a foundation replacement.

What BCG found — and why it matters beyond Silicon Valley

The report identifies three structural moves at companies ahead of the curve:

1. Reorganizing around autonomous agents. Instead of using AI to accelerate human tasks, these companies build workflows where AI agents execute entire decision sequences — ticket triage, lead qualification, contract analysis, financial report generation — with minimal, targeted human oversight.

2. Flattening management layers. When an agent can consolidate data, identify anomalies, and generate recommendations, the role of a middle manager who merely aggregates and relays information loses its purpose. This is not futurist speculation; it is what is happening right now at companies like Klarna, which reduced its customer service headcount from 700 to fewer than 50 people using generative AI while maintaining the same ticket volume.

3. Shorter product cycles and lower cost to serve. With fewer manual handoffs and fewer intermediate reviews, the time between a strategic decision and its execution shrinks. The cost per transaction, per customer served, per bug fixed, falls in a measurable way.

The direct consequence: pressure for measurable return has intensified. It is no longer enough to say "we are exploring AI." The market — and investors, in the case of startups — wants to see the cost line move.

The mistake most SMBs are making right now

I work with SMBs in Brazil, Italy, and the US, and the pattern I see repeated is always the same: the company subscribes to ChatGPT Teams, Copilot 365, or a CRM with built-in AI, trains the team for two days, and declares that it "already has AI." Six months later, adoption has dropped to 20% of users, the process is still the same, and ROI is hard to articulate.

The problem is not the tool. It is that the tool was inserted into a workflow that was never rethought around it.

Imagine a commercial proposal generation process that involved five manual steps — gathering client data, formatting the scope, pricing, legal review, and sending. If you add AI only at the "help write the text" step, you saved perhaps 30 minutes. If you redesign the workflow so that an agent pulls data from the CRM, structures the scope based on validated templates, calculates pricing with embedded business rules, and generates the draft for final review in a single execution, you have gone from five steps spread across two days to a delivery in 20 minutes. That is the leap BCG is describing.

The question every SMB should ask before buying any AI tool

Before signing any software contract, the right question is: which process, if redesigned from scratch with AI at its center, would reduce our operational cost or increase our speed in a way that customers would actually feel?

Not "where can I fit AI in." But "which workflows deserve to be rebuilt?"

What changes in practice for those operating in Brazil

The Brazilian context adds specific layers. Domestic SMBs face rising labor costs, tax complexity that consumes entire teams' hours, and sales cycles that depend heavily on human relationships. This creates both urgency and friction.

The urgency: automating the back office — tax, finance, customer service — has fast, measurable payback. Tools like N8N (workflow automation), Make (formerly Integromat), and agents built on GPT-4o or Claude 3.5 Sonnet APIs are already being used by SMBs to process invoices, respond to collection emails, qualify inbound leads, and generate P&L reports automatically. This is not science fiction — it is a matter of weeks of configuration, not months.

The friction: most SMB owners still lack clarity on which processes are candidates for structural automation versus which require irreplaceable human judgment. Making that triage is, very often, the most valuable work a consultant can offer — even before touching any tool.

The decision that separates who will win from the next wave

The BCG report is not a prediction of a distant future. It is a photograph of the competitive present. The companies growing with better margins in technology and services sectors are not necessarily those with the most capital or the most engineers. They are the ones that decided AI is neither a department nor a tool — it is the operating system of the business.

For SMBs, that decision rarely requires a two-year transformation with eight-figure investment. What it does require is clarity about which two or three core processes, if redesigned with agents at their center, would permanently change the cost structure of the business.

That clarity starts with the right question. And the right question is not "which AI should I buy?" — it is "how does my business need to be operated so that AI is a structural advantage, and not just another cost?"