Analysis · AI for Business
AI agents are coming to your back office. Here is what is real.
The software you already pay for is about to start acting on its own. That part is real. So is the failure rate. The businesses that win will not be the ones that adopt the most AI, they will be the ones that redesign the work.
Key takeaways
- AI agents are being built into the software SMBs already pay for. Gartner expects them in 40% of enterprise apps by the end of 2026, up from under 5% in 2025.
- Adoption is not the advantage. 88% of organisations use AI, but only 39% report any enterprise profit impact (McKinsey, 2025).
- Most agent projects will fail. Gartner expects over 40% to be cancelled by the end of 2027.
- Small businesses are early, roughly 18% adoption, so the opening is genuinely open.
- What works is redesigning one workflow end to end and owning it, not bolting on a chatbot.
The agents are arriving whether you ask or not
You will not have to go looking for AI agents. They are being built into the tools you already use. Gartner expects task-specific AI agents to feature in 40% of enterprise applications by the end of 2026, up from less than 5% in 2025. Your accounting tool, your CRM, your online store: the software is learning to act, not just to store.
What this means in practice is that agent capability is arriving as a default setting rather than a project. That is good, because features you once paid a developer for now come bundled. It is also a trap, because a feature switched on inside someone else's software is not a system you control, and control is where the value sits.
Using AI is not the same as getting value from it
Here is the number that should reframe the whole conversation. In McKinsey's 2025 global survey, 88% of organisations said they regularly use AI in at least one part of the business. Almost everyone. Yet only about a third have scaled it across the company, and just 39% report any measurable impact on enterprise profit, most of them under 5%.
Read that gap slowly. Access to the technology is nearly universal; measurable financial return is rare. The bottleneck is not the model. It is the work around the model: the process was never redesigned, so the AI was bolted onto a workflow that still assumes a human in the loop at every step.
Small businesses are early, and that is the opportunity
Adoption among small businesses is still low and climbing fast: roughly 18% were using AI by the end of 2025, up from about 5% two years earlier, according to the JPMorgan Chase Institute's analysis of business banking data, which lines up with the US Census Bureau's own survey.
The honest read for a Portuguese SMB is that you are not behind, you are early. Early is where the cheap wins are, before every competitor has automated the same task and the advantage is priced away.
Different perspectives
Agents genuinely compress back-office cost. Routine, high-volume tasks, invoicing, order confirmation, stock updates, first-line support, are exactly what current agents do well, and they arrive embedded in tools SMBs already own. For a small team, that is real capacity without real hiring.
The same technology has a high failure rate. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, killed by rising costs, unclear value and weak controls. McKinsey's data agrees from the other side: only about 6% of organisations are genuine high performers. Agents are easy to start and hard to land, and a demo is not a deployment.
Comparison
Three ways a small business can respond to AI agents
| Approach | Time to value | You own it | Ongoing cost | Best for |
|---|---|---|---|---|
| Bolt a chatbot onto one channel | Days | No | Per-seat subscription | A quick win on a single channel |
| Wait for your software to add agents | Months to years | No | Bundled and rising | Low effort, low control |
| Redesign one workflow and build the system | 2 to 4 weeks | Yes | Low, and it is yours | Durable efficiency that compounds |
Our view
We build these systems, so here is the version without the hype. The winners in the McKinsey data are not the ones who bought the most AI. They are the roughly 6% who redesigned a workflow around it and owned the result. For a small business that is not a strategy deck, it is one job done properly: pick a single high-volume task that leaks time, rebuild it so the system does the work rather than a person, measure it, then move to the next one.
That is exactly what we did for A Batina, a 32-year-old retailer: one system across POS, online store and stock, with invoicing automated. The outcome was not a chatbot. It was 10 hours a week given back to the team, customer acquisition up 15%, and €200 a month in software they stopped paying for. Cut the busywork. Build the system. Keep the growth.
What to do
- Pick one high-volume task that clearly leaks time: invoicing, stock, follow-ups, order confirmations.
- Map it end to end and redesign it so the system does the work, not a person.
- Measure the before, hours, errors, cost, so the after is provable.
- Own the result, then repeat on the next task rather than buying ten tools at once.