There is no faster way to fill a room in Lisbon right now than to put "AI" in the event title. In the space of eighteen months, the city has gone from a handful of practitioners doing serious applied AI work to a market flooded with consultants, agencies, and advisory firms, all of whom will tell you, with great confidence, that they can transform your business with artificial intelligence.
Some of them can. A meaningful number cannot. The difference between the two groups is not always obvious from the outside - both tend to produce impressive presentations, speak fluently about large language models, and reference the same set of case studies. Here's how to tell them apart.
The gold rush dynamic
Every major technology shift creates a gold rush, and AI is no exception. The pattern is consistent: a genuinely transformative technology emerges, demand for expertise outpaces supply by an enormous margin, and the gap fills quickly with people who have read about the technology rather than built with it. This is not cynicism - it's the normal market response to a real shift. The problem is that clients who engage during the gold rush phase often spend significant budget on advice rather than outcomes.
In Lisbon specifically, the pattern has been accelerated by the city's growing reputation as a tech hub and the influx of international businesses looking for local AI partners. The demand signal has been strong enough to attract providers of every capability level, and the market hasn't yet matured to the point where reputation and track record do the sorting work reliably.
What buzzword bingo actually looks like
The tell-tale signs of an AI consultancy that is selling slides rather than solutions are consistent across providers. Watch for:
- Strategy deliverables as the end product. A provider whose engagement model ends at the strategy deck - no build, no integration, no handover of working software - is a provider that cannot be held accountable for whether the strategy works. Frameworks and roadmaps have value, but they are inputs to implementation, not outcomes.
- Use case catalogues without specificity. Presenting a list of "AI use cases for your industry" without engaging with your specific data, systems, and operational constraints is not consulting - it's content marketing delivered in a meeting room.
- Model name-dropping without architecture thinking. A provider who talks primarily about which AI models they use (GPT-4, Claude, Gemini) without discussing the systems and data pipelines that need to exist for those models to be useful hasn't thought seriously about implementation.
- No prior work they can show you running. This is the equivalent of the portfolio test for design agencies. If a provider cannot show you AI systems they have built and deployed - not prototypes, not demos, but production systems handling real business operations - treat their capability claims with significant scepticism.
What a genuine AI opportunity audit looks like
Before building anything, a credible AI partner should help you understand where in your operations AI can generate real, measurable return - and where it cannot. This is different from telling you that AI is generally transformative. It requires engaging with the specifics of your business.
A proper opportunity audit maps your existing processes, identifies where repetitive, high-volume, or pattern-dependent tasks exist, assesses the data you have available to train or prompt AI systems effectively, and produces a prioritised view of which use cases offer the highest ROI relative to implementation complexity. The output should be specific enough that you could brief an engineering team on what to build - not a set of inspiring possibilities, but an ordered list of concrete problems with a proposed technical approach for each.
The question we ask in every audit is not "where could AI help?" - it's "where does the cost of human time or the quality ceiling of human execution create a bottleneck that AI can specifically address?" Those are different questions, and the second one has a much shorter list of honest answers.
ChatGPT wrappers vs. real AI infrastructure
This is the distinction that separates a significant proportion of what's being sold as "AI implementation" from what serious practitioners would recognise as infrastructure. A ChatGPT wrapper is an interface that sends user inputs to a large language model API and returns the output. It can be built in a day, it does not require deep AI expertise, and it does not create durable competitive advantage because anyone can build the same thing.
Real AI infrastructure involves: properly structured data pipelines that give AI systems the specific, current, and accurate information they need to operate effectively; retrieval-augmented generation systems that connect language models to proprietary business knowledge; evaluation and monitoring systems that track output quality and flag degradation; and integration with existing business systems so that AI-generated outputs flow into actual workflows rather than existing in isolation. This is engineering work, not consulting work, and it requires people who build software for a living. It's the same discipline we bring to AI and automation work, and to using AI in web projects where the model is one component of a larger system rather than the whole product.
Use cases generating real ROI in Portugal right now
Based on implementations we're running or have completed with Portuguese businesses, the use cases that are generating measurable, auditable return in 2025 are more mundane than the conference presentations suggest - and more valuable:
- Customer support automation. For businesses handling high volumes of Portuguese and English language enquiries, AI triage and response systems that handle tier-one queries without human intervention are generating 30–50% reductions in support cost with measurable customer satisfaction improvements when implemented with proper escalation design.
- Content operations at scale. Not AI-generated content published without review, but AI-assisted content workflows where first drafts, metadata, social adaptations, and Portuguese/English variants are generated automatically and reviewed by a single editor. Teams that ran this at 20 pieces per month are running it at 80. This is part of a broader shift in how AI is reshaping digital marketing.
- Internal knowledge systems. Portuguese businesses with significant institutional knowledge trapped in PDFs, email threads, and the heads of long-tenure employees are building internal AI assistants that make that knowledge accessible to new hires and junior staff in a way that training programmes never managed.
- Sales intelligence. AI systems that research prospects, surface relevant context before sales calls, and generate personalised outreach are reducing the preparation time on each opportunity and improving the quality of the first conversation.
Questions to ask a potential AI partner
The interrogation that separates credible providers from slide merchants is direct. Ask:
- Can you show me a production AI system you've built and deployed, with the client contact I can call?
- What does your discovery process look like before you recommend a technical approach?
- How do you evaluate whether an AI system is working correctly once it's in production?
- What happens when the AI produces a wrong or harmful output - what systems do you build to catch that?
- What do you charge, and what does the client own at the end of the engagement?
The last question matters particularly. Consultancies that deliver strategy docs own nothing at handover - the intellectual property is in the document, which you paid for, but the expertise required to execute it remains with the consultancy. Build-oriented partners leave you with running software, documented systems, and engineering decisions that a future team can maintain and extend.
Why we approach it as a product team, not a consultancy
Incremento Labs was built on the premise that the advice-implementation gap in AI consultancy is the central problem worth solving. We don't produce AI strategies as standalone deliverables. We build the systems, run them until they're stable, and then hand over something working. Our discovery process is structured like a product sprint - specific, time-boxed, and oriented toward a build decision rather than a strategy document.
The businesses that get the most from AI engagements are the ones that treat it as a product problem - "what should we build, for whom, and how will we know it's working?" - rather than a strategy problem. That framing changes everything: who you hire, what success looks like, and how quickly you find out whether you were right.
The AI opportunity in Lisbon and Portugal is real. The noise around it is real too. The way through the noise is the same as it always is: ask for working evidence, not polished presentations.