Observador: Turning subscription sales into conversations with AI

Project name: Subscriptions Concierge

Newsroom: Observador (Portugal)

Solution: Conversational AI agent engaging readers on WhatsApp to negotiate and personalise subscription offers

Organisation staff size: 101 - 200 staff

Tech stack: WhatsApp Business API, AI-assisted conversational logic, custom CRM data pipelines

Impact: Positive engagement and conversion signals with high audience acceptance of transparent AI interactions

Core lesson: Audiences respond well to transparent AI interactions, provided human oversight approves binding decisions


Observador launched in 2014 as a digital disruptor to Portugal's legacy newspapers. Twelve years on, it has grown from 40 to 170 people, runs a 24-hour FM talk radio station and is the country's leading podcast hub. Its latest experiment asks whether an AI agent, chatting with readers one-to-one on WhatsApp, can persuade them to subscribe.

The problem: A small market and the limits of the phone call

Observador has four roughly equal revenue streams: subscriptions, display advertising, sponsored content and audio advertising. Subscriptions matter most. “The one that is most resilient to market fluctuations and to news cycles is of course subscriptions,” says Leo Xavier, Chief Technology and Innovation Officer at Observador.

That resilience is hard-won. Observador came close to breaking even midway through its life before choosing to invest in launching an FM radio station. It is now nearing sustainability again, while many Portuguese outlets have had to take drastic measures to survive.

The market, however, is small. Purchasing power is lower than in much of Europe, and Portugal's entire market for news subscriptions amounts to roughly 150,000 subscribers, shared among three main players. Xavier believes mass, undifferentiated communication is reaching a plateau, and that selling subscriptions is becoming “much more of a personal conversation”. The key question, he says, is: “What is it in your usage that justifies paying €6 (£5) a month, €7 (£6) a month for a subscription?”

Outsourced phone marketing delivers results but has clear limits. “It's a bit invasive. It's ringing someone up,” says Xavier, and that risks damaging the brand. External callers also lack Observador's data and follow a script. Text messaging offered a calmer, complementary channel, letting customers reply “when and if they want and at the pace that they're comfortable with”.

Building the solution: The subscriptions concierge

Observador had prototyped this idea a year earlier, but it stalled. “We prototyped it, it was fun, looked great, but then it died,” Xavier recalls. The JournalismAI Innovation Challenge enabled them to “bet fully on it and get it done quickly”.

The result, the subscriptions concierge, was built from scratch with strong support from AI-assisted coding and runs around the clock, sandboxed from its main site and with full engineering review and oversight. It takes a list of target customers and gathers what Observador knows about each: whether they use the app or read opinion pieces, which plans and campaigns they have had and their renewal price. It combines this with a prompt that the subscriptions team owns and keeps refining.

The concierge sends a single WhatsApp message that always identifies it as an AI agent. If the customer does not reply, there is no follow-up. The tone is deliberately playful, almost teasing: “Hey, why don't you push? I'm an AI agent. I'm quite capable.”

Once a customer replies, the conversation is free-form, with a long prompt full of guidelines and guardrails steering the AI towards understanding the customer's needs and securing a subscription or upgrade. Within set rules, it can even negotiate on price when cost is the sticking point.

Humans stay in control. The subscriptions team picks the audience and prompt, launches campaigns and can take over any conversation. The AI never completes a sale itself: it asks the customer to explicitly confirm the agreed terms, and a staff member reviews the summary before anything becomes binding.

The main obstacle was the channel. WhatsApp's rules require the all-important opening message to follow approved templates, limiting personalisation. SMS has no such limits, but in Portugal it is barely used for conversations any more. The team is working on a verified WhatsApp badge, which should lift response rates a little.

The opportunities: An audience open to AI

Beyond the usual KPIs, such as conversion rates and cost per contact, an initially secondary metric became the biggest finding: how many people engaged positively with the first message. Fewer than 20 per cent responded, but complaints were negligible, despite what Xavier calls “a very vocal customer base”. Uninterested readers replied politely, and anyone asking not to be contacted again was removed immediately. “No one said, why are you using AI?” he says.

That acceptance opens the door to new uses, which is why the tool is called a concierge rather than a sales bot. Next, Observador plans to contact readers who abandon the checkout page, a valuable moment when speed matters and nobody can phone them in the middle of the night. Further ahead, they want to use conversations for onboarding and engagement, checking in midway through a subscription to suggest unused features or hear concerns.

Lessons for newsrooms

  • Test your assumptions about AI acceptance: Observador's readers engaged with an agent that declared itself as AI from the first message, and almost none objected. Transparency, a light tone and backing off when asked appear to have helped earn that tolerance.

  • Keep humans accountable: Let AI hold the conversation, but ensure a person can step in at any time and sign off on anything binding.

  • Clear thinking is the new bottleneck: As AI speeds up coding, "the bottleneck stops being code," says Xavier, and becomes the thinking behind it. Briefing an AI poorly, like briefing an engineer poorly, exposes vague goals, only faster. Invest in experienced people who can set a strategy and document intent precisely.

Explore case studies from the 2025 cohort

The 2025 Innovation Challenge Report, Prompting Sustainability, collects case studies from all 12 grantees and reveals their journeys, challenges, and opportunities.

Read about their experiences, discover best practices, and find inspiration in their innovative approaches to integrating AI into journalism.


The JournalismAI Innovation Challenge, supported by the Google News Initiative, is organised by the JournalismAI team at Polis – the journalism think-tank at the London School of Economics and Political Science, and it is powered by the Google News Initiative.