Santander, BBVA and Sabadell are already putting AI to work, and what is changing is that agents stop merely answering and start executing tasks. Santander cites more than 50% lower customer-service costs in an experimental case; BBVA is joining Poppy, a protocol for AI agents to operate with safe permissions; and Sabadell has agentic AI integrated at scale in its systems and processes. We summarise what our carousel reports, based on sources published between 14 and 19 October 2026, and what a company that is not a big bank can take from it.
This article expands on our LinkedIn carousel. If you are already thinking about agents inside your organisation, it pairs well with the agentic company: how to govern your AI agents and the 8 decisions before you let an agent act.
What figure does Santander give for customer service?
According to Banco Santander’s press release of 19 October 2026, an experimental case with NVIDIA shows more than 50% lower customer-service costs with AI. The carousel sums it up as an index: 100 before, under 50 with AI. The description of the case mentions models specialised by process and higher-quality conversations.
Read the figure carefully: it is an experimental case from a single bank, not a general result, and the carousel does not detail the methodology. What is useful for your company is not copying the 50%, but the underlying idea: specialise the model by process and measure the cost per conversation before and after.
What are BBVA and Sabadell doing?
The carousel sums up the change as a move from answering questions to executing tasks: before, a question got an answer; now, a request reaches an AI agent and comes back as a finished task. According to El Plural (14 October) and Computing (19 October), both banks are moving in that direction:
- BBVA: is joining Poppy, a protocol for AI agents to operate with safe permissions.
- Sabadell: agentic AI integrated at scale in its systems and processes.
The carousel gives no further detail on Poppy or on the scope of Sabadell’s rollout, so we do not fill in the gaps.
Why does moving from answering to executing matter?
Because the control question changes. An assistant that only answers is judged on the quality of its answer; an agent that executes tasks is also judged on what it can touch and with which permissions. That is our own reading, but it fits with BBVA joining a protocol designed precisely for agents to operate with safe permissions.
In practice, before you give an agent access to your tools, decide what it can do on its own, what needs approval and what it must never do.
In which four areas can you measure AI in your company?
You do not need to be a big bank to start. The carousel proposes four areas where AI can be measured, each with an example and an indicator:
- Customer service (AI agents): resolves repetitive queries 24/7 and hands only the complex cases to a person. Indicator: lower cost per conversation.
- Sales (sales automation): replies to every lead within minutes, qualifies it and books the meeting. Indicator: lead response time.
- Operations (intelligent workflows): orders and invoices move from email to your ERP on their own, with no typing. Indicator: hours saved.
- Documentation (extraction and analysis): reads contracts and case files, extracts the key data and summarises it. Indicator: time per case file.
An important caveat: these are KPIs proposed by JAV BRAIN, not client results.
How do you start putting AI to work in your company?
This is JAV BRAIN’s approach, not data from the banks:
- 1. We identify the processes that gain most from AI.
- 2. We design and integrate agents into your tools.
- 3. We measure results and scale what works.
If you start on your own, the same order holds: one process, one indicator and a before-and-after comparison, like the 100 versus under 50 index the carousel uses to sum up Santander’s case.
Which process would you automate first?
That is our closing question: the big banks are already making their move, so when is it your company’s turn? Tell us which process you would put to work with AI first.
Related. If you are interested in the productivity, savings and compliance side, read AI in Spanish Business: Productivity, Savings and Compliance.
Frequently asked questions
What cost saving does Santander cite for AI in customer service?
More than 50% lower customer-service costs, in an experimental case with NVIDIA, according to Banco Santander’s press release of 19 October 2026. It is an experimental case, not a general result.
What is Poppy and what does it have to do with BBVA?
According to the carousel, BBVA is joining Poppy, a protocol for AI agents to operate with safe permissions. The cited sources are El Plural (14 October) and Computing (19 October); the carousel gives no further detail.
Can a company that is not a bank use AI agents?
Yes, it can get started: the carousel’s message is that you do not need to be a big bank. Pick one process, define an indicator (for example, cost per conversation in customer service) and measure the result before and after.
Which process in your company would you put to work with AI first? We identify the processes that gain most and integrate agents into your tools through our AI agents and automation service. Book a free exploratory session and we will look at your case.
More analysis like this in the original LinkedIn post.
