Banco Santander is extending AI tools to all 185,000 employees worldwide in June 2026 after saying early deployments generated €35 million of value in the first quarter and put the Spanish bank on course for more than €200 million this year. The move turns AI from a specialist productivity experiment into a default layer of work inside one of Europe’s largest banking groups. That is the real news: not that Santander is using chatbots, but that it now believes the economics are visible enough to justify mass access.
For WindowsForum readers, this is more than another “AI transformation” press release from a large enterprise. Santander’s rollout captures the next phase of corporate AI adoption: Microsoft Copilot for everyday office work, multiple frontier models behind the scenes, and hundreds of automation agents aimed at the unglamorous plumbing of regulated business. The question for banks, sysadmins, and software vendors is no longer whether AI can produce a convincing demo. It is whether it can survive contact with compliance, fraud, customer service, and the monthly cost report.
The first wave of enterprise generative AI was heavy on access and light on accountability. Companies bought licenses, announced partnerships, ran training sessions, and let departments discover whether large language models could help with emails, summaries, code, or internal search. Santander’s latest update suggests a more mature posture: AI is being treated as a measurable business system, not merely a workplace perk.
That distinction matters because banks are unusually hostile environments for vague technology narratives. They have legacy systems, regulatory constraints, audited processes, data residency concerns, and customers who do not forgive errors involving money. A bank cannot simply “move fast” if the failure mode is a blocked card, a mishandled fraud claim, or a compliance breach.
Santander says around 40,000 employees are already actively using AI tools. Extending access to the full workforce is therefore not the beginning of adoption; it is the scaling of a model that management believes has crossed an internal threshold. The bank is now telling investors, employees, and competitors that AI has moved from experimentation into execution.
The numbers are deliberately framed in business language. Santander is targeting more than €200 million in AI-driven value in 2026 and more than €1 billion between 2026 and 2028. That value appears to include both cost savings and revenue upside, which is important because the most credible enterprise AI deployments will probably be those that improve margins in several small ways rather than magically replacing entire business units overnight.
But Santander is not presenting this as a single-vendor bet. The bank says its multi-provider approach includes OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, startups, other technology partners, and G42. That mix tells us something important about where enterprise AI is heading: Copilot may be the user interface for many employees, but the model economy underneath is becoming more plural.
This is the same pattern cloud buyers learned years ago. Enterprises may standardize on a primary platform, but they rarely want to be trapped by one provider’s roadmap, pricing, model behavior, or regional constraints. In AI, the risks of lock-in are even sharper because model performance can shift quickly, context windows matter, safety filters vary, and specialized agents may perform better on one stack than another.
For Windows and Microsoft administrators, Santander’s approach is a reminder that Copilot adoption does not end the architectural conversation. Identity, data classification, logging, endpoint management, retention, and conditional access still matter. If anything, AI makes those controls more important because a user’s prompt can become the new place where sensitive business context leaks.
Santander says it has more than 280 process automation agents in production. That phrase deserves scrutiny. “Agent” has become one of the most overloaded words in technology, covering everything from a glorified macro to a semi-autonomous system that can plan, call tools, check outputs, and pass work to another system. In a banking context, the responsible version is likely to be constrained, monitored, and deeply integrated with existing workflow controls.
That is not a weakness. It is the only sane way to deploy AI in finance. The market may be fascinated by autonomous agents that roam across applications, but banks need systems that perform specific jobs reliably, produce audit trails, and know when to hand off to a human.
The Brazil fraud-claims example is especially revealing. Santander says AI has made the process around 95 percent faster, with up to 90 percent automation and an error rate below 1 percent. If those figures hold under sustained production load, they point to an important category of AI value: faster resolution of emotionally charged customer problems that are expensive to handle manually and damaging when delayed.
Those numbers are easy to admire and easy to misunderstand. Customers do not object to automation because it is automated; they object when automation becomes a maze, blocks escalation, or misunderstands the problem. A well-designed AI voice system can reduce waiting, handle routine card queries, and free humans for edge cases. A badly designed one becomes another layer of corporate deflection.
This is where banks have a narrow path. Fraud, cards, and account access are areas where speed matters, but so does confidence. Customers need to know whether they are speaking with a bot, when a human can intervene, and how errors will be corrected. The promise of AI in service channels is not fewer humans at any cost; it is better routing of human attention.
Santander’s framing suggests it understands this, at least in public. The bank talks about giving service teams time back to focus on more complex needs. That is the language every responsible AI deployment should use, though the real test will be staffing decisions over the next several years.
That matters because too much of the AI debate is trapped in a labor-replacement frame. Cost reduction is real, and banks will pursue it aggressively. But the more durable opportunity may come from making existing products smarter, more contextual, and more adaptive.
Payments are a natural proving ground. They generate large volumes of structured and semi-structured data, involve real-time decisions, and sit at the intersection of fraud, customer preference, merchant economics, and regulatory obligations. AI can help identify when to offer a currency choice, how to detect suspicious behavior, or how to reduce friction without increasing risk.
The same logic applies across banking. AI can compress back-office work, but it can also personalize offers, detect churn risk, improve credit workflows, and make digital channels less brittle. The winners will not be the banks that bolt a chatbot onto their website. They will be the banks that rebuild decisioning, servicing, and operations around governed data flows.
This is where AI business cases often get slippery. If a workflow was going to be automated anyway, how much value should be credited to AI? If a department saves time but headcount stays flat, does that count as cost savings, capacity creation, or theoretical productivity? If a customer-service bot deflects calls but worsens satisfaction, does the spreadsheet still look good?
Santander’s use of quarterly value reporting is therefore significant. It imposes a cadence on the AI program. Once executives attach euro amounts to AI initiatives, internal teams must define baselines, measure outcomes, and defend assumptions. That can be uncomfortable, but it is healthier than the alternative: a haze of “innovation” activity with no profit-and-loss consequence.
The bank’s first-quarter figure of €35 million also gives the target a plausible ramp. To exceed €200 million in 2026, Santander needs adoption and use-case deployment to accelerate through the year. That seems possible if the employee rollout, automation agents, and country-level deployments compound. It is not guaranteed, and investors should remain alert to how the bank defines value.
The obvious concerns are familiar: confidential data in prompts, hallucinated outputs, model bias, overreliance, and shadow IT. But the deeper concern is consistency. In a bank, two employees using AI differently in the same workflow can create compliance and customer-treatment problems, even if neither intends to.
Training helps, and Santander has already emphasized AI education across the group. But training alone is not governance. Effective enterprise AI control requires clear approved-use policies, technical guardrails, model access management, logging, retention rules, human review points, and consequences for misuse.
This is where Microsoft’s enterprise footprint becomes strategically useful. Entra ID, Purview, Defender, Intune, Microsoft 365 audit logs, and sensitivity labeling give large organizations a governance foundation that consumer AI tools cannot match by default. But a multi-provider model complicates that picture. Santander will need consistent policy enforcement across Copilot, ChatGPT Enterprise-style environments, Claude, Gemini, internal agents, and partner-built systems.
The upside is flexibility. The bank can route tasks to the most appropriate model, benchmark providers against one another, avoid total dependency on a single vendor, and take advantage of rapid improvements. It can also negotiate from a stronger position if providers know they are not the only option.
The downside is operational complexity. Different models produce different outputs, follow instructions differently, handle context differently, and carry different contractual and data-handling terms. A workflow validated on one model may not behave identically on another. In regulated environments, that matters.
This creates a new role for enterprise architecture. Model selection can no longer be treated as a procurement afterthought. It becomes part of system design, risk management, and operational resilience. Banks will need internal model catalogs, evaluation harnesses, fallback strategies, and change-control processes that look more like software release management than traditional vendor sourcing.
That puts IT administrators in the middle of the transition. They will be asked to enable tools quickly while protecting data, preserving compliance, and keeping costs predictable. The most important AI work in many organizations will be unglamorous: permissions cleanup, SharePoint hygiene, identity governance, endpoint baselines, browser controls, and data-loss prevention.
Copilot and similar tools are only as safe as the information architecture they can see. If employees have excessive access to internal documents, AI can surface that overexposure at conversational speed. The old mess of permissions becomes a new kind of search and summarization risk.
This is why AI readiness is not just a licensing project. Before broad deployment, organizations need to understand what data users can access, which repositories are stale, which labels are enforced, and where sensitive material lives. Santander has the scale and resources to build those controls. Smaller organizations adopting the same tools may discover that the hard part begins after the purchase order.
But banking is a labor-intensive industry, and large-scale automation inevitably raises workforce questions. If AI can automate 90 percent of a fraud-claims process, resolve 40 percent of a call category through self-service, and deploy hundreds of agents across operations, it will eventually affect staffing models. The effect may show up through attrition, slower hiring, redeployment, outsourcing reductions, or direct cuts.
The honest answer is that all of these can be true at once. AI can improve jobs for some employees while reducing demand for others. It can create new governance, analytics, and process-design roles while shrinking repetitive operational work. It can make frontline teams more effective while also giving executives a new lever for cost-to-income improvement.
Santander’s responsibility will be to show whether its “AI for everyone” strategy is genuinely a workforce enablement program or primarily an efficiency program with better branding. The distinction will become clearer as value targets rise toward 2028. Productivity gains eventually demand allocation decisions.
Santander’s claimed error rate below 1 percent in a Brazilian card fraud process is encouraging, but it also illustrates the standard the industry must meet. Error rates need context: what counts as an error, how severe errors are, how often humans review outputs, and how quickly mistakes are corrected. In financial services, not all errors are equal.
This is why the best AI deployments will be boringly engineered. They will constrain model behavior, use retrieval from approved sources, log decisions, require human review above thresholds, and separate recommendation from execution. The goal is not to make the model seem magical. The goal is to make the system dependable.
For customers, the key will be transparency and recourse. If AI affects a fraud claim, a card interaction, or a service outcome, the bank must be able to explain the decision path well enough for support, audit, and dispute handling. “The model said so” will not satisfy regulators or customers.
The phrase “AI-first” can sound like marketing excess. In practice, it usually means something more prosaic: standardizing data, modernizing platforms, connecting workflows, and building feedback loops. The model is the visible layer; the data estate is the machinery underneath.
This is also why AI transformation favors large institutions that can fund long-running modernization programs. Santander can invest across countries and businesses, reuse patterns, centralize expertise, and negotiate with major providers. It can spread the cost of governance over a huge employee base and customer footprint.
That does not mean large banks will automatically win. Their legacy complexity is formidable. But if they can turn AI projects into reusable group-wide capabilities, the economics become powerful. Santander’s announcement is essentially a claim that it has found that repeatable model.
For WindowsForum readers, this is more than another “AI transformation” press release from a large enterprise. Santander’s rollout captures the next phase of corporate AI adoption: Microsoft Copilot for everyday office work, multiple frontier models behind the scenes, and hundreds of automation agents aimed at the unglamorous plumbing of regulated business. The question for banks, sysadmins, and software vendors is no longer whether AI can produce a convincing demo. It is whether it can survive contact with compliance, fraud, customer service, and the monthly cost report.
Santander Moves AI From Pilot Theater to the Operating Model
The first wave of enterprise generative AI was heavy on access and light on accountability. Companies bought licenses, announced partnerships, ran training sessions, and let departments discover whether large language models could help with emails, summaries, code, or internal search. Santander’s latest update suggests a more mature posture: AI is being treated as a measurable business system, not merely a workplace perk.That distinction matters because banks are unusually hostile environments for vague technology narratives. They have legacy systems, regulatory constraints, audited processes, data residency concerns, and customers who do not forgive errors involving money. A bank cannot simply “move fast” if the failure mode is a blocked card, a mishandled fraud claim, or a compliance breach.
Santander says around 40,000 employees are already actively using AI tools. Extending access to the full workforce is therefore not the beginning of adoption; it is the scaling of a model that management believes has crossed an internal threshold. The bank is now telling investors, employees, and competitors that AI has moved from experimentation into execution.
The numbers are deliberately framed in business language. Santander is targeting more than €200 million in AI-driven value in 2026 and more than €1 billion between 2026 and 2028. That value appears to include both cost savings and revenue upside, which is important because the most credible enterprise AI deployments will probably be those that improve margins in several small ways rather than magically replacing entire business units overnight.
The Microsoft Layer Is Only the Front Door
Microsoft Copilot is the most visible part of Santander’s employee-facing rollout, and that is not surprising. For large enterprises already standardized on Microsoft 365, Copilot has the advantage of living where employees spend much of their day: Outlook, Teams, Word, Excel, PowerPoint, and the broader Microsoft identity and compliance stack. For IT departments, that makes it easier to govern than a patchwork of consumer AI accounts and browser extensions.But Santander is not presenting this as a single-vendor bet. The bank says its multi-provider approach includes OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, startups, other technology partners, and G42. That mix tells us something important about where enterprise AI is heading: Copilot may be the user interface for many employees, but the model economy underneath is becoming more plural.
This is the same pattern cloud buyers learned years ago. Enterprises may standardize on a primary platform, but they rarely want to be trapped by one provider’s roadmap, pricing, model behavior, or regional constraints. In AI, the risks of lock-in are even sharper because model performance can shift quickly, context windows matter, safety filters vary, and specialized agents may perform better on one stack than another.
For Windows and Microsoft administrators, Santander’s approach is a reminder that Copilot adoption does not end the architectural conversation. Identity, data classification, logging, endpoint management, retention, and conditional access still matter. If anything, AI makes those controls more important because a user’s prompt can become the new place where sensitive business context leaks.
The Bank’s Strongest Evidence Is Boring, Which Is Why It Matters
The most persuasive parts of Santander’s announcement are not the broad claims about becoming a “data and AI-first bank.” They are the operational examples: card fraud claims in Brazil, AI-supported voice channels in the UK, automation agents in credit, fraud, know-your-customer checks, and back-office operations. These are not glamorous use cases. They are exactly the kind of repetitive, rules-heavy, document-heavy workflows where AI can generate real value if it is tightly governed.Santander says it has more than 280 process automation agents in production. That phrase deserves scrutiny. “Agent” has become one of the most overloaded words in technology, covering everything from a glorified macro to a semi-autonomous system that can plan, call tools, check outputs, and pass work to another system. In a banking context, the responsible version is likely to be constrained, monitored, and deeply integrated with existing workflow controls.
That is not a weakness. It is the only sane way to deploy AI in finance. The market may be fascinated by autonomous agents that roam across applications, but banks need systems that perform specific jobs reliably, produce audit trails, and know when to hand off to a human.
The Brazil fraud-claims example is especially revealing. Santander says AI has made the process around 95 percent faster, with up to 90 percent automation and an error rate below 1 percent. If those figures hold under sustained production load, they point to an important category of AI value: faster resolution of emotionally charged customer problems that are expensive to handle manually and damaging when delayed.
Customer Service Is Where Efficiency Meets Trust
Santander UK’s planned use of AI in voice channels for card-related queries is another practical test. The bank is targeting around 240,000 calls, or roughly 40 percent of annual volume for that category, to be resolved through self-service. It expects that to save customers around 26,000 hours and return about 45,000 hours to service teams for more complex work.Those numbers are easy to admire and easy to misunderstand. Customers do not object to automation because it is automated; they object when automation becomes a maze, blocks escalation, or misunderstands the problem. A well-designed AI voice system can reduce waiting, handle routine card queries, and free humans for edge cases. A badly designed one becomes another layer of corporate deflection.
This is where banks have a narrow path. Fraud, cards, and account access are areas where speed matters, but so does confidence. Customers need to know whether they are speaking with a bot, when a human can intervene, and how errors will be corrected. The promise of AI in service channels is not fewer humans at any cost; it is better routing of human attention.
Santander’s framing suggests it understands this, at least in public. The bank talks about giving service teams time back to focus on more complex needs. That is the language every responsible AI deployment should use, though the real test will be staffing decisions over the next several years.
Payments Show Why AI Is Not Just a Cost-Cutting Machine
The Getnet example broadens the story beyond productivity. Santander says its payments unit is using AI to improve experiences for international customers and merchants, including scenarios where customers paying abroad may prefer to pay in their home currency. The business case is not just faster processing; it is better conversion, better merchant outcomes, and more advanced cross-border payment services.That matters because too much of the AI debate is trapped in a labor-replacement frame. Cost reduction is real, and banks will pursue it aggressively. But the more durable opportunity may come from making existing products smarter, more contextual, and more adaptive.
Payments are a natural proving ground. They generate large volumes of structured and semi-structured data, involve real-time decisions, and sit at the intersection of fraud, customer preference, merchant economics, and regulatory obligations. AI can help identify when to offer a currency choice, how to detect suspicious behavior, or how to reduce friction without increasing risk.
The same logic applies across banking. AI can compress back-office work, but it can also personalize offers, detect churn risk, improve credit workflows, and make digital channels less brittle. The winners will not be the banks that bolt a chatbot onto their website. They will be the banks that rebuild decisioning, servicing, and operations around governed data flows.
The €1 Billion Target Is a Management Discipline Test
Santander’s three-year AI value target is ambitious enough to attract attention but not so wild that it belongs in a venture-capital pitch deck. A global bank with 185,000 employees, millions of customers, and sprawling operations has plenty of process waste to attack. The harder part is proving that claimed value is incremental, measurable, and not merely a rebranding of ordinary digitization.This is where AI business cases often get slippery. If a workflow was going to be automated anyway, how much value should be credited to AI? If a department saves time but headcount stays flat, does that count as cost savings, capacity creation, or theoretical productivity? If a customer-service bot deflects calls but worsens satisfaction, does the spreadsheet still look good?
Santander’s use of quarterly value reporting is therefore significant. It imposes a cadence on the AI program. Once executives attach euro amounts to AI initiatives, internal teams must define baselines, measure outcomes, and defend assumptions. That can be uncomfortable, but it is healthier than the alternative: a haze of “innovation” activity with no profit-and-loss consequence.
The bank’s first-quarter figure of €35 million also gives the target a plausible ramp. To exceed €200 million in 2026, Santander needs adoption and use-case deployment to accelerate through the year. That seems possible if the employee rollout, automation agents, and country-level deployments compound. It is not guaranteed, and investors should remain alert to how the bank defines value.
The Real Governance Problem Is Now Scale
Giving AI access to every employee changes the risk profile. A small pilot can be monitored closely. A 185,000-person deployment becomes a behavior-management challenge across countries, roles, languages, regulatory regimes, and risk appetites.The obvious concerns are familiar: confidential data in prompts, hallucinated outputs, model bias, overreliance, and shadow IT. But the deeper concern is consistency. In a bank, two employees using AI differently in the same workflow can create compliance and customer-treatment problems, even if neither intends to.
Training helps, and Santander has already emphasized AI education across the group. But training alone is not governance. Effective enterprise AI control requires clear approved-use policies, technical guardrails, model access management, logging, retention rules, human review points, and consequences for misuse.
This is where Microsoft’s enterprise footprint becomes strategically useful. Entra ID, Purview, Defender, Intune, Microsoft 365 audit logs, and sensitivity labeling give large organizations a governance foundation that consumer AI tools cannot match by default. But a multi-provider model complicates that picture. Santander will need consistent policy enforcement across Copilot, ChatGPT Enterprise-style environments, Claude, Gemini, internal agents, and partner-built systems.
Multi-Model Banking Is Powerful and Messy
Santander’s willingness to work across OpenAI, Anthropic, Google, G42, startups, and other partners reflects a pragmatic view of the AI market. No single model family is best at everything. Some systems may be better for coding, some for summarization, some for multilingual customer service, some for tool use, and some for controlled enterprise deployments in particular jurisdictions.The upside is flexibility. The bank can route tasks to the most appropriate model, benchmark providers against one another, avoid total dependency on a single vendor, and take advantage of rapid improvements. It can also negotiate from a stronger position if providers know they are not the only option.
The downside is operational complexity. Different models produce different outputs, follow instructions differently, handle context differently, and carry different contractual and data-handling terms. A workflow validated on one model may not behave identically on another. In regulated environments, that matters.
This creates a new role for enterprise architecture. Model selection can no longer be treated as a procurement afterthought. It becomes part of system design, risk management, and operational resilience. Banks will need internal model catalogs, evaluation harnesses, fallback strategies, and change-control processes that look more like software release management than traditional vendor sourcing.
Windows Shops Should Watch the Admin Story, Not the Hype Reel
For the WindowsForum audience, the Santander rollout is a useful case study because it shows how AI enters the enterprise through familiar channels. The average employee will not experience this as an abstract model marketplace. They will experience it as Copilot in Microsoft 365, AI-assisted service tools, workflow automation, smarter CRM fields, and internal agents embedded into line-of-business systems.That puts IT administrators in the middle of the transition. They will be asked to enable tools quickly while protecting data, preserving compliance, and keeping costs predictable. The most important AI work in many organizations will be unglamorous: permissions cleanup, SharePoint hygiene, identity governance, endpoint baselines, browser controls, and data-loss prevention.
Copilot and similar tools are only as safe as the information architecture they can see. If employees have excessive access to internal documents, AI can surface that overexposure at conversational speed. The old mess of permissions becomes a new kind of search and summarization risk.
This is why AI readiness is not just a licensing project. Before broad deployment, organizations need to understand what data users can access, which repositories are stale, which labels are enforced, and where sensitive material lives. Santander has the scale and resources to build those controls. Smaller organizations adopting the same tools may discover that the hard part begins after the purchase order.
The Jobs Debate Is Coming Whether Santander Wants It or Not
Santander’s announcement emphasizes employee support, better customer service, and operational efficiency. That is the expected language, and some of it is likely true. AI can remove tedious work, reduce customer wait times, and help staff handle complex cases more effectively.But banking is a labor-intensive industry, and large-scale automation inevitably raises workforce questions. If AI can automate 90 percent of a fraud-claims process, resolve 40 percent of a call category through self-service, and deploy hundreds of agents across operations, it will eventually affect staffing models. The effect may show up through attrition, slower hiring, redeployment, outsourcing reductions, or direct cuts.
The honest answer is that all of these can be true at once. AI can improve jobs for some employees while reducing demand for others. It can create new governance, analytics, and process-design roles while shrinking repetitive operational work. It can make frontline teams more effective while also giving executives a new lever for cost-to-income improvement.
Santander’s responsibility will be to show whether its “AI for everyone” strategy is genuinely a workforce enablement program or primarily an efficiency program with better branding. The distinction will become clearer as value targets rise toward 2028. Productivity gains eventually demand allocation decisions.
Banking AI Will Be Judged by Its Failure Modes
The most dangerous mistake in enterprise AI is judging it only by successful demos. In banking, the more important question is what happens when the system is wrong, uncertain, outdated, manipulated, or overconfident. A model that performs well 99 percent of the time can still create serious risk if the remaining 1 percent lands in fraud, credit, compliance, or customer vulnerability.Santander’s claimed error rate below 1 percent in a Brazilian card fraud process is encouraging, but it also illustrates the standard the industry must meet. Error rates need context: what counts as an error, how severe errors are, how often humans review outputs, and how quickly mistakes are corrected. In financial services, not all errors are equal.
This is why the best AI deployments will be boringly engineered. They will constrain model behavior, use retrieval from approved sources, log decisions, require human review above thresholds, and separate recommendation from execution. The goal is not to make the model seem magical. The goal is to make the system dependable.
For customers, the key will be transparency and recourse. If AI affects a fraud claim, a card interaction, or a service outcome, the bank must be able to explain the decision path well enough for support, audit, and dispute handling. “The model said so” will not satisfy regulators or customers.
The AI-First Bank Is Really a Data-First Bank
Santander’s own language pairs AI with data, and that pairing is not incidental. AI value depends on data quality, data access, and data governance. A bank with fragmented customer records, inconsistent process definitions, and unmanaged document stores will struggle to get dependable results from even the best models.The phrase “AI-first” can sound like marketing excess. In practice, it usually means something more prosaic: standardizing data, modernizing platforms, connecting workflows, and building feedback loops. The model is the visible layer; the data estate is the machinery underneath.
This is also why AI transformation favors large institutions that can fund long-running modernization programs. Santander can invest across countries and businesses, reuse patterns, centralize expertise, and negotiate with major providers. It can spread the cost of governance over a huge employee base and customer footprint.
That does not mean large banks will automatically win. Their legacy complexity is formidable. But if they can turn AI projects into reusable group-wide capabilities, the economics become powerful. Santander’s announcement is essentially a claim that it has found that repeatable model.
The Santander Signal for Every Copilot Rollout
Santander’s expansion offers a preview of what large-scale enterprise AI will look like when it leaves the keynote stage. It is not one chatbot, one vendor, or one killer app. It is a layered operating model in which productivity tools, automation agents, customer-service systems, payment intelligence, and compliance controls all have to work together.- Santander is extending AI access from roughly 40,000 active users to its full 185,000-person workforce.
- The bank says AI generated €35 million of value in the first quarter of 2026 and is on track for more than €200 million this year.
- Microsoft Copilot is the everyday productivity layer, but Santander is also using a multi-provider strategy that includes OpenAI, Anthropic, Google, G42, startups, and other partners.
- The most concrete gains are coming from process automation, fraud claims, customer-service routing, and payments rather than from generic chatbot usage.
- The biggest risks now shift from experimentation to governance, including data access, model consistency, auditability, employee training, and cost control.
- The long-term workforce impact remains unresolved because productivity gains in banking almost always become staffing and operating-model questions.
References
- Primary source: Finextra Research
Published: 2026-06-21T23:03:11.979285
Loading…
www.finextra.com - Related coverage: santander.com
Loading…
www.santander.com - Related coverage: computerweekly.com
Loading…
www.computerweekly.com - Related coverage: resultsense.com
Loading…
www.resultsense.com - Related coverage: zonebourse.com
Loading…
www.zonebourse.com - Related coverage: bloomberg.com
Loading…
www.bloomberg.com