Lumıary
Vendor reportedLive since Sep 2026Added Oct 2026

Bizum

Bizum launches a generative AI assistant for customer support on its website and WhatsApp

Spanish payments service Bizum worked with Telefonica Tech to replace pre-programmed support answers with a generative AI assistant on its website and WhatsApp Business. Telefonica reports 90% accuracy in answers and 30% fewer email contact requests, with the project completed in three months.

Improve customer experienceCut costs

Key figures

Copied exactly from the sources. Copy any figure with its citation.

90%

Accuracy of the assistant's answers, as reported in the press release1

30%

Reduction in email contact requests for questions or help, as reported in the press release1

What it took

CostNot disclosed
TeamNot disclosed
TimelineCompleted within three months1

The problem

Bizum is the Spanish mobile payments service that works through the country's banks. Its customer support relied on pre-programmed answers. Customers still wrote in by email with questions about incidents, general queries and possible scams. Bizum wanted support that was more efficient, more up to date and available on more than one channel.

What they built

Bizum put a virtual assistant based on generative AI on its website and on WhatsApp Business. Generative AI is software that writes answers in natural language. The assistant works alongside Bizum's staff and is available 24 hours a day, every day. It answers questions about incidents, general queries and possible scams.

How they did it

Telefonica, through its Telefonica Tech unit, developed the assistant on its own generative AI platform. The design uses several assistants coordinated together, each handling a specific type of question, and draws on documentation provided by Bizum as its knowledge source. Telefonica also built a connection (an API) that looks up bank contact details, so questions that reach Bizum but belong to a bank can be redirected. For payment incidents, the assistant can give the user the right bank's details and even send an email on the user's behalf. Telefonica says it completed the project within three months.

Results

Telefonica reports that Bizum reached 90% accuracy in the assistant's answers and cut email contact requests by 30%. These are the vendor's own figures and the release does not explain how they were measured.

Lessons for your project

  1. Ground the assistant in your own documentation. Bizum feeds it material it supplies, which keeps answers consistent with official guidance.

  2. Split the work between several narrow assistants. Bizum uses several, each focused on a topic such as incidents, general questions or scams.

  3. Route what is not yours to the right owner. The assistant looks up the contact details of the relevant bank and passes payment issues on.

  4. Track both answer quality and ticket volume. Bizum reports accuracy and a drop in email requests, which show quality and workload effects.

Caveats

  • The only source is a press release from Telefonica, which built the assistant. The 90% and 30% figures are its claims, with no method, baseline period or independent check given.
  • The release is in Spanish and gives no go-live date. The deployed date shown is the month of the release, and the metric and timeline wording here is our English rendering.
  • The language model behind the assistant is not named.

Sources

  1. Telefonica crea un asistente virtual para modernizar la atencion al cliente de Bizum

    Telefonica · 2026-09-08 · Press release · telefonica.com

Summarized in our own words from the sources above. Spotted an error? See corrections.

Related cases