Apps that understand and act, designed around AI first.

Apps where AI reads, sees, listens and acts for the user. One plan for web, iOS and Android, tested on every release.

Why a chat window on the old app is not enough

Most companies added AI to their apps as a separate chat window. The forms and steps behind it stayed the same. Customers still retype data you hold and wait for a callback. The assistant can explain the process, not shorten it.

The models underneath have changed. They read scanned documents, interpret photos, transcribe speech and call your systems. So the user names a goal, the app does the work: it gathers what it needs, proposes the next step and acts once the user confirms.

Building this is harder than adding a feature. Product, design, engineering and AI testing need one plan and clear limits on what the app may do alone. Without measuring quality on every release, pilots stay pilots and the core app never changes.

What we build

  • Product strategy and design

    We pick the journeys where AI changes the outcome, then design what the app does and what it only suggests.

  • Web, iOS and Android

    One product plan, one design system and shared logic across platforms, native or cross-platform as each product needs.

  • Documents, photos and speech

    The app reads contracts, invoices and forms, understands photos and speech, and checks what it extracts against your records.

  • Personalisation from your data

    Recommendations and next steps based on each customer's history and products, drawn from your CRM and core systems, never guessed.

  • Actions on the user's behalf

    The app changes limits, books appointments and files claims. The user confirms anything touching money, contracts or personal data.

  • Evaluation and safety built in

    Tests built from real cases run on every release, with guardrails and live tracking of quality, completion, escalations and cost.

How it works

One request, from the user's first input to a completed, logged outcome.

  1. The user states a goal

    Typed, spoken, a photo or a document, in the user's language. The app works out what they want.

  2. The app gathers context

    It reads the document, pulls the customer's data from your systems and checks what the user may do.

  3. It proposes the next step

    It shows an answer or a prepared action to check. Unclear or out-of-limit cases go to your team.

  4. It acts with confirmation

    After confirmation, the app completes the order, change, claim or booking with the user's permissions, and logs it.

  5. Every outcome is measured

    Quality, completion and cost are tracked per journey. Every failure becomes a test for the next release.

How we roll it out

  1. Discovery

    3 to 4 weeks

    We pick the journeys where AI moves a business number and agree what the app may do on its own.

    • Prioritised journey map with target metrics
    • Clickable prototype of the first journey
    • Architecture, data and risk assessment
  2. First release

    10 to 14 weeks

    One journey built end to end on web and mobile, on your systems, released to a defined user group first.

    • Production release on web, iOS and Android
    • Evaluation suite and quality dashboard
    • Runbooks for support and escalation
  3. Scale and run

    Ongoing

    We add journeys in short cycles, widen access as quality holds, and hand over as much as you want.

    • Release plan tied to business targets
    • Monthly quality, usage and cost reporting
    • Handover and training for your teams

In production elsewhere

Public examples from other organisations, not unus clients, showing AI in the core of an app, at scale.

Public case studyCommerzbankBanking · Germany · 2025 Banking app assistant that completes credit card tasks 30,000+ conversations a month
The situation
Commerzbank wanted to modernise digital support and unify outdated systems while meeting regulatory, fraud protection and transparency rules. Customers expected instant, personal service, and routine requests like card blocking or limit changes had to work at any hour.
What they deployed
Ava, an AI agent in the Commerzbank app, runs on Azure OpenAI, grounded in bank data via Azure AI Search. It handles credit cards from ordering to blocking, filters sensitive data, takes spoken requests and escalates to a human.
The outcome
According to Microsoft's customer story, Ava manages more than 30,000 customer conversations per month and resolves 75% of requests autonomously.

An app assistant earns its place when it can complete regulated transactions inside the conversation, with clear limits and a route to a person.

Source: Microsoft Customer Stories, Commerzbank AG fuels 30,000 monthly conversations with Foundry Agent Service, November 2025
Public case studyZalandoFashion retail · Germany · 2024 AI fashion assistant brought to all 25 markets 25 markets, local languages
The situation
Search and filters work when customers know what they want. Many purchases start instead with an occasion, a destination or a season. Zalando wanted to help customers in its European markets find their own style, in their own language.
What they deployed
The Zalando Assistant combines Zalando's own models with OpenAI's. Customers describe a need, such as what to wear to a November birthday in another city, and get personalised advice and suggestions. It weighs location, weather and occasion, alongside Trend Spotter.
The outcome
In October 2024 Zalando made the assistant available in beta to logged-in customers in all 25 of its markets, in local languages.

Personalisation built into the main shopping flow, in each market's language, can be rolled out across a whole European footprint, not just a pilot country.

Source: Zalando corporate newsroom, Zalando brings its AI-powered assistant to all markets and adds four new cities to its Trend Spotter, October 2024
Public case studyAllianzInsurance · Austria · 2024 Claims copilot that reads documents and images From motor to property claims
The situation
A motor claim means gathering claim and contract data, reading invoices, images and incident descriptions, checking them against policy terms and writing to experts. Claims staff wanted quick help with these steps, while coverage and payout decisions stayed with professionals.
What they deployed
Allianz Technology built Insurance Copilot, a generative AI tool for claims handlers. It summarises claim and contract data, analyses images and documents, checks claims against invoices and policy terms, flags discrepancies, suggests next steps and drafts emails. The handler decides.
The outcome
Launched in 2024 for motor claims in Austria, Insurance Copilot has since been expanded to property claims, and a cross-functional team is scaling it to other Allianz operating entities.

Employee apps that read documents and images can start in one country and one claim type, then scale once the human decision points are clearly defined.

Source: Allianz, AI at Allianz: Smarter claims management, smoother settlements, February 2025

These are public deployments by other organisations, summarised from the sources linked. They are not unus clients.

Built for European rules

Data stays where agreed

Models and data run in EU or Swiss cloud regions or on your infrastructure, and on the device where practical.

GDPR and Swiss nDSG

Impact assessments per journey, minimal data in prompts and logs, fixed retention for uploads, and records your data protection officer can review.

EU AI Act duties

AI use is disclosed (Article 50). Annex III makes credit scoring and life or health insurance pricing high-risk from 2 December 2027.

Oversight and security

Changes to money, contracts or personal data need the user's confirmation. Every action is logged, and agreed cases always reach your team.

Questions buyers ask

We already have an app. Do we need to start again?

Usually not. We start with the journeys where AI changes a business number and rebuild those flows inside your existing app, reusing your backend, design system and release process where they hold up. Some products need a new architecture underneath; we tell you during discovery, with the reasons and the cost, before anything is built.

Native or cross-platform?

It depends on the product. Cross-platform frameworks suit most service and commerce apps and keep web, iOS and Android in step. Native code makes sense where camera, audio or on-device processing must perform at its best. We decide per product, often mixing both, and the choice is documented in the architecture.

Which AI models do you use?

We choose per task: large language models for conversation and reasoning, specialised models for document extraction, image recognition and speech, and smaller models where cost or speed matters. Models are accessed through a layer we control, so they can be replaced as prices, quality or your hosting requirements change, without rebuilding the app.

How do you know the AI is good enough to release?

Before the first release we build test sets from real cases, including difficult and adversarial ones, and agree pass thresholds with you. Every release is evaluated automatically against them. In production we track answer quality, task completion, escalations and complaints per journey, and widen the user group only when the measured quality holds.

What happens when the AI is wrong or unsure?

The app shows the user what it understood before acting, asks for confirmation on anything that changes money, contracts or personal data, and hands cases below agreed confidence thresholds to a person with the full context. Errors found in production are added to the test sets so the same failure is caught before the next release.

Who owns the code and who runs the product?

You own the code, the prompts, the evaluation sets and the data. We build inside your repositories and cloud accounts, and run the product with your teams for as long as you want us to. Handover is planned from the start, so your people can take over build and operation in stages.

Bring one process. We'll bring the loop.