We already have RPA bots: what changes?

RPA moves data between screens and breaks when a screen changes. IPA adds document reading and an agent for the cases rules do not cover, and puts RPA only where it is really needed: where there is no API. The bots you have stay, and we measure them with the same cost per case.

What happens when the AI gets it wrong?

Every decision has a confidence threshold, and below it the case goes to a person with the reasoning already written. Decisions are traced with data, model version and rationale, so an error is found, explained and fixed upstream. In a bank that is the condition for getting through compliance.

Do we have to change the core banking or insurance systems?

No. We integrate via API where there is one, and via RPA where there is not. The process changes around the systems, not the systems.

Do customer documents leave the company?

No. Document reading and agents run in your infrastructure or in your European cloud tenant, and the data stays where it is. It is the requirement banks and insurers pick us for.

How much do we save, exactly?

The Diagnosis says, not us: two weeks, free, ending with the cost per case of your five best processes and the automatable share of each. Industry averages say 20-35% once live, but your number is yours to decide, looking at your own.

ENTERPRISE

Intelligent Process Automation area

Casework is still done by hand.

We make it run itself.

Every case passed from hand to hand costs someone's minutes and days of waiting. With Intelligent Process Automation (IPA, also called Cognitive Automation or iRPA) 50-70% of the steps run on their own, cycle time halves and people work only where judgement is needed.

Book the Diagnosis

The problem in numbers

The cost is in the handoffs, not the people

A banking case goes through several systems and several hands before it closes. Every handoff is waiting, retyping and a chance of error, and almost nobody books it as a process cost.

50-70%: Share of a process's tasks that can be automated
20-35%: Run-rate cost reduction once live
−50-60%: Straight-through processing time per case
>100%: First-year ROI in documented cases

Public sources: McKinsey, Intelligent process automation: the engine at the core of the next-generation operating model; banking and insurance industry benchmarks.

Diagnosis
2 weeks
Free
The process map
Inventory of repetitive processes with volumes, time per case and exception rate. The five best candidates with a €/year cost and how much of each can be automated.
Pilot Build
3 weeks
The first automated process
One end-to-end process in production on a narrow scope: document reading, rules, an agent for the ambiguous cases, people on the exceptions. Cost per case before and after.
Build
8-12 weeks
The process factory
Orchestration, connectors to the core systems, controls and audit trail. The mapped processes come in one at a time, each with its own measured cost per case.
Stewardship
Retainer
Automation stewardship
Volumes, exception rates and cost per case under watch; the bots are kept in step when systems change. Monthly report and quarterly review.

Where we apply it

Three banking, finance and insurance processes, worked end to end

The scope is that of a mid-sized bank or insurer. No client names: the cases are told by the shape of the process.

Specialty insurance

Tailor-made policies underwritten with AI

A specialty insurer built its own underwriting platform with agent coding: the customer describes the risk, the AI composes the tailored policy and prices it, and the underwriter steps in only on risks outside appetite.

Risk intake from questionnaire, documents and external sources
Clause composition and pricing by AI agents, within written appetite limits
Automatic issue; risks over threshold go to the underwriter with the reasoning already written

The underwriter moves from filling in policies to deciding on the risks that matter. The platform was written entirely with coding agents, measured like everything else.

Retail banking

Account opening and KYC without retyping

ID document, proof of address and questionnaire arrive through different channels and end up in three systems. The case assembles itself, and the operator only sees the ones with a doubt.

Reading of documents and selfie, with consistency checks across sources
AML and watch-list checks by rules, and an agent for the name-match cases
Opening in the core systems via API, or via RPA where there is no API

On clean cases opening goes from days to minutes; the operator works the exception queue, not the whole queue.

Business lending

Credit files ready before the committee

Financial statements, credit bureau data, company records and business plans read and reconciled on their own. The analyst receives the file already worked up, with the ratios computed and the points to probe highlighted.

Structured extraction of statements and records, reconciled across sources
Ratio computation and comparison against the credit policy
Pre-assessment written by the agent, with sources cited field by field

The analyst starts from judgement instead of data collection. Every number in the file has its source, and the committee sees it.

What you get

From “passed along” to “closes by itself”

Every item has a name and a measure: it goes in the contract and gets checked on delivery.

Process map

Which processes repeat, what each case costs and how much of it can be automated. We start where the ratio is best.

Cost per case

What every case costs before and after, in euros. The number that makes the saving verifiable rather than claimed.

Document reading

Structured extraction from PDFs, scans, e-mails and attachments, with a confidence level per field and human checks below threshold.

Agents for the ambiguous cases

Where rules are not enough, an AI agent decides within written limits, explains its choice and hands over when it is not sure.

Audit trail and controls

Every decision traced with its data, model version and reasoning: what compliance and internal audit ask for.

Exception workstation

A queue for the operator with the case already prepared: what is missing, why it landed there, what it takes to close it.

Inside your perimeter

The AI passes compliance, or it doesn't ship

Regulated companies are not afraid of AI: they are afraid of where the data ends up and of who answers when it gets it wrong. Three precise fears, each met with a written guarantee and a deliverable with a name.

Local models

Open-weight models in your infrastructure or in your European cloud tenant. No prompt, document or line of code leaves the perimeter for an outside API. The model is picked per task, not by habit, and where a local model is enough that is the one we use.

Deliverable: deployment architecture and register of the models in use, with where the data sits for each.

GDPR and privacy by design

Impact assessment before the first prompt, minimisation of the data that enters the model, defined logs and retention, roles and legal bases written together with your DPO. Privacy is in the design, not in an annex signed afterwards.

Deliverable: DPIA and updated record of processing, ready for the DPO.

AI Act

Risk classification of every system, transparency duties, human oversight where the regulation asks for it, an audit trail of every decision. Most obligations apply from 2026: no longer a plan, a requirement.

Deliverable: register of AI systems and a compliance file for each.

Questions we get

What people ask before signing

: RPA moves data between screens and breaks when a screen changes. IPA adds document reading and an agent for the cases rules do not cover, and puts RPA only where it is really needed: where there is no API. The bots you have stay, and we measure them with the same cost per case.

Contact us

A quote, a particular request, or simply a coffee to meet us? Write to us and we will reply to a "nano-second".

Address: Piazza Maestri del Lavoro 7
20063, Cernusco sul Naviglio (MI)
Italy
Address: Piazza dei Martiri 1
40121, Bologna (BO)
Italy
Telephone: +39 02 479 23365
Email: [email protected]