AI & Digital Solutions: the Infini team, now part of Forvis Mazars in Spain

Truly effective automation always begins with a rigorous diagnosis. Before developing any solution, we map out the existing process architecture: we identify dependencies between systems, detect redundancies, eliminate activities with no added value and determine where artificial intelligence will have the greatest impact. This preliminary redesign ensures that what is being automated is already an optimised process, not a digitised version of your existing problems.

Automating an inefficient process merely amplifies its problems. Our six-step methodology ensures that the process is optimised before it is automated.

Building on this foundation, we construct bespoke automation pipelines, combining the orchestration of language models, business logic and access to existing systems —ERPs, CRMs, databases, internal APIs. Going beyond traditional RPA, we integrate LLMs to handle unstructured cases, interpret documents and make contextual decisions that rule-based robots cannot resolve. We apply the same level of rigour to high-volume administrative processes as we do to highly technically critical workflows, such as validations in genetic analysis laboratories or real-time decision-making systems.

Automation methods

We always select the approach best suited to the technical context and the level of complexity of each process:

Robotic Process Automation (RPA)

Software robots that replicate human interaction with systems for repetitive, rule-based tasks. We work with Microsoft Power Automate, UiPath and Automation Anywhere, particularly effective for standardised administrative processes with a high volume of transactions.

Bespoke development with integrated AI

Pipelines in Python, JavaScript or C# that incorporate calls to language models, classifiers and information extractors. Suitable for processes with high variability, unstructured documentation or complex business rules requiring contextual reasoning.

Platform-based automation

Use of platforms such as Salesforce, SAP and Microsoft, alongside integration tools such as Make and Zapier, to create connected workflows between applications, suitable when rapid deployment and low technical maintenance are required.

We develop agents with persistent memory, access to external tools and multi-step reasoning capabilities using frameworks such as LangChain or LlamaIndex. We implement RAG (Retrieval-Augmented Generation) systems that connect models to corporate knowledge bases —technical documentation, manuals, case histories— enabling accurate and traceable responses without hallucinations or data leaks. Agents can be deployed in both public customer service environments and fully private infrastructures for internal use.

We design agents tailored to your processes that understand context, interact naturally, access corporate databases and execute actions with precision. Conversational solutions can reduce enquiry resolution times by up to 70%, whilst keeping information fully secure within the corporate environment.

Our approach begins with an analysis of business-critical conversation flows and continues with the design of intents, entities, memory and dialogues that ensure consistent, high-value experiences. Integration with external APIs, vector databases and internal systems enables the agents to operate as true digital collaborators.

Benefits of AI Agents

• Customised and contextually accurate interactions with users and customers.
• Continuous 24/7 availability without any reduction in the quality of the response.
• Significant reduction in the operational workload on support and IT teams.
• Highest protection of confidential information through deployment in a private environment.
• Full traceability of sources consulted thanks to the RAG architecture.

We deploy and adjust open-source models —Mistral, LLaMA, Qwen— on on-premises NVIDIA infrastructure, applying supervised fine-tuning and RLHF techniques to specialise the models for each organisation’s domain and tone. Multi-model orchestration enables each task to be routed to the most efficient model according to its nature: extraction, synthesis, classification or generation, optimising computational cost without sacrificing response quality. We design architectures that are completely isolated from the public cloud, maintaining total control over business data.

AI deployments in private environments can reduce long-term operational costs by up to 60% compared to models based exclusively on cloud services.

The project is developed following a comprehensive analysis of requirements and data, followed by the design of a secure and fully isolated architecture, the selection and fine-tuning of the most suitable base model, and integration with existing information systems via controlled internal APIs.

Advantages of Private AI

• Complete privacy of corporate and scientific data.
• Enhanced security by avoiding any exposure to external networks.
• Absolute control over model versions, updates and deployments.
• Fully specialised models through fine-tuning using proprietary data.
• Significant long-term savings on licences and cloud usage.
• Guaranteed compliance with data protection regulations (GDPR, sector-specific).

We offer our own methodology and technology platform to harmonise and standardise the development of software applications within a single controlled environment, covering all phases of the Software Development Life Cycle (SDLC). From the drafting of functional and technical specifications to automated deployment into production via CI/CD pipelines, every stage of the process is aligned, documented and governed according to the same quality and security standards.

The AI Software Factory not only accelerates development: it ensures that every application produced is consistent, maintainable and auditable, regardless of the team or project responsible for it.

The platform integrates AI support into every phase of the cycle: specification generation and validation, architectural design, code review, automated test generation and coverage analysis. This enables development teams to work faster without compromising quality and allows technical leads to maintain visibility and control over the status of each project in real time.

SDLC phases covered

We carry out an independent and structured assessment of the actual state of artificial intelligence within the organisation: which models and tools are in use, how they are being used, what technical, ethical and regulatory risks they present, and what gap exists between the available potential and the value being realised. The audit covers both in-house developed solutions and implemented third-party tools —co-pilots, assistants, automations, model APIs— and produces an executive report with findings, prioritised risks and actionable recommendations.

Many organisations have more AI than they realise, but less governance than they need. Audit transforms that opacity into visibility and control.

Areas of assessment

We design and implement a comprehensive AI strategy tailored to each organisation’s actual maturity level. Many companies begin their journey with productivity tools, such as Microsoft Copilot or ChatGPT, without an overarching vision to guide adoption, avoid duplication and ensure that every investment contributes to a common goal. The Excellence Plan builds on this maturity assessment to create a coherent roadmap: from the adoption of initial tools through to the implementation of advanced capabilities for model development, orchestration and governance.

Adopting Copilot is a valuable first step. Building an AI strategy is what turns that step into a sustainable competitive advantage.

The plan addresses three dimensions in an integrated manner: technology strategy —which capabilities to develop and in what order—, governance —how to manage risks, data and regulatory compliance— and people —how to build the internal capabilities needed so that the organisation does not remain indefinitely dependent on third parties. The establishment of an AI Centre of Excellence (CoE) is the organisational mechanism that institutionalises these three dimensions, creating a hub of knowledge, standards and decision-making that accelerates adoption and ensures long-term consistency.

Plan Components

• Maturity assessment: assessment of the current level of AI adoption, capabilities and culture.
• Strategic roadmap: prioritisation of initiatives by impact, feasibility and alignment with the business.
• Governance framework: usage policies, risk management, model quality control and regulatory compliance.
• Reference architecture: technology standards, integration patterns and tool selection criteria.
• Establishment of the AI Centre of Excellence (CoE): structure, roles, processes and operational KPIs.
• Training programme: tailored training by profile —managerial, technical and end-user— to build internal autonomy.

AI Centre of Excellence

The CoE acts as the internal reference body for all matters relating to artificial intelligence: it centralises knowledge, sets standards for development and use, evaluates new tools and models, and supports business teams in identifying and implementing initiatives. Its establishment transforms AI from a collection of disparate projects into a structured, scalable and governed organisational capability, capable of generating value continuously and sustainably.