How tax advisors use AI in nuwacom

Tax firms work across DATEV, ELSTER, commercial registers, client documents, spreadsheets and email every day. A large share of the time goes not into the professional judgement itself, but into the steps before it: collecting documents, retrieving data, comparing figures, looking for deviations and preparing the result so that a qualified professional can review it.

With nuwacom, firm knowledge, skills, agents, workflows and connected systems come together in one centrally managed AI environment. The result is repeatable review processes with clear rules and human approvals, instead of individual prompts without shared context.

This article presents a number of setups for tax firms. All of them prepare and pre-check, they do not decide. The professional assessment and the final responsibility always remain with the qualified professional.

Which AI agents and workflows can tax firms build with nuwacom?

With nuwacom, firms can model recurring retrieval, reconciliation and review processes as skills, agents and workflows. This works particularly well for tasks that follow fixed rules, recur regularly and draw on firm knowledge or connected data sources.

Agent templates are already available in nuwacom for a number of these tasks, for example the Document Intelligence Agent, the Contract Review Agent and the Knowledge Base Assistant.

How can firms connect nuwacom directly to DATEV?

nuwacom can integrate MCP servers as connectors, alongside the prebuilt app connectors it supports standard MCP, custom MCP and your own APIs. For DATEV this opens two established routes without the firm having to run its own integration project.

MCP for DATEV by riecken.io connects the firm's DATEV environment, including locally operated DATEV installations. Available data includes master data, accounting with BWA, open items, trial balance and account sheets, as well as document search in the DMS including text recognition on scans. Access is separated into read and write, visibility can be restricted per user and per client, every request is logged, and write operations run through a change plan with preview and explicit confirmation. The provider states compliance with § 62a StBerG and § 203 StGB.

The DATEVconnect Gateway with MCP by klardaten.com provides the same access through gateway infrastructure hosted in Germany, and additionally offers a REST interface for your own applications and automations, including search across DATEV DMS and document storage.

For the setups on this page this means agents and workflows operate on the actual client records rather than on exported snapshots, while permissions stay where they belong. Which data is available depends on the DATEV modules in use and the permissions granted. Before productive use, data processing agreements, confidentiality undertakings and the professional law assessment need to be clarified.

How MCP works technically is described in the article on MCP in practice. The full range of connectable systems is listed under nuwacom connectors.

How does client acceptance work with nuwacom?

Client acceptance is the process that shows the principle most clearly: many sources, fixed criteria, a result that has to be documented, and a decision that must stay with the qualified professional.

A possible sequence:

1. Record the firm's acceptance criteria, meaning legal forms, sectors, risk classes, exclusion grounds and required documents.

2. Define the sources: submitted documents, register extracts, publicly available information and the existing client base.

3. A skill defines which information must be complete and what counts as an open point.

4. An agent brings the documents together, maps them to the criteria and flags gaps and contradictions.

5. A workflow obtains approval from the responsible professional before the engagement is set up.

6. The result goes into the file as a dossier, with source and retrieval timestamp for every data point.

nuwacom workflows carry the process logic, AI skills hold the review rules, and agents do the consolidation.

What could a client acceptance skill look like in nuwacom?

A nuwacom skill records the rules the agent works by. The team does not have to rewrite the working method into a prompt on every run, and two employees receive the same review framework for the same task.

A possible skill contains:

• the firm's acceptance criteria and exclusion grounds

• required documents per legal form and engagement type

• rules for mapping shareholdings and representation arrangements

• how to handle missing, outdated or contradictory information

• mandatory fields for source and retrieval timestamp

• the required output format

• clear limits on which decisions must not be automated

The output per engagement can compactly contain:

• client, legal form and registered office

• mapped shareholdings and representation arrangements

• documents present and missing

• matches against the exclusion criteria

• open points for professional review

• source and retrieval timestamp per data point

The same pattern carries the remaining setups: the skill holds the review framework, the agent applies it, the workflow brings it to the right person at the right time.

How do context, skill, agent and workflow work together in nuwacom?

nuwacom separates the layers of an AI process so firms do not have to build an isolated system for every use case.

• Context engine: makes approved firm knowledge available to chat, apps, agents and workflows.

• Skill: defines rules, review steps, mandatory fields and output formats.

• Agent: handles a clearly defined task based on the approved context.

• Workflow: connects triggers, conditions, agents, integrations and human approvals into a repeatable process.

Through the context engine and connectors, existing firm knowledge and connected systems can be brought into these processes.

"Ask the AI" becomes a controlled workflow that applies the same standards on every run.

How can assessments and tax accounts be reviewed with nuwacom?

Three processes follow the same structure: the data is already there or is retrieved through the existing channel, and the actual work lies in a structured comparison.

Income tax assessment review. An agent compares the figures in the assessment against the submitted return data, sets out deviations line by line and maps them to the relevant code. Deviations that cannot be explained are flagged rather than interpreted. A workflow can carry the appeal deadline as a date so the finding does not sit unattended.

Account reconciliation review. Balances, open items and payment positions are compared under fixed rules. The agent reports differences with document references and separates explainable items such as shifted due dates from the cases that need clarification. Through the DATEV MCP, the review runs on the current position rather than on an export.

Tax account and VaSt retrieval. The retrieval itself continues through the designated channel and with the existing authorisations. nuwacom starts at the point where the data is available: it is prepared consistently, compared against the firm's own records, and deviations are presented in a list a person can work through in a few minutes.

In all three cases nuwacom is the AI and orchestration layer. The data comes from defined sources, and which of it can be processed depends on the access routes available and permissible to the firm.

How can commercial register extracts and shareholding structures be analysed?

Register extracts are the case where the effort grows with the number of client numbers while the review logic stays the same.

An agent can, for example:

• read extracts for one or several client numbers

• extract shareholders, management and representation arrangements

• map shareholdings across multiple extracts

• build a shareholding overview or org chart from them

• flag differences against the stored master data

• report changes against an earlier retrieval

This becomes particularly valuable for affiliated companies, where the structure is spread across several extracts and nobody has had it in one place before. The Document Intelligence Agent handles the document analysis, while the mapping follows the rules in the skill.

How can payroll data be checked for plausibility?

In payroll, errors rarely occur where someone was not paying attention. They occur where an entry looks plausible and is not.

Company car benefits in kind. The vehicle description from the payroll run is compared against a vehicle database. Not every vehicle with "hybrid" in its name is a tax-privileged hybrid, and those are exactly the cases a visual check misses. The agent does not deliver an assessment but a list of vehicles where the description and the database entry do not match unambiguously.

Deviations over time. Payroll runs from several periods are compared, and the agent flags notable changes: new or discontinued wage types, sudden jumps, amounts outside the range usual for that client. The firm sets the thresholds in the skill. Only cases above those thresholds are passed on, so the list stays short enough to actually be reviewed.

How do firms make internal knowledge available to agents?

Through the nuwacom context engine, approved documents and connected systems can serve as a shared knowledge base for agents and workflows.

This is relevant for, among others:

• the firm's review instructions and working procedures

• checklists per engagement type and legal form

• client history and documented decisions

• templates for correspondence and client communication

• internal technical notes and interpretation decisions

• onboarding material and process descriptions

An agent then works not only with general model knowledge but with the way this particular firm works. That is also the lever against knowledge loss during engagement handovers.

How does the use of AI agents in a firm stay controllable?

nuwacom combines agents and workflows with central governance. Roles, access and model approvals can be managed centrally, while teams use their own AI applications within their permissions.

For use inside a tax firm, the following are particularly relevant:

• confidentiality under § 203 StGB and the requirements of § 62a StBerG

• approved knowledge sources and per-user client visibility

• roles, permissions and controlled model approvals

• traceable workflows and logged access

• human in the loop before every step with external effect

• separation of read and write access to DATEV

• data processing agreements and hosting options

More on this is covered by nuwacom AI governance and the article on § 203 StGB and DORA compliance in regulated industries.

What experience exists from comparable processes?

One documented example comes from an adjacent field. BHB & Partners is an independent financial advisory firm in Luxembourg, not a tax firm, but the process patterns are the same: document-heavy reviews, fixed criteria, results that have to be documented.

In its AML and KYC process the firm reduced research and documentation effort with nuwacom by 65 percent. In M&A data room management the reduction is 40 to 50 percent, with full consistency checking across all documents. In both cases the professional assessment stayed with the partners. What changed was the preparatory work.

The BHB & Partners case study describes both processes in detail. Why the pressure in professional firms comes from the same direction is set out in the article on how lawyers, auditors and tax advisors can meet the talent shortage with AI.

When is nuwacom the right AI platform for a tax firm?

nuwacom is particularly relevant when a firm wants to do more than provide individual AI chats, and instead run firm knowledge, reusable review standards, agents, workflows and the DATEV connection on one shared platform.

Before making a choice, firms should check, among other things:

• Can the AI access approved firm knowledge?

• Can the firm's own review standards be stored as skills?

• Can agents handle multi-step reviews?

• Can DATEV be connected via MCP, with separated read and write rights?

• Are human-in-the-loop approvals possible?

• Can different models be used for different tasks?

• Can roles and model approvals be managed centrally?

• Are confidentiality, data processing and logging covered?

• Is there a traceable path from pilot to regular operation?

This checks not only whether a model writes good answers, but whether the platform fits the way the firm operates. A full overview is available under the AI operating system for tax advisors and auditors.

How does a firm get started with nuwacom AI agents?

With nuwacom a firm can start by modelling one clearly defined process and then extend the same approach to further teams and client groups.

A sensible entry point:

• pick a frequent, clearly defined process, for example the assessment review

• define the required knowledge and data sources, including DATEV access

• record the existing review standard as a skill

• test agent and workflow on real cases with human control

• measure time saved, quality of findings and adoption

• only then connect further processes

Anyone who wants to walk through a specific firm use case can map the process together with the team through the nuwacom demo.

What does nuwacom cost for tax firms?

nuwacom offers Business and Enterprise plans. For larger firms and firm networks the Enterprise context is often the relevant one, where individual integrations, firm-wide governance, hosting options, SLA or dedicated onboarding are required. Costs for DATEV MCP access are charged separately by the respective provider.

Current terms are available on the nuwacom pricing page.

In short

With nuwacom, tax firms can bring firm knowledge, review standards, AI agents, workflows, the DATEV connection and governance together on one platform.

The most immediate areas of use are:

• client acceptance and acceptance documentation

• account reconciliation and open item review

• income tax assessment review

• preparation of tax account and VaSt data

• commercial register analysis and shareholding structures

• company car benefit in kind checks

• plausibility checks on payroll deviations over time

The decisive step is not to build as many agents as possible. What works is one clear, recurring review process with measurable benefit, which the firm first puts into controlled production and then extends. In every setup, the professional decision stays with the qualified professional.