In most law firms, tax advisory practices and audit firms, a quietly remarkable piece of work is performed each day that never appears on a fee note. It is the repeated answering of questions that have already been answered, often more than once, often by the very same firm. The cause is rarely that the underlying matter is intellectually difficult, although it sometimes is. The cause is that the earlier answer, which sits somewhere in the house, cannot be retrieved in time to be useful.
Anyone who has spent a few seasons inside a knowledge-intensive professional services firm will recognise the choreography. A client rings with a question that feels strangely familiar to the partner who picks up the phone. An associate half-remembers a similar matter from the previous spring. A note about it almost certainly sits in the document management system and probably also in the inbox of the colleague who handled the original file. The professional who locates the note gains half an hour. The one who does not begins again from a blank screen. In the average firm, somebody begins again several times a week, with admirable patience and at a perfectly respectable hourly rate.
The economic size of an invisible activity
Before examining the individual professions, the macro picture deserves a glance. The widely cited McKinsey study The Social Economy already estimated in 2012 that knowledge workers spend roughly 19 per cent of their time searching for information. Later research, including Microsoft's annual Work Trend Index, studies by IDC, Atlassian and Coveo, places the figure somewhere between eight and eleven hours per knowledge worker per week, depending on methodology. In the advisory professions, where research, drafting and correspondence are the actual product, the share sits at the upper end of that band rather than the lower.
The economic gravity of this comes from the billing structure. Hourly rates in commercial law firms across the DACH region, by the figures reported by the German Federal Bar (Bundesrechtsanwaltskammer) and the customary market surveys, sit somewhere between roughly two hundred and fifty and six hundred euros, depending on seniority. In international full-service firms, partner rates can comfortably exceed that. In tax advisory and audit the picture is more heterogeneous, although here too the average billed rate sits firmly in the three-figure euro range. Every hour spent looking for knowledge that already exists in the house is an hour that is either silently absorbed into the margin or, more often, that the client never sees.
This sum is rarely calculated honestly. If it were, it would comfortably run into six figures a year in any mid-sized practice and into seven figures in the larger houses. That is the unsentimental reason why the question of what a firm does with its own matter knowledge is quietly turning into a question of commercial leadership rather than back-office hygiene.
Why these three professions in particular
Law firms, tax advisors and audit practices share a structural feature that makes them unusually exposed to questions of knowledge infrastructure. Their core service consists in applying normative texts and earlier interpretations to a concrete client situation. That application is almost never genuinely novel. It draws on the accumulated understanding of prior matters, on commentary literature, on administrative guidance, on professional pronouncements and on the silent reservoir of judgement that resides in individual practitioners.
A second feature follows from professional conduct. Client confidentiality and professional secrecy make it impossible to outsource the work to publicly available knowledge once the question becomes concrete. A practitioner who wishes to advise a client cannot rely on what a public language model has overheard about comparable matters. He or she is dependent on the prior work that exists inside the house. In most houses, that prior work is fragmented.
A third feature deserves to be acknowledged with the slight discomfort it warrants. The fee structure rewards hours rather than reuse. A junior who works through the same question for the fourth time in a similar configuration produces a commercially questionable output and a perfectly billable timesheet. The professions therefore contain a quiet incentive to redo work, dressed in the respectable language of due care. The professional who spent five hours researching looks diligent. The professional who found the firm's earlier answer in fifteen minutes and applied it to the new fact pattern looks, faintly, as if he might have cut a corner.
Law firms: the matter memory
In commercially oriented law firms the pattern is visible every day. A mid-sized firm with three offices and forty fee earners will have produced several thousand matter files over fifteen years. In each of those matters, pleadings have been drafted, contractual clauses negotiated, review schemes applied, litigation strategies articulated. The material exists. It is filed, often even ordered. Yet once the responsible professional has left the firm or is simply busy with another matter, the material is, in any practical sense, inaccessible.
A recurring example is employment law. The question of what constitutes valid termination for cause in a suspected breach of duty appears in any commercially oriented firm several times each year. In each case a comparable matter exists somewhere in the house, including a serviceable client letter and a tactical recommendation already thought through. In each case the new matter begins from a blank page, because while the document management system searches successfully, it ranks results by file number or date rather than by relevance. The professional who cannot tell from the hit list whether a document is a substantive memo or a passing email opens twenty files and then concludes that re-reading the underlying authorities himself will be faster.
The same applies to transactional work. Negotiating a Limitation of Liability clause in an international supply agreement is not a rare event for a mid-sized firm. The formulations developed in prior matters, including the variants the firm has agreed and the variants it has consistently refused, constitute concentrated negotiating intelligence. They sit across ten or twenty separate matter files, each with the context that makes their portability comprehensible. A firm that systematically opens this reservoir not only negotiates faster, it also negotiates more consistently, which is the rarer and more valuable property of the two.
The professional conduct dimension is immediate. Client secrecy and the duty of care prohibit dropping draft agreements into publicly available AI tools, even with names redacted. In practice that step nevertheless occurs, often quietly, often in the best of intentions, often to move a matter along on a tight evening. A firm that does not absorb this behaviour into its own controlled system effectively delegates a slice of its matter work to an infrastructure whose data flows it does not oversee, which is an unusual position for a profession built on the careful stewardship of other people's secrets.
The most structurally telling moment comes at partner transitions. A partner retires, leaves or scales back. With her departs, in many cases, a large fraction of what made her stewardship of those matters possible. It is not the file that is lost. It is the knowledge of which arguments the opposing side typically advances, which judge in which chamber tends to favour which reading, which client is sensitive to which register of correspondence. None of this is filed anywhere. It exists only to the extent that the firm has built an environment in which it can be captured at all.
Tax advisory: the recurring interpretive question
In tax advisory the phenomenon may be even more pronounced, because the frequency of recurring client queries is higher. A practice with twenty professionals and three thousand clients will field several hundred telephone and written enquiries on standard themes in any given quarter. Travel expenses, business entertainment, gifts to business contacts, VAT on cross-border services, marginal employment, company car treatment, home office. Each question has been answered before, multiple times, drawing on the same circulars, the same court decisions, the same administrative guidance, often on the same internal template letters.
A concrete configuration. A client runs an online business shipping to consumers across several EU member states and is asking, for the first time, whether the One Stop Shop scheme can be applied to a particular constellation or whether registration in the destination states is required. The practice has handled this question in the past three years more than once. Memos exist, advisory notes exist, a usable client letter exists, an internal decision tree exists that distinguishes the typical fact patterns. The substantive answer is in the house. The organisational question is how the responsible advisor finds it in the fifteen minutes available between two appointments.
The same logic applies to payroll. Questions on benefits in kind, on flat-rate taxation under §37b EStG, on the taxable benefit of an electric company car with a home charging station, on the payroll-tax treatment of benefits delivered through third-party benefits platforms or on the social-security classification of hybrid employment models recur across the client base with a regularity that would be comical if it were not also expensive. Each query is professionally clear in isolation. Collectively, they are answered afresh hundreds of times each year, because the firm's own earlier work cannot be brought to bear on the next instance.
The Federal Chamber of Tax Advisors (Bundessteuerberaterkammer) has for several years documented a widening shortage of professional successors. Enrolment in the relevant degree programmes is flat, the pass rates of the tax advisor examination remain demanding and the age structure of the profession is visibly shifting. In this constellation the question of how a firm makes the knowledge held in the heads of its experienced professionals available to the incoming generation moves from comfort to viability. Whoever is to run the practice in ten years cannot run it if the relevant knowledge continues to reside exclusively in the heads of those who will by then have retired.
A frequently underestimated dimension is client communication. A tax practice that answers its most common client questions from a coherent base of its own earlier answers communicates not only more efficiently but also more consistently. Clients who receive two subtly different answers to the same question over the course of two years sometimes draw the unflattering conclusion that the firm is not taking their matter seriously. A consistent house voice, one that does not depend on the mood and the personal notes of the individual answering on a given afternoon, is a quiet quality safeguard that translates into client retention.
Audit practices: the audit memory
In audit, the phenomenon carries its own slightly self-conscious name. It is usually called lessons learned and refers to the attempt to preserve from a completed audit those insights that might matter for next year's audit of the same client or for comparable engagements. In practice the process is more ritualised than lived. The lessons are entered into a template, the template is filed, and the next audit begins with a phone call to the colleague who ran the previous year.
A concrete configuration. A mid-sized industrial group has been audited by the same firm for eight years. In each of those years, interpretive questions on the capitalisation of development costs under IAS 38 have been discussed, in each year the threshold for capitalisation has been reassessed on the facts, in each year similar questions on the recoverability of the capitalised amounts have arisen. The answers from these eight years exist in the working papers and final reports of each annual audit. Collectively, they exist nowhere in consolidated form. The new engagement partner who takes the file this year therefore begins with several days of trawling through prior-year papers, performing exactly the consolidation that a living audit memory would deliver in an hour.
The standards themselves, the International Standards on Auditing together with the supplementary pronouncements of the Institute of Public Auditors in Germany (IDW) and equivalent bodies in Austria and Switzerland, have grown considerably over the decades. Their application to concrete audit situations is an exercise that requires experience, because the standards deliberately leave room for professional judgement, the exercise of which is the actual product. That judgement lives in the heads of experienced partners and senior managers, and is rarely filed in a form that surfaces its value for future audits.
A further dimension arises from the regulatory framework. Since the EU audit reform and its national transpositions, the requirements on audit quality, documentation and internal inspection have risen considerably. Inspection reports by the Auditor Oversight Body (APAS) in Germany, RAB in Switzerland and equivalent supervisory authorities elsewhere consistently identify consistency of execution across offices and reporting periods as a recurring theme. A knowledge infrastructure that surfaces previously exercised judgement is therefore not primarily an efficiency measure. It is a contribution to audit quality.
Demography matters here as well. The auditing profession is ageing, the qualifying examination remains demanding, and the inflow of new entrants has not, on the reporting of the Chamber of Public Auditors (Wirtschaftsprüferkammer) and the customary industry surveys, kept pace with the retirement wave the coming decade will bring. In that setting, the question of how a firm preserves the experiential knowledge of its departing partners is no longer a matter of nicety. It is a matter of operability.
What runs through all three
For all the differences between legal, tax and audit work, four shared patterns are worth naming, because they explain why these professions are unusually exposed to the question of knowledge infrastructure.
First, reuse of knowledge is the rule rather than the exception. A client question that arises in a firm for the first time is rare. Far more often the same question appears for the third, fifth or twentieth time. The efficiency of handling depends directly on whether the firm knows how to draw on its own reservoir.
Second, the language in which the knowledge sits is technically demanding. It follows the conventions of its profession rather than general conventions, with its specific terms, cross-references, abbreviations and rhetorical patterns. Generic search tools that do not understand this language deliver hits whose relevance is only knowable by reading them. The reading consumes precisely the time saving that the search was supposed to deliver.
Third, the legal and professional sensitivity of the material is high. Client confidentiality, professional secrecy, data protection and, in audit, independence requirements impose constraints on handling that are foreign to most other industries. A knowledge infrastructure that intends to survive in the advisory professions must be designed under those constraints from the very beginning rather than retrofitted to them later.
Fourth, the dependence on individuals is unusually tight. A matter is not only what stands in the file. It is also what the responsible professional carries in his or her head. This layer travels with the person. When the person leaves, so does the layer. A knowledge infrastructure that does not take this layer seriously describes only half of the matter memory.
What the unofficial infrastructure already shows
In most firms, the staff have already answered the underlying question, although nobody asked them to. The answer is not labelled internal knowledge management. It travels under the rather more honest heading of publicly available AI tools. Walk the corridors of any decent-sized practice in the early evening and the same scene plays out. Junior professionals use freely available language models to read into unfamiliar matters, to sketch out argument structures, to assemble first drafts of pleadings and to polish client correspondence. They do so in a grey zone that professional conduct has not expressly forbidden and has very carefully not endorsed either.
From the perspective of the individual, the behaviour is perfectly rational. The tools deliver a noticeable acceleration of work and a far gentler entry into complex matter. From the perspective of the firm, the behaviour is uncomfortable, because it sits outside professional oversight and because it does not draw on the knowledge the firm itself has produced. From the perspective of the client, it is awkward, because information that strictly belongs to the matter is being processed in systems whose data flows the firm does not control. From the perspective of the supervisor, it is unresolved, because the question of where responsibility lies when an answer has been produced under the influence of an external model has not yet been mapped in any settled way.
The strategic conclusion is not that these tools should be pushed back. That is not realistic in any house with a recognisable workforce. The conclusion is that professional reality is asking for an infrastructure that acknowledges the evident usefulness of these tools and offers it under the conditions the profession requires. That is the point at which the conversation about Private AI Knowledge Systems begins.
What a Private AI Knowledge System delivers, concretely
Read in this light, a Private AI Knowledge System is not yet another tool added to a tool landscape that is already crowded. It is the organisationally responsible form of what the workforce is already doing. It takes the function that public language models perform for the individual professional and relocates it into an environment that belongs to the firm, whose data flows the firm understands, and into which the firm's own matter knowledge enters as a source.
For a law firm this means that pleadings, internal memos, matter notes, prior advice, draft contracts in their various negotiation states and the relevant commentary literature become searchable in a controlled environment that ranks not by file number but by substantive proximity. The junior associate confronted for the first time with a Limitation of Liability clause does not see a blank document. He sees the formulations the firm has worked through in recent years, the negotiation history that surrounds them and the colleagues responsible for the relevant matters. The advice is not delegated. It is far better prepared.
For a tax practice this means that the firm's existing memos, client letters, internal decision trees and the relevant circulars become accessible in connection with the concrete configurations from the client base. An enquiry on the One Stop Shop scheme no longer begins with re-reading §18j UStG. It begins with an overview of the cases in which the firm has already examined a comparable configuration, together with the conclusions and the reasoning. The hours that are released can be spent in two directions. They can go into the more demanding matters that genuinely require fresh work. They can go into the development of a considered client communication that binds the firm over the longer term.
For an audit practice this means that working papers, final reports, internal interpretive memos, lessons learned and the prevailing professional pronouncements become accessible in a form that actually reconstitutes the audit memory. A new engagement leader taking over an existing client does not begin with several days of file mining. She begins with a thematically ordered overview of the interpretive questions handled in prior audits, together with the professional judgements that were taken. The audit is not abbreviated. In the parts that require experience, it is conducted with the experience of the practice rather than only the experience of the individual currently in charge.
In all three settings, the professional's own responsibility remains entirely intact. An AI-supported answer is preparation, not substitution. It makes the judgement faster, it does not pronounce it. Precisely this separation is what the professional framework demands and what a controlled environment realises more cleanly than ad hoc use of public tools without professional embedding.
What a Private AI Knowledge System is not
The boundaries are worth drawing, because the term wears out quickly in the current market rhetoric. A Private AI Knowledge System is first of all not the use of public AI tools accompanied by a polite reminder about discretion. Prompt behaviour is not infrastructure. Whatever is entered into a public service leaves the house, however carefully it is phrased.
It is, secondly, not the improvised wiring of a language model to a stack of PDF files in an architecture that demos beautifully and fails the firm's role and confidentiality model on the first production day. A serious architecture separates roles, respects matter assignments, records access and makes the provenance of every answer traceable.
It is, thirdly, not a generic copilot that lifts the firm into the same standard environment as every other organisation. The value of a knowledge system in the advisory professions arises from its connection with the firm's own material. A tool that does not allow that connection accelerates generic tasks without addressing the matter memory.
It is, fourthly, not a project to hand to the IT department. The question of which knowledge should be available, in which form, to which role, is a professional, strategic and partnership question. It calls for a decision by the firm's leadership, not for a procurement process.
What the market data suggest
Robust surveys on the use of language models in the daily work of the advisory professions across the DACH region are still young, although the picture they sketch is reasonably consistent. A series of professional-body surveys, including the reporting of the German Federal Bar (Bundesrechtsanwaltskammer), publications of the German Bar Association (Deutscher Anwaltverein), the annual industry studies by DATEV in the tax advisory segment and the relevant work of the Institute of Public Auditors in Germany (IDW) and the Swiss audit chamber (EXPERTsuisse), point in the same direction. Individual adoption of generative AI is widespread. Institutional embedding in a firm-owned, governed environment remains the exception.
An international observation reinforces the point. The annual Wolters Kluwer Future Ready Lawyer Survey, the reports of Thomson Reuters on generative AI in the legal profession and the studies by the major international consultancies on generative AI in professional services suggest that the firms that invested early in their own controlled knowledge infrastructure perceive themselves as more productive, and are perceived by their clients as more consistent. The time between a client enquiry and the first substantive answer shortens, and the quality of the answer becomes less dependent on whether the most experienced person happens to be available that afternoon.
None of this constitutes proof. It is a pointer, towards a direction that the DACH region is only beginning to engage with seriously. The firms that take that direction early and carefully will, in five years' time, occupy a position that slower-moving competitors will then need to reconstruct at considerable cost.
The question that remains
In most firms the conversation about AI is currently being conducted as a technological question. Which tools are deployed, which vendors are credible, which models perform what. These questions are not wrong, they are simply secondary. The first question is what a law firm, a tax practice or an audit firm actually wishes to do with the knowledge it has accumulated over the years. Whether it intends to keep entrusting that knowledge to the personal memory of individual professionals, whether it wishes to bring it into a form that belongs to the partnership as a whole, or whether, by quiet default, it would prefer to hand it over to external tools that lighten the daily load and give very little back in exchange.
The question is professional, it is strategic and, as the figures suggest, it is also commercial. Answering it does not require a decision about a particular vendor. It requires a settled view on what the firm's own matter memory is worth to it, and to whom that memory should belong in five, ten and twenty years' time.
In the coming instalments of this series the same thread will be developed for further segments. For consultancies and agencies, for engineering practices and technical service providers, for software companies, for industrial groups distributed across many sites, and for the overarching questions of governance and technical architecture. The advisory professions stand at the start of the sequence because they exhibit the pattern in unusually pure form. They live by knowledge, they lose a portion of it each day, and they now have, for the first time, a means of resolving both at once.
