Every Domain Has Its Substrate: A Walkthrough Across Six Professions

This article works through the substrate of six professions. If you work in one of them, you should recognise your section and be able to start drafting your own.

The lists are not complete. They are a starting point. The six sections are independent, so go straight to yours.

What a Substrate Is

A substrate is a firm's captured expertise. It has three layers and three wrappers.

Three substrate layers — quantitative, subjective, procedural — surrounded by three wrappers: question archetypes, constraints and safety, and the grounding constraint

The three layers are what the firm already knows.

Quantitative. The metrics, frameworks and calculations the field uses: materiality in accounting, bracing tables in construction, the PMC in cycling coaching. This is textbook material, plus the firm's own calibrations of it.
Subjective. What the metrics miss. Pattern recognition built up over years, such as knowing something is wrong before the figures show it. It is rarely written down, but it affects decisions.
Procedural. How an expert responds: the structure of a good answer, the order of the steps, the decisions nobody has written down.

Three wrappers are defined from the layers.

Question archetypes. The kinds of question that come up in the work, grouped by how they are answered.
Constraints and safety. What the expert refuses to do, what is escalated, and what is out of scope. These matter as much as what the AI can do.
Grounding constraint. What every output is tied to, so that it is correct for this firm and this client, not just plausible.

Every answer the AI gives is tied back to both the layers and the wrappers.

Six professions follow.

Accounting (Tax / Audit / Advisory)

Quantitative. The relevant tax statute (Income Tax Act 2007 in NZ; IRC in the US; your jurisdiction's equivalent). Accounting standards (NZ IFRS / IFRS / GAAP). Materiality thresholds as the firm calibrates them, not the textbook ones. The audit risk model (inherent × control × detection). Depreciation and amortisation conventions. Transfer-pricing rules where relevant. The firm's chart-of-accounts standards. Engagement profitability ratios: realisation, utilisation.

Subjective. "Something is wrong with this balance sheet." The patterns a senior partner recognises and a junior takes years to learn: related-party transactions with an unusual structure, revenue recognition that happens to track management bonus thresholds, working-capital movements that suggest liquidity stress before the cash flow does. Client temperament: who can be told a hard finding directly, and who needs to be led to it gradually.

Procedural. How a partner triages a tax-planning question. The order of operations in a year-end engagement. When to escalate to peer review. How to word a contentious finding in a management letter so the client accepts it and the relationship continues. Recognising an aggressive structure the firm has seen before.

Question archetypes. Tax planning and structuring. Year-end close support. Audit-finding interpretation. Compliance deadline tracking. Client onboarding and engagement scoping. Industry-specific recurring patterns (farming, tech, property). Off-topic: anything that amounts to legal or investment advice.

Constraints and safety. No individualised tax advice without partner sign-off. No legal opinions. No statements on contentious estate planning. AML/CFT flags follow the firm's protocol, not the AI's reasoning. An AI summary does not replace professional scepticism in audit.

Grounding constraint. Every recommendation is traceable to the firm's prior engagements with this client (or comparable ones), the relevant statute, and the partner's documented position. Not generic tax advice from the web.

Wealth Management

Quantitative. Portfolio metrics: Sharpe, Sortino, max drawdown, tracking error, beta. Asset-allocation frameworks: strategic-tactical splits, glide paths, risk parity. Tax-aware structures: KiwiSaver / PIE / PIR (NZ); 401(k) / IRA / Roth (US); ISA / SIPP (UK); the CGT and inheritance-tax thresholds in force. Cash-flow modelling, Monte Carlo retirement projections, sequence-of-returns risk. Trust and estate structures. Fee and platform-cost schedules.

Subjective. The adviser's judgement of the gap between stated risk tolerance and actual risk capacity. Behavioural patterns: who panics in a drawdown, who chases yield, who mistakes concentration risk for conviction. Which family relationships will complicate estate planning. Which provider's fact sheet is reliable, and which one makes the fee schedule hard to find.

Procedural. The structure of an annual review. The order of operations after a major life event: inheritance, redundancy, divorce, business sale. When to bring in a tax specialist, an estate lawyer or an insurance specialist. How to word a risk warning that holds up to regulator review (FMA / FANZ in NZ; FCA in the UK; SEC / FINRA in the US).

Question archetypes. Portfolio review. Withdrawal-strategy planning. Tax-efficiency questions. Major-life-event response. Product comparison and platform selection. Off-topic: legal advice, insurance underwriting, any product recommendation outside the firm's authorised menu.

Constraints and safety. No product recommendations outside the authorised investment menu. No tax opinions beyond firm scope. No predictions of market direction. Suitability rationale recorded in the regulator's framework, every time. AML/KYC protocols cannot be bypassed. Fiduciary duty cannot be delegated to AI.

Grounding constraint. Every recommendation is tied to the client's actual portfolio, their documented goals, the suitability assessment on file, and the firm's investment-committee positions. Not generic personal-finance advice.

Law Practice

Quantitative. The controlling statutes and binding authorities in each practice area. Limitation periods, statutory deadlines, court timetables. Quantum tables and settlement ranges from prior comparable matters. Costs schedules: court costs, billing rates, the firm's realisation and utilisation. Statutory thresholds that change the analysis (jurisdictional thresholds, summary-judgment thresholds, Companies Act notification thresholds, Privacy Act breach-notification triggers). The matter-management metrics partners manage by: WIP, lock-up days, leverage, profit per equity partner.

Subjective. "Something is wrong with this deal." The patterns a senior partner notices and a junior takes years to learn: earnouts structured to hide risk, indemnity caps that do not cover the actual exposure, reps that do not match the diligence findings, limitation-period and waiver issues that are about to become a problem. The judge's or arbitrator's preferences: who is procedural, who is commercial, who takes a long view. The counterparty firm's negotiating approach: which firms argue in good faith, which are positional. Judging whether a witness's evidence will hold up under cross-examination. Client temperament: who wants reassurance, who wants direct counsel, who will give way under regulatory pressure and who will contest it.

Procedural. The structure of a competent advice memo (issue → facts → analysis → conclusion). The order of operations in a transaction: term sheet → DD → drafting → negotiation → completion → post-completion. The order of operations in litigation: pleadings → interlocutories → discovery → briefs → hearing → judgment → enforcement (or settlement at any stage). Privilege protocols: what is solicitor-client, what is litigation privilege, what loses privilege when it is circulated. How to word advice that is clear without overstating certainty. When to brief counsel and when to handle in-house. The conflict-check protocol. Waiting a day before sending anything drafted in anger.

Question archetypes. Contract review and negotiation. Statutory and regulatory interpretation. Litigation strategy and risk assessment. Due-diligence summarisation. Precedent and template selection. Drafting (clauses, advice memos, letters of demand, opinions). Compliance interpretation. Off-topic: legal advice without admitted-lawyer sign-off, and financial or tax advice the firm does not authorise.

Constraints and safety. No legal advice without sign-off from a practitioner admitted in a jurisdiction where the firm practises. Privilege must be preserved: privileged material is never sent to a third-party model that retains data. Conflict checks are mandatory. Court-imposed deadlines are absolute, and are always shown to the lawyer, never handled silently. Ethical judgements that belong to the partner (when to withdraw representation, when to disclose to a tribunal, candour duties) cannot be delegated. AML/CFT obligations are statutory and follow firm protocol. Citations to authorities must be verifiable. Invented case citations are a known failure of legal AI and have ended careers. The system cites from the firm's verified precedent database and never generates a citation.

Grounding constraint. Every recommendation is traceable to the firm's prior matters on the issue, the controlling statute and binding authorities (verifiably cited), the facts of the matter on file, and the partner's documented position. Not generic legal commentary, and never an unverified citation.

Environmental Consultancy

Quantitative. Site-assessment frameworks (Phase I / Phase II ESA, Preliminary Site Investigation). Contaminated-land standards (NES-CS in NZ, RBSL/CCME in North America, your jurisdiction's equivalent). Trigger values for soil, sediment, groundwater, surface water. Modelling tools and their assumptions: dispersion (AERMOD), groundwater (MODFLOW), noise (SoundPLAN). Regulatory frameworks (RMA, NEPA, Clean Air Act). HHRA / ERA risk-assessment frameworks.

Subjective. "I've seen a site like this before." Knowing the likely contaminants from the historical land use. Field judgement: which monitoring well looks suspicious, which sample is unrepresentative, which iwi or hapū consultation will need more time. Which council officer is risk-averse on which issue. The estimator's doubt about a scope that was priced too low.

Procedural. The structure of a Phase I assessment. How to structure a consent application narrative. When to recommend additional sampling and when to accept the data as adequate. How to write a finding that holds up to an expert-witness challenge in the Environment Court.

Question archetypes. Site-assessment scoping. Contaminant pathway analysis. Consent or permit application support. Environmental impact statement drafting. Regulatory interpretation. Iwi and community engagement comms. Off-topic: legal advice, and anything that would amount to certification.

Constraints and safety. No certification of findings. That is the licensed engineer's or scientist's decision. No definitive predictions of regulatory decisions. Iwi and Treaty-settlement matters defer to the relevant team and tikanga. Any contamination above trigger values is referred for licensed review, not resolved inside a recommendation.

Grounding constraint. Every recommendation is tied to the actual site data, the relevant standard, the consultancy's prior comparable assessments, and the licensed practitioner whose name will go on the report.

Residential Construction / Builders

Quantitative. Building codes (NZ Building Code; IBC/IRC; your jurisdiction). Bracing and lateral-load calculations. Span tables. Material take-offs and waste factors. Pricing: SQM rates, labour productivity assumptions, sub-trade rate cards. Programme dependencies: foundation cure, frame, cladding, lockup, lining. Margin and overhead percentages. Retention schedules. The H&S risk register the firm uses on site.

Subjective. What a site visit shows: "this clay subgrade is going to need engineered fill." Sub-trade reliability: which framer can work to a tight programme, which electrician will leave you waiting two weeks. Client temperament: who wants a fixed price and no involvement, who needs weekly visits. Which BCO is strict about which detail. The estimator's doubt about a price that came in too low.

Procedural. The structure of a variation conversation with a client. The order of operations in a typical build. When to call the engineer and when to decide on site. How to write a defect notice that holds up in a building dispute. The pre-line walkthrough.

Question archetypes. Estimating and take-off support. Programme and sequencing decisions. Variation pricing and comms. Code-compliance interpretation. Client communication drafting: variations, delays, defect responses. H&S incident response wording. Off-topic: Producer Statements, engineering sign-off, legal positioning in contract disputes.

Constraints and safety. No structural-design certification. That is the CPEng's decision. Consent-condition interpretation defers to the council engineer. Weathertightness risk is flagged for senior review, not decided. Contractual disputes go to legal. H&S incident reporting protocols are mandatory.

Grounding constraint. Every recommendation is tied to the job's specs and programme, the firm's standard details, prior comparable jobs, and what was agreed in writing. Not generic construction advice.

Logistics and Supply Chain

Quantitative. Network metrics: OTIF, fill rate, lead time, cost-per-mile, inventory turnover, demurrage and detention exposure. Forecasting models: ABC classification, exponential smoothing, ARIMA. Routing maths: VRP variants, line-haul utilisation, dock-door scheduling. SLA thresholds for each customer tier. Carrier-rate cards in force.

Subjective. The dispatcher's judgement that "this driver isn't going to make it through customs by 5." The carrier reliability ranking that is not in any system: who handles a peak-season exception calmly, and who does not cope. Which lanes have weather risk in which season, which ports have a history of labour action, which customer's procurement team always disputes the recovery invoice. Seasonal habits such as "we always pre-position there in October."

Procedural. The exception triage decision tree: delay confirmed → reroute attempt → customer comms → cost-recovery path. The order of comms when something goes wrong: which customer first, which carrier, which executive. The structure of a peak-season standup. How to explain the root cause of a service failure to a customer and keep the relationship.

Question archetypes. Exception triage. Capacity and routing decisions. Customer-service comms drafting. Carrier performance review. Forecast adjustment. Tender response support. Off-topic: anything that touches safety regulation or commits to rates outside authorised tariffs.

Constraints and safety. No commitments to rates outside approved tariffs. Dangerous-goods classification stays with a certified DGSA. Customs and compliance protocols cannot be bypassed. Hours-of-service and driver-safety rules are statutory and cannot be overridden.

Grounding constraint. Every recommendation is tied to the current state of the network, the carrier's documented performance for this customer, and the contract terms in force. Not generic supply-chain best practice.

Workflows Fit the Document Substrate to Your Firm

A written substrate needs software to apply it to real work. That software is the Document Substrate, and a workflow is how it is fitted to your firm. A workflow defines a job or engagement. It has three parts.

Slots. The documents the job needs, each named. For a rental property: the bank statement, rent summary, loan statement, rates bill and invoices.
Rules. Checks that run in code over the figures read from those documents: do they agree, is a month missing, is anything counted twice. No model takes part, so the same documents always get the same verdict.
Outputs. A report on screen and a workpaper in Excel or Word, stamped DRAFT until every slot is filled and every rule passes.

Example rental property workflow for an accounting practice: five document slots with the loan statement missing, rules run by code with no model, and two outputs, a report on screen and a workpaper stamped DRAFT until complete

The Document Substrate is the same for every firm. The workflows, and the practice manual that explains each rule, are what change from one firm or industry to the next.

How Document Substrate works, explained simply

Now Do This For Yours

If you recognised your section, the next step is to draft your own. Do it quickly, on paper, and do not try to make it complete.

Three questions help:

Pick a recurring question your senior people answer well, and ask: what would the answer have to be based on to be defensible? That list of inputs is most of your quantitative layer, and it is what every answer has to be tied back to: the metrics, the standards, the prior cases, the firm's documented position. Write them down.

Pick something a junior gets wrong in their first six months, and ask: what does the senior person know that the junior does not yet? It is almost always in the procedural or subjective layer: a procedure or a judgement call that nobody has written down. It is the most valuable thing to capture, because at present only a few people have it.

Pick an hour of work someone repeats every week, and ask: how much of it needs them? This is a better place to start than asking what can be automated.

You do not need a complete list. You need enough to know what every output has to be tied to, what the AI has to refuse to do, and what kinds of question it has to recognise.

Map → Architect → Build, in practice is a structured version of this exercise, scoped to your business, that ends with a concrete plan.

Why "substrate"? explains the choice.