For most of its history, professional advisory has been priced and delivered as a function of time. A diagnostic took six weeks because it took six weeks for a team of analysts to gather documents, read them, reconcile the numbers and draft the findings. Clients paid for those hours, and firms staffed engagements with pyramids of junior people whose main job was to process information so that a small number of senior people could interpret it.
That model is now under genuine pressure. Not because artificial intelligence can replace an experienced advisor (it cannot) but because it has collapsed the cost of the processing layer that sat underneath advisory judgment. When the reading, extraction, reconciliation and first-draft work can be done in hours rather than weeks, the question every client should be asking changes. It is no longer "how many people will you put on this?" It is "how much of your senior team's thinking will I actually get?"
This article sets out how we see the shift, where AI genuinely changes the quality of advisory work, where it does not, and what clients should expect, and demand, from the firms they engage.
The processing layer is being automated
Consider what a typical regulatory gap assessment involved five years ago. A team would collect several hundred policies, procedures and board papers. Analysts would read each document, map its content against a regulatory framework clause by clause, and record gaps in a spreadsheet. A manager would review the spreadsheet, and a partner would review the manager's summary. Most of the elapsed time, and most of the fee, sat in the first two steps.
Today, a well-designed AI pipeline can ingest the same document set, extract the relevant obligations, map them against the framework and flag probable gaps with citations back to the source text. The output is not a finished assessment. It is a structured, evidence-linked first pass that an experienced reviewer can interrogate, correct and build on.
The same pattern repeats across disciplines:
- Financial due diligence: extracting and normalising years of management accounts, ledgers and contracts before an analyst begins testing quality of earnings.
- IFRS transitions: reviewing large lease or contract populations to identify the terms that drive recognition and measurement.
- Cybersecurity assessments: correlating configuration exports, vulnerability scans and policy documents to identify control weaknesses.
- Compliance monitoring: screening communications, transactions or product documentation against defined rules and risk indicators.
In each case, the work that used to consume the majority of an engagement's hours is the work that AI now does best: high-volume, pattern-based, well-bounded reading and extraction.
Where AI genuinely improves the work
It would be easy to describe the change purely in terms of speed and cost. Those benefits are real, but in our experience the more important effect is on quality: specifically, on coverage and consistency.
Coverage: from samples to populations
Traditional assurance and advisory work relies heavily on sampling because reviewing every item was never economical. An auditor tests 60 transactions from a population of 60,000. A consultant reviews 40 contracts from a portfolio of 2,000. Sampling is statistically defensible, but it is also a compromise: it tells you about the sample and infers the rest.
When extraction and classification are cheap, the compromise disappears. We can review the entire contract population, every access log, every policy. The anomalies that a sample would probably miss (the one lease with an unusual extension option, the three dormant administrator accounts, the policy that contradicts another policy) surface as a matter of course.
Consistency: the same standard on item one and item two thousand
Human reviewers get tired. The two-thousandth contract does not receive the same attention as the twentieth, and two analysts will often interpret the same clause differently. A well-tested AI extraction step applies the same criteria every time. That does not make it infallible; it makes its errors systematic, and systematic errors can be found and corrected through testing in a way that random fatigue errors cannot.
Speed that changes decisions, not just timelines
When a diagnostic that took six weeks takes ten days, the benefit is not only a shorter project plan. It is that the findings arrive while the decision they inform is still open. A board considering an acquisition, a CFO preparing for a regulatory inspection or a CISO responding to an incident all operate on timelines that traditional advisory frequently could not meet. Faster evidence means advice that is used rather than filed.
Where AI does not replace the advisor
Every serious practitioner we work with shares the same caution, and it is worth stating plainly: the parts of advisory work that matter most are the parts AI is least equipped to do.
Framing the right question. Most engagements begin with a client describing a symptom: slow month-end close, repeated audit findings, a failed transformation. Identifying the underlying problem requires conversations, context and an understanding of organisational dynamics that no model has access to.
Weighing trade-offs that involve values. Should a bank accept a longer implementation timeline to preserve data lineage? Should a retailer adopt a simplified accounting policy that reduces cost but increases volatility? These decisions depend on risk appetite, stakeholder expectations and strategy. AI can inform them. It cannot make them.
Taking accountability. A regulator, an auditor or a board does not accept "the model said so" as a basis for a conclusion. Every significant finding we issue is reviewed and owned by a named professional who can explain it, defend it and stand behind it.
Delivering difficult messages. Some of the most valuable advice we give is advice clients do not want to hear. Delivering it well, with evidence, empathy and a credible path forward, is a human skill, and it will remain one.
AI makes the evidence cheaper. It does not make judgment cheaper, and it should not.
What AI-augmented delivery looks like in practice
The phrase "AI-powered consulting" is now widely used and rarely defined. In our practice, it means a specific operating model with four characteristics.
1. AI is embedded in the method, not bolted on
We do not run a traditional engagement and then use AI to format the report. Each methodology (gap assessment, due diligence, control testing, transformation diagnostic) has been redesigned so that the processing steps are automated and the human steps are concentrated where judgment is needed. That redesign is where most of the benefit comes from.
2. Every output is traceable to source evidence
An AI-generated finding that cannot be traced back to a specific document, clause or data point is worthless in a professional context. Our pipelines keep citations attached to every extracted fact, so a reviewer, or a client's auditor, can check any statement against its source in seconds.
3. Humans review everything that matters
We classify outputs by consequence. Low-consequence steps, such as document classification or data normalisation, are tested through sampling and control totals. High-consequence outputs (findings, recommendations, numbers that will appear in financial statements or regulatory submissions) are reviewed item by item by qualified professionals.
4. Client data stays governed
Client documents are processed in controlled environments with clear rules on retention, access and model usage. Client data is not used to train general-purpose models. For regulated clients, we document the processing arrangements in the same way we would document any other outsourced activity, so they can satisfy their own third-party risk requirements.
The economics are changing, and clients should benefit
If AI removes a large share of the hours in an engagement, the natural question is who keeps the saving. Some firms will quietly bank the margin and continue to price on the old basis. We think that is short-sighted, and clients should push back on it.
In our view, the change should show up for clients in three ways:
- Lower cost for well-defined work. Diagnostics, assessments and reviews that are largely evidence-gathering should cost materially less than they did.
- More senior time per engagement. The saving from automation should be partly reinvested in experienced people spending more time on interpretation, stakeholder engagement and implementation support.
- Pricing tied to outcomes and scope, not hours. When effort is no longer a good proxy for value, fixed-fee and outcome-linked arrangements become more appropriate and more transparent.
Questions clients should ask their advisors
Whether or not you work with us, these are the questions we would encourage any client to put to a prospective advisory firm:
- Which steps in your methodology are automated, and which are performed by people?
- How do you verify AI-generated outputs before they reach us?
- Can every finding be traced to the underlying evidence?
- Where is our data processed, how long is it retained, and is it used to train any model?
- How has your use of AI changed your pricing and team structure?
- Who is personally accountable for the conclusions in the final report?
A firm that cannot answer these clearly is either not using AI in a meaningful way or not governing it properly. Neither should be acceptable to a client in a regulated industry.
The risks we take seriously
Being candid about the limitations is part of using AI responsibly. The main risks we manage on every engagement are:
- Confident errors. Language models can produce fluent, plausible text that is wrong. This is why traceability and human review are non-negotiable rather than optional extras.
- Automation bias. Reviewers can become less critical when presented with polished machine output. We counter this with structured review checklists and by deliberately testing reviewers with seeded errors during quality assurance.
- Hidden gaps in source data. AI can only analyse what it is given. If key documents are missing, the analysis will be confidently incomplete. Scoping and data-completeness checks remain a human responsibility.
- Regulatory expectations. Supervisors increasingly expect firms to explain how AI is used in processes that affect financial reporting, customers or risk management. Documentation and model governance are part of the deliverable, not an afterthought.
What this means for the next five years
We expect the advisory market to separate into two groups. The first will continue to sell time, supported by AI tools used informally by individual consultants. The second will rebuild their methods around AI, pass a meaningful share of the efficiency to clients, and compete on the quality and seniority of their judgment.
For clients, the practical implication is straightforward. You should expect faster answers, broader coverage and more time with experienced people, at a price that reflects the reduced effort. You should also expect transparency about how the work is done and who stands behind it.
That is the standard we hold ourselves to. It is why we describe our model as advisory excellence amplified by AI, rather than AI in place of advisors. The technology is extraordinary. The judgment, accountability and relationships remain human, and they remain the reason clients engage us in the first place.