AI for compliance consultancies: the policy was never the work

November 22, 2026

A client asks ChatGPT for an information security policy. 40 seconds later there is a tidy 6 page document with headings, roles and a revision table. Then the same client opens your quote of 18,000 euro for an ISO 27001 project and asks the question you hear more often this year: what am I actually paying for?

The honest answer is that the policy was never the work. The work is a scope that holds, a risk assessment that fits this company, the evidence you collect from the client month after month, the justification for every control, and year 2, when everything has to hold again for the surveillance audit. That part costs the hours. It is also where AI for compliance consultancies can do the most, as long as the consultant keeps making the calls.

This article is for owners and managing directors of compliance consultancies with 10 to 200 people: ISO 27001, ISO 9001, NEN 7510, NIS2 and the Dutch Cybersecurity Act, GDPR, BIO. You run dozens of projects a year and keep a steady base of clients in maintenance. Your growth stops at the number of consultants you can find. It is not for you if you are the compliance officer of 1 company, a SaaS startup that wants SOC 2 through a tool, or an independent consultant with 4 clients.

In this article you will learn where the hours in a compliance project really go, why the policy is the cheapest part, which part AI does and which part fixed rules do, a worked example with 60 projects and 140 clients in maintenance, what not to automate, and how to start without putting your own clients at risk.

‍

Key takeaways

• Since 15 August 2026 the Dutch Cybersecurity Act applies to more than 8,000 organisations. Demand for consultants is rising. Supply is not.

• Writing policy documents is the smallest part of a project. Collecting, checking and maintaining evidence is the largest.

• A generic AI tool writes a generic policy. Your value is your standard: your templates, your judgement per control, your past files.

• The core of the automation is an evidence register: 1 line per control per client, with the evidence, its date, its owner and your verdict.

• AI reads what the client sends and maps it to controls. Fixed rules check that everything is complete and current. The consultant decides and signs.

• 1 piece of evidence often counts for ISO 27001, the Cybersecurity Act and NEN 7510 at once. Record it once and you sell it 3 times.

• In the worked example a project drops from 180 to 117 hours. With the same team the firm goes from 60 to about 90 projects a year.

• Keep scope, risk acceptance and the verdict on evidence with the consultant. An auditor asks those questions to a person.

‍

AI for compliance consultancies: why a policy generator does not fix it

The market is full of tools that write policies. You enter a few details, pick a standard and get 30 documents back. For a client that has nothing yet, it feels like progress. For you as a consultant, it feels like competition on the part you already did fastest.

How long does it take to write a policy once you have guided 80 clients? You have templates, you adapt them, you get them reviewed. That is a day, maybe 2. The weeks of a project sit somewhere else.

They sit in deciding which systems are in scope and which are not. In the risk assessment where the managing director hears for the first time what happens if the planning software supplier is down for a week. In the 40 emails it takes to get an export of access rights. In the screenshot of the backup settings that turns out to be from last year. In the internal audit, the management review, and the moment the auditor asks why control 8.12 was excluded.

A policy generator does nothing for those hours. Worse, it makes the client feel the work is done when it has barely started. That is why compliance consultancies with an AI tool often deliver documents faster and still do not run more projects.

‍

Where the hours in a compliance project really go

Ask a consultant where the time goes and you rarely hear "writing documents". You hear: chasing evidence. That work takes a few fixed forms, and they come back with every client.

• Scope and baseline: interviews, a walk through the systems, the gap between what the client thinks happens and what does.

• Risk assessment: threats and vulnerabilities per asset, a score, a treatment, and a board that has to sign for what it accepts.

• Documentation: policies, procedures, and the statement of applicability with a justification per control.

• Evidence: exports, screenshots, logs, contracts, training records. Request, check, send back, request again.

• Internal audit and management review: samples, findings, corrective actions, and the records of all of it.

• Certification audit support: preparing, attending, closing nonconformities.

Evidence is where it goes wrong. Not because it is hard, but because it is scattered. It lives in email threads, in a shared folder per client, in a spreadsheet each consultant keeps a little differently. If you want to know how far a client is, you ask the consultant. And if that consultant is ill or leaves, a colleague starts searching from scratch.

We saw the same pattern at a subsidy advisory practice: 5 advisors worked 5 ways in 5 tools. The craft was the same, the records were not. Compliance work has that problem even more, because every standard uses slightly different words for the same measure.

‍

The policy is cheap, your standard is the value

This is the shift most firms miss. A generic AI can write a generic policy. Every client can now do that themselves. What a client cannot do is apply your judgement: which controls you exclude for a 40 person company that runs only SaaS systems, which evidence you accept for access management, which wording an auditor accepts without a discussion.

That judgement lives in the heads of your best consultants and scattered across past files. As long as it stays there, no AI can use it. Neither can a new consultant, except by shadowing for 2 years.

So the first step is not buying an AI tool. It is writing down your standard: per control what you expect, which evidence counts, what the typical deviation looks like and how you assess it. That is the same foundation we describe in the article on an AI knowledge base for service firms. Without that standard, AI mostly gives you a faster junior.

With the standard in place, the client's question changes. They no longer pay for a document ChatGPT can also write. They pay for your standard, applied to their company, with a name underneath.

‍

The evidence register: what AI does, what rules do, what the consultant does

The core of a good automation is not a chatbot. It is an evidence register. Per client, 1 line per control or requirement: what the standard asks, which evidence exists, its date, who owns it at the client, and what the consultant thinks of it. Everything AI does updates that register. Everything the consultant does starts from it.

What AI does

• Reads what the client sends, an export, a screenshot, a PDF, and proposes which controls it belongs to.

• Pulls the date, the system and the coverage out of the evidence and records them in the register.

• Drafts policies and procedures from your own templates and past files, not from the internet.

• Drafts the justification per control for the statement of applicability.

• Writes a draft request email per owner at the client, listing exactly what is missing.

What fixed rules do

• Every control has a decision, applicable or excluded, and every exclusion has a justification.

• Every piece of evidence has a date, an owner and a status. Evidence older than your agreed term turns amber automatically.

• Every risk treatment points to a measure, and every measure points to evidence.

• A file only goes to the auditor when no red lines are open.

What the consultant does

• Sets the scope and discusses risk acceptance with the board.

• Judges whether the evidence is sufficient. AI proposes, the consultant ticks the box.

• Runs the internal audit, the conversation with the auditor and the hard discussions.

That split is not a detail. AI is good at reading and sorting, and weak at knowing when something is just short of enough. Fixed rules are boring and therefore reliable. The consultant is expensive and therefore used where judgement is needed. It is how we built it for a BPM file processor too, where a file went from 4 to 8 hours to under 5 minutes: AI reads, code applies the rules, a person signs.

AI for compliance consultancies: consultant hours per ISO 27001 project today and with AI, from 180 to 117 hours, and who decides each step

Want to put this on paper for your own firm before you talk to anyone? The free AI product requirements template helps you put the processes, the evidence, the integrations and the cost of a mistake in 1 place. It also gets you quotes you can compare.

‍

Record it once, sell it 3 times

The Cybersecurity Act exposes another advantage. Many clients now face several frameworks at once. A healthcare provider has NEN 7510, may fall under the Cybersecurity Act and wants ISO 27001 for a tender. A government supplier works with the BIO.

The words differ, the measures mostly overlap. Access management, backups, supplier management, incident reporting, awareness. If you keep a separate file per standard, you ask the client 3 times for the same export.

In an evidence register you record once that the access rights export of 2 October is sufficient. Then you link it to the control in ISO 27001, the duty of care under the Cybersecurity Act and the requirement in NEN 7510. That mapping table is your work and your asset. AI helps keep it current when a standard changes.

For the client it means fewer emails. For you it means a second or third framework is not a new project from zero, but an extension of a file that already exists. According to Forvis Mazars, organisations under the act must be able to show that they identified their risks and chose their measures on that basis, and the board keeps final responsibility. That is exactly what a maintained register shows.

‍

Worked example: a firm with 22 consultants

An example, not a client. A compliance consultancy with 22 consultants runs 60 new ISO 27001 and NIS2 projects a year. It also has 140 clients in maintenance, each with an annual surveillance audit. A new project takes 180 consultant hours on average. A maintenance client takes 40 hours a year, half of it collecting and checking evidence.

The hours of a new project, today and with AI and an evidence register:

• Scope and baseline: 30 hours becomes 22, because AI reads the documents upfront and the interviews get shorter.

• Risk assessment: stays at 25 hours. This is judgement and board work.

• Documentation and statement of applicability: 35 hours becomes 15, with drafts from your own standard.

• Collecting and checking evidence: 50 hours becomes 20, through mapping, request emails and automatic checks on dates and completeness.

• Internal audit and management review: 25 hours becomes 20.

• Certification audit support: stays at 15 hours.

Total: from 180 to 117 hours per project. That is 63 hours less, times 60 projects, so 3,780 hours a year. For maintenance clients the evidence part goes from 20 to 10 hours and the rest from 20 to 18. That is 12 hours per client, times 140, so 1,680 hours.

Together that is 5,460 hours a year. A consultant bills about 1,400 hours. That is the capacity of almost 4 consultants, without a vacancy. Put differently: with the same team the firm can run about 90 projects a year instead of 60.

What does it cost? In this example a Blueprint of 7,500 euro, a foundation for the standard and the register of around 20,000 euro, and 2 automations: the evidence register with mapping, and the document drafts, together 65,000 euro. That is 92,500 euro, plus 2,500 euro a month to run. Say the firm sells only half of the freed hours, at 110 euro an hour. That is 300,300 euro in revenue a year. Payback is then well under a year.

These numbers are an example. Your own numbers come from measuring your own projects, not from an article. But the ratio shows up at almost every firm: the evidence costs more than the policy.

‍

What not to automate

There is a line, and in compliance it is sharper than in most sectors. You sell trust. A mistake an auditor finds costs you more than a few hours.

• The scope. What falls inside and outside the management system is a choice with consequences. The consultant makes it with the client.

• Risk acceptance. A board that accepts a risk has to do so itself and understand it. Under the Cybersecurity Act that is explicitly the board's responsibility.

• The verdict "this evidence is sufficient". AI may propose it, but the tick belongs to a person with a name.

• The conversation with the auditor. An auditor asks why. A consultant should answer that, not generated text.

Then there is the data. Clients send you their access rights, their network diagrams and their incident logs. Those do not belong in a free chatbot. A security consultant who pastes a client's firewall export into a public AI tool has written the first nonconformity in the file personally. A proper setup keeps the data in the EU, separated per client, with a log of who saw what. It is also the first thing your clients will ask you about.

‍

How to start

Do not start with the tool. Start with 1 standard and 5 closed files.

• Pick the standard where you have the most volume, usually ISO 27001.

• Take 5 recently closed projects and measure where the hours went. Use your time records, not your memory.

• Write down with your 2 best consultants what you expect per control and which evidence counts. That is your standard.

• Build the evidence register for 1 client type and let AI only map and sort. No documents yet.

• After 3 months, measure how many request emails and hours each file saved. Only then add the document drafts.

The order matters. Start with documents and you speed up the part that was already fast. Start with the register and you hit the hours that make a project long. It is the same logic as in our piece on how to productise your consulting service with AI: find the part that is the same every time and make it repeatable.

‍

From project to subscription

There is a step further, and for many firms it is the real opportunity. Once the register exists, the client can look into it. Upload evidence, see what is missing, get a notice when something expires. Maintenance then stops being a yearly sprint before the surveillance audit. It becomes an ongoing service at a fixed monthly price.

That is the difference between a firm that sells hours and a firm that sells a product. The consultant is still needed for the judgement. But the volume comes from the platform, not from more people. This is usually a year 2 step. First the register has to work inside your own team.

‍

Who builds it, what happens after launch, and if month 1 disappoints

Codelevate turns Dutch B2B service companies into software companies: we map the expert process you run by hand, automate it so the same team handles the volume, and build the platform your customers pay for, with 2 engineers by name from the blueprint to year 2.

Who works on it? The same 2 engineers who read your files and wrote down your standard with your consultants. No rotating team, no handover after the sale.

What happens after launch? From month 4 it runs under Operations: monitoring, a named engineer on the alert, the AI cost per file capped, a 1 page monthly report, and updates when a standard or a model changes. The response time is set in the Blueprint. A new version of ISO 27001 or a change to the Cybersecurity Act then becomes a planned update, not a new project.

And if month 1 disappoints? Exit after month 1: pay for what is delivered, keep everything. Your standard, your register and your code stay yours.

Want to work on it yourself first? The free AI product requirements template helps you map your own projects before you have a conversation with anyone.

‍

Send us what you have

Send us what you have: an overview of your projects, an anonymised file or just your time records. 1 of the 2 engineers reads it with you in 20 minutes, and you get 1 of 3 answers in writing the same day: do it in house, hire, or the Blueprint.

Table of Contents
Share this article

Common questions

How can AI help a compliance consultancy?

Mostly in evidence, not in policies. AI reads what the client sends, maps it to controls, flags missing or outdated evidence and drafts request emails, while the consultant judges and signs.

Can AI write ISO 27001 policies?

Yes, but that was never the expensive part, and a generic tool writes a generic policy that an experienced consultant would adapt anyway.

What is an evidence register?

One line per control per client with the evidence, its date, its owner at the client and the consultant's verdict, so anyone can see how far a client is.

Does the Dutch Cybersecurity Act change anything for consultancies?

Yes, since 15 August 2026 it applies to more than 8,000 organisations, so demand rises and evidence mapped once can count for ISO 27001, NIS2 and NEN 7510.

What should a compliance consultancy not automate?

The scope, risk acceptance, the verdict that evidence is sufficient and the conversation with the auditor, because those need a named person.

Is it safe to put client evidence into an AI tool?

Not in a public chatbot; a proper setup keeps data in the EU, separated per client, with a log of who saw what.

Get started with
an intro call

This will help you get a feel for our team, learn about our process, and see if we’re the right fit for your project. Whether you’re starting from scratch or improving an existing software application, we’re here to help you succeed.