The AI Marketing Audit: Assess Before You Automate

Why audit first

AI makes whatever process it touches faster. If the process is sound, that is good news. If conversions are mis-tracked, the landing page leaks, or the email list was collected without the right permissions, automation multiplies the damage and makes it harder to see. An audit before automation answers one question: which parts of your marketing are sound enough to speed up?

This is different from a readiness score. The readiness checklist asks whether the business is prepared to invest. The audit below inspects the marketing itself and produces a list of findings with evidence.

The five areas

The five-area audit

Area	What to check	Evidence to collect	Red flag
Measurement	Whether the actions that matter are tracked and match your own records	Analytics key events list; a month of platform conversions next to orders or CRM leads	Platform figures that differ from your records by more than you can explain
Journey	Where people drop out between first visit and conversion	Visits, landing page views, first conversions and repeat conversions for the last 90 days	Most traffic landing on pages with no clear next step
Workload	Which repeatable tasks consume the most time	A one or two week time log by task	Nobody can say where the marketing hours go
Context	Whether the business is written down well enough to brief anyone	Offer, customer, proof, tone and banned claims documents	Key facts exist only in someone's head
Permissions	Whether you may use the data and channels you plan to automate	Where each list came from and what people agreed to	Bought or scraped lists used for email to individuals

Measurement

Start here because every other finding depends on it. In Google Analytics, the actions that matter to your business should be marked as key events; Google's help documentation describes a key event as an action that is particularly important to the success of your business. List yours and check that each one fires once per real action.

Then compare a month of conversions reported by each ad platform with your own order or CRM records. Some difference is normal: platforms attribute differently, and Google Ads help describes consent mode modelling, which estimates conversions for visitors who decline cookies and reports them alongside observed ones. Check your own account's conversion settings to see whether modelling applies to you. The point of the audit is not to make the numbers identical; it is to know how and why they differ, so an automated system is not optimising towards a figure you cannot trust.

Journey

Take the last 90 days and write down the number at each stage: visits, views of the pages that matter, first conversions, and the next valuable action (a repeat order, a sales call, an upgrade). Divide each stage by the one before. The lowest ratio relative to what you would reasonably expect is usually where the money is. If that stage is a landing page, generating more traffic with AI will not help until the page is fixed.

Workload

Ask everyone who does marketing work to log their time by task for a week or two. You are looking for tasks that are repeatable, frequent and rule-based, because those are where AI saves real hours. Tasks that are rare or depend on fine judgement rarely repay the effort of automating.

Context

Collect whatever describes the business: offer, pricing logic, customer description, testimonials you have permission to use, tone of voice, words you avoid, claims you cannot make. Grade it simply: could a capable stranger write something accurate and on-brand from this alone? If not, writing that document is a finding in its own right, and it comes before any content automation.

Permissions

For each list or channel you might automate, record where the contacts came from and what they agreed to. The Information Commissioner's Office explains that sending unsolicited marketing email to individuals requires their consent or all the conditions of the soft opt-in, and that the soft opt-in does not cover bought-in lists. Read the ICO's current guidance before automating email to any list you did not collect yourself, because an automation that emails the wrong list simply breaks the rules at scale.

Grading the findings

Write each finding in one line with its evidence and a severity:

• • Critical: automation would cause harm or waste (broken tracking, unlawful list, a leaking checkout).
• • Important: automation would work but underperform (thin context, unclear journey data).
• • Minor: worth fixing when convenient.

Illustrative example of a finding: "Critical. The contact form key event fires on page load as well as on submission, so reported leads are roughly double the CRM count. Evidence: event debug view and a March comparison."

The decision rule

Sort every workflow you were thinking of automating into one of three lists:

1. • Fix first: anything touching a critical finding. Do not automate until it is resolved.
2. • Automate: repeatable, frequent tasks whose inputs are sound, whose output can be checked, and whose outcome you can measure.
3. • Leave alone: rare, high-judgement or relationship-heavy work where AI can assist a person but should not run the task.

Most audits produce a short automate list and a longer fix first list. That is a good result: it tells you where the next month should go.

Doing it yourself or getting an outside view

You can run this audit internally in one or two weeks. An outside view helps when the people who built the tracking are also the ones checking it. AI for Marketing's free written breakdown is a narrower starting point: it uses public information and any figures you choose to share, so it cannot inspect private analytics, but it can point to where the journey appears to leak.

FAQ

How long does an AI marketing audit take?

One to two weeks for a small or mid-sized business, mostly because the workload area needs a real time log. The measurement and permissions checks can often be done in a day each.

Who should run the audit?

Someone who can read analytics and ad platform reports and who will be honest about gaps. Ideally not only the person who set up the tracking, because people rarely spot their own configuration mistakes.

What if the audit finds nothing ready to automate?

Then the fix first list is your plan for the next month. Fixing tracking and the biggest journey leak often improves results before any AI is added, and makes later automation measurable.

Is this the same as a data protection audit?

No. The permissions area only checks whether the lists and channels you plan to automate can lawfully be used that way. A full data protection review covers far more and may need specialist advice.