How to choose a digital marketing agency: 13 things to check

How to choose a digital marketing agency comes down to five questions. Who will do the work, and which work does the proposal cover? Can the agency prove its claims with your data, and which recommendations does its fee model reward? When you part ways, who keeps the accounts and the data? A pitch deck, a client list or a badge does not answer any of them on its own.

The 13 checks below work through those questions in order, from preparing before the first call to reviewing the first 90 days. The fee, ad account and measurement examples come from paid media, where platform rules are written down and can be checked. The team, scope and production checks apply the same way to any service, such as SEO, content, social media or customer relationship management (CRM). The fee comparison is a hypothetical example, not a market rate. No agency is named, ranked or recommended.

Preparing before you talk to agencies sets the yardstick for every comparison that follows. If you have not written down what you want and where you stand today, you end up comparing pitches on how persuasive they are.

1. Write down the goal, the success metric and today's number

The goal should be one business outcome, such as new customers, cost per new customer or the margin from one product line. "More visibility" is not a goal, because it does not say what gets counted. The performance marketing metrics guide maps each metric to the decision it informs.

Define the metric precisely: what is counted, per which unit and over which period. For example, a new customer is someone placing a first order in your order system, counted monthly, with cancellations and returns removed. Pull the last few months from your own system using that definition. If you calculate customer acquisition cost (CAC) the same way with the CAC calculator, you have a baseline for every number the agency reports later.

There is a reason to take the baseline from your own system rather than the ad platform. Google Ads counts conversions using the tracking settings and attribution rules set for each conversion action. If the agency changes those settings, the platform number moves. Your order system does not.

2. List who controls each account and data source today

Before the calls, write down who owns what. If you have worked with an agency or freelancer before, the list may hold surprises. Check these:

• People with Admin access to the Google Ads account, and any manager accounts linked to it
• The person or company named in the "who pays" field of the payments profile
• People with the Administrator role in the Google Analytics 4 (GA4) property
• People with full control of the Meta business portfolio
• Who manages the website tags and campaign links

This list is the starting point for what check 10 puts into the contract. Google warns that an account with a single administrator can lose access to its tags if that person becomes unavailable. It suggests adding a second administrator.

Checking an agency's claims means tracing each case study back to its evidence. That is the account data, the measurement method and the people who will do the work. The next five checks do that step by step.

3. Ask for read access to a real account, or a screen share

The most direct way to check a case-study number is to look at the account's own report. Ask the agency to walk you through an account it manages by screen share, with that client's consent. Temporary read access works too. In Google Ads, Read-only access lets a user view campaigns but not edit them.

Look for three things in the account. Check that the case-study figure matches the report for the same date range. See which actions count as conversions, and whether the result sits next to the spend that produced it.

Client confidentiality may rule this out. In that case, ask for an anonymised report with its definitions and date range written out.

4. Ask who will manage your account

An agency's references show the firm's track record, not the team that will run your account. This check moves the evaluation to the people assigned to you. Ask these questions in the meeting:

Will the person pitching also manage the account, or do they sit in sales? Who is the responsible specialist, and how long have they worked on which channels and account sizes? How many clients do they handle at once, and how much time a week does that leave for yours? Which work, such as creative production, the product feed or measurement setup, goes to another team or an outside supplier? If the specialist changes, how is the handover done and when are you told?

These questions have no right answer, only consistent ones. Ask to speak with the specialist during the proposal stage, and consider naming them in the contract. The Google Partner rules only require half of the agency's account strategists to be certified. So ask separately whether the person assigned to you is certified.

Once work starts, the Google Ads change history shows who manages the account. According to Google, it lists changes from the past two years. For changes made in the interface, it shows the e-mail address of the person who made them.

5. Ask how they measure incrementality

Incrementality counts the conversions that would not have happened without the ads. Google describes incrementality experiments as tools for understanding how well ads drive the action you care about. Conversion Lift splits the audience into a group that sees ads and one that does not, then measures the difference in conversions. The same page notes that it is not available for every account.

Ask the agency which method it uses, how it builds the control group, how long tests run and which decision a result changed. The control group guide explains how such groups are formed. A good answer names a method, states its limits and explains how it judges whether a gap is chance. The statistical significance guide covers that logic.

An answer that only cites platform return on ad spend (ROAS) answers a different question. In Google's definitions, ROAS divides attributed conversion value by total spend. Incremental ROAS (iROAS), in user-based Conversion Lift, divides incremental conversion value by total spend in the test.

6. Ask which metrics they report, and why

Ask for a sample report, and for each metric ask for its definition, its source and the decision it informs. The source is either the platform or your order system. Impressions and reach do not show a business result by themselves. If a report leads with them, ask where the sales and cost figures are.

Splitting traffic by source takes two separate checks. For Google Ads, ask whether auto-tagging is on and Google Ads is linked to GA4. Google calls auto-tagging the recommended approach for GA4 and keeps manual tagging for special cases. For campaign links on Meta, e-mail and other channels, ask for one UTM (Urchin Tracking Module) naming rule. The UTM builder produces the parameters.

You can check the arithmetic of a reported ROAS with the ROAS calculator. Does the report's revenue and spend give the same ratio? The calculator does not check attribution. Entering your company's total revenue does not verify platform ROAS either. Total revenue includes orders that would have come without the ads, and sales from other channels.

7. Read the Google Partner badge for what it measures

The Google Partner badge rests on performance, spend and certification requirements. According to Google's Partner requirements page, they are checked daily. The table shows each threshold and what it is measured on.

The certification requirement is capped at 100 users. Premier Partner adds one condition: the agency must rank in the top 3% of participating companies in its country. That cut is set once a year. It looks at factors such as client spend growth, client retention and product mix.

The thresholds look at the agency's registered manager account, the client accounts it manages and its account strategists. None of them looks at your account or at the specialist assigned to you. The badge shows that the agency works in Google Ads at a certain volume. It does not show what results you will get in your sector.

The optimization score is based on Google's recommendations, which partners can apply or dismiss. Asking which recommendations the agency dismisses, and why, tells you more than the badge.

When you compare proposals, what the fee covers and what it depends on matter as much as its size. Scope shows whether you are comparing prices for the same work. The fee model decides which recommendations earn the agency more.

8. Put the scope of each proposal side by side

Two fees cannot be compared if they do not cover the same work. One agency may price campaign management only, while another includes creative production and measurement upkeep. Fill in the table below for each proposal.

Production capacity belongs in this table too. If creative or content work is in scope, ask how many pieces arrive each month, who makes them and how fast. If tests need new creatives, that number sets how often a campaign can change. If landing page support is in scope, ask which changes and speed fixes it includes. The landing page speed guide covers how speed affects the ad side. For an SEO service, you can fill the same table with the tasks in the SEO checklist.

The price of out-of-scope work is the row people miss most. A low monthly fee can cost more overall if a frequent task is billed separately.

9. Work out the incentive each fee model creates

Agency fees are usually a percentage of ad spend, a fixed monthly fee, a performance fee or a project price. Some deals combine them. The example below compares the first two.

Hypothetical example: percentage of spend vs fixed fee

Agency A charges 12% of monthly ad spend. Agency B charges a fixed $2,400 a month. B's scope is assumed to stay the same as spend grows. The amounts were chosen for the example and are not market rates.

The two fees meet at $20,000 of spend. Below that A costs less, above it B does. The bigger difference shows up when you ask what the agency earns from a recommendation.

Say the agency proposes raising the monthly budget from $20,000 to $30,000. Those two budgets are the two cases compared. A's fee rises by $1,200, while B's stays put.

Assume a 40% contribution margin. The extra $10,000 breaks even at $10,000 of extra contribution with B and $11,200 with A. Dividing that by the margin gives the incremental revenue the extra budget must bring: $25,000 with B and $28,000 with A.

Dividing that revenue by the extra spend gives the marginal incremental return of the budget increase. This is not Google's iROAS. iROAS looks at total spend, while this ratio looks only at the added $10,000.

The table shows the store's net result at three levels of extra incremental revenue. The first two rows are the break-even points for B and A (2.5 and 2.8). The extra spend, the margin and both fees stay fixed. Only the extra incremental revenue changes.

In all three rows A earns an extra $1,200, while the store's result depends on the row. Under these assumptions, the percentage model raises the bar for extra spend by 12%. The rule that "a percentage fee aligns the agency with growth" holds here only for growth in spend. In the first row spend grows, the store gains nothing with B and loses $1,200 with A.

That does not make a fixed fee better in every case. Under a fixed fee the agency is paid the same regardless of workload or results. If the scope from check 8 is not written down, it becomes unclear what you are paying for.

Whichever model you choose, decide on extra spend from the incremental revenue of the extra budget, not platform ROAS. A lift test that compares people who see ads with people who do not gives the average effect of the whole budget. It does not, by itself, give the effect of the added $10,000.

Add a hypothetical measurement to the same example. Say incremental revenue over a no-ads group is $70,000 at a $20,000 budget and $95,000 at $30,000. The average return falls from 3.5 to 3.17, still above A's bar of 2.8.

But the extra $10,000 brings only $25,000, a marginal return of 2.5, so the store loses $1,200 with A. Seeing the effect of the extra budget takes a test that compares the two budget levels over the same period.

The contract matters most on the day the relationship ends. It should settle who controls the accounts, how access is removed and what the reports count. This section is a checklist drawn from platform rules, not a legal template.

10. Write down account control and how access is removed at exit

In Google Ads, control of the account, payment and manager account ownership are separate matters. Check each one:

1. Does someone from your company hold direct Admin access to the account?
2. Which linked manager account is the owner, and what does ownership let it do?
3. Who pays on the payments profile, and in whose name are invoices issued?
4. Can the account keep running, with its history and payment method, after the agency leaves?

Google describes the payments profile as holding the name and address of the person or company legally responsible for it. That does not say who controls the account. Access levels and manager account ownership do.

A manager account that creates an account becomes its owner automatically. A linked existing account does not grant ownership by default, and Google says linking keeps its history intact. Admins of an owning manager can add and remove users and grant or revoke Admin access in the client account. Google recommends that a manager be made owner only if it needs those privileges.

An agency having created the account is not a problem on its own. According to Google, the client account still owns its data and can always unlink an owning manager. Unlinking takes Admin access in the client account. The real risk is that nobody at your company holds that access, and that the exit is not in writing.

In GA4, only the Administrator role can add and delete users. You can give the agency Editor or Marketer, depending on its work, and keep Administrator yourself.

On Meta, the assets should stay in your business portfolio, with the agency added as a partner. Meta says adding a partner requires full control of the portfolio. A partner given full control can manage an asset but cannot share it with another business. Partial access is limited to tasks such as managing ads or replying to messages.

At exit, you remove the agency's users in Google Ads and unlink its manager account. In GA4, deleting a user needs the Administrator role at account level. On Meta you remove the partner from your portfolio. If the person who created a Google Ads script loses access to the account, the script stops running. If the agency set up scripts, list which ones run under whose authorization, using the steps in the Google Ads Scripts guide.

Write into the contract who owns these accounts, the audience lists and the creative source files. Add the access removal date and the delivery of off-platform files.

11. Write down report definitions and the notice period

What each report counts belongs in an annex to the contract. For each metric, list the definition, source, attribution window, period and the conversion actions included. Add that any change to a definition needs written approval. That keeps reports comparable from month to month. Put the scope table from check 8 in the same annex.

The notice period should leave enough time to hand over access and move campaigns to a new team. Its length is your call. What matters is that the contract says which work continues during notice and on which date the agency loses access.

The first 90 days are when you confirm the agency's setup, then make the first comparison against your baseline. In this period, checking the setup and the delivery of scoped work matters more than judging results.

12. Verify setup and delivery in month one

At the end of the first month, redo the list from check 2 and compare it with the contract. Check that the actions counted as conversions match the definitions in the annex. Confirm that auto-tagging is on in Google Ads and that links on other channels follow the agreed UTM rule.

Compare sales and new-customer figures in the agency's first report with your order system, and ask in writing why they differ. A gap is not always an error, because the platform and your order system count by different rules. What matters is that the cause is known and explained the same way each month. Also check this month whether the scoped work was delivered and who actually manages the account.

13. Judge ad effect and agency contribution separately at day 90

At day 90, recalculate the metric from check 1 with the same definition over a period of the same length. If your sales are seasonal, compare with the same months last year rather than the previous months. Also note events outside the agency's control, such as price changes, new products or stock problems.

Split this comparison into two questions. The first is the incremental effect of the ads. A test with a control group that sees no ads shows what the ad activity adds over no ads at all. If you plan one, work out the group size in advance with the method in the sample size guide.

The second is the agency's management contribution. A no-ads control group does not measure it, because it compares the agency with no ads, not with an in-house team or another agency. Judge the agency's contribution on three things. Look at the quality of the scoped work it delivered, and at the changes in the change history with the reasons it gave for them. Then look at how the metric moved against the baseline under comparable conditions.

A before-and-after comparison cannot fully separate the agency's effect from seasonality or price changes. That is why the three need reading together. To track customer lifetime value (LTV) over time, use the LTV to CAC ratio calculator. It compares what the agency's customers cost with what they are worth.

Red flags are the signs, across the checks above, of a risk that is hard to undo later. The table sums them up with the check each one comes from.