Ask a room of founders which marketing agency they would hire again, then listen to why. Almost nobody answers with a skill. Nobody says the keyword research was exceptional. They say: she actually understood how we made money. They told us not to do the rebrand. The report told me what failed. I could get hold of someone on a Tuesday.

That pattern holds across hundreds of these answers, and it is the most useful thing in them. A good marketing agency is one that understands how the client makes money, declines work that will not move that number, reports in terms a CFO recognises, and is honest about when the retainer should end. Channel craft is table stakes: necessary, broadly available, and almost never the thing a happy client names first.

One thing to sit with before the detail. Every behaviour clients praise costs the agency money in the short run. Saying no shrinks the scope. Honest reporting surfaces failure. Telling a client to move the work in-house ends the invoice. Not soft skills, then, but expensive choices, which is why they are rare and why they are worth screening for.

They ask about the business model before they ask about the channel

The agencies clients keep are the ones whose first call was uncomfortable because of the questions. Not “what’s your budget and which channels are you running,” but closer to this:

• What does a customer pay you in year one, and what do you keep after cost of delivery?
• How long from first touch to closed, and who is involved in saying yes?
• What percentage of trials or enquiries convert, split by where they came from?
• What does a customer look like at month twelve: renewed, expanded, gone?
• Who closes the demand we create, and how much capacity do they have?

Everything downstream depends on those numbers. Gross margin sets the ceiling on acquisition cost; without it nobody can say whether a $400 cost per customer is excellent or ruinous. Sales cycle sets the reporting window: on a five-month cycle a monthly revenue report is noise, and an agency that does not know this will either panic in month three or quietly start optimising a lead count, because lead counts move in weeks.

Watch how differently the same money gets spent once the answers exist. Product A: $1,200 a year, self-serve, most churn inside 45 days. Product B: $40,000 a year, seven-month cycle, a committee of four, two salespeople. Give both $10,000 a month. Product A needs volume at a low cost per activated account, and most of its leverage sits in onboarding rather than advertising. Product B needs thirty named accounts and material a champion can forward to a finance director. An agency pitching both the same “full-funnel demand engine” never asked.

This gap is normal on the inside too. In the Spring 2026 CMO Survey, run by Duke’s Fuqua School with 308 senior US marketing leaders responding between 7 and 29 January 2026, only 49.5 percent said marketing and finance work together on growth. An agency opening with the finance questions is doing something most in-house teams are not yet doing routinely. Google’s guidance on hiring a search provider puts it plainly: your provider “should ask questions about what makes your business unique, who your competitors are, how search results can help, and how your customers find you.” If a proposal arrives without those numbers in it, the agency is guessing and has priced the guess.

Saying no is the single most cited behaviour

Go back through the praise and this appears more than anything else. Clients remember the no. It shows up in three forms, and they are not equally hard.

Declining work that will not move the number. A client asks for a podcast, a rebrand, a presence on a new platform. The agency prices it, then says the honest thing: this takes four months, it will not change your pipeline this year, here is what we would do instead. Turning down billable work already on the table is a real cost, which is why it is a real signal.

Telling a client the positioning is the problem, not the ads. This is the one clients quote years later, and most have never had the mechanism explained. Paid media amplifies a message; it does not create one. If your message does not separate you from three alternatives, extra spend buys more of the small group who were going to be persuaded anyway, and as you exhaust them your cost per acquisition climbs while the account merely looks badly managed. The agency saying “this is not an ads problem, it is that nobody can tell what you do” is refusing the easy diagnosis.

Refusing a channel the business is not ready for. Content and search when nobody is there to work the enquiries. Paid social for a $40,000 product with a committee and a seven-month cycle. Any paid programme at all when trial-to-paid is running at 1.5 percent, because you would be buying more trials for an onboarding that does not work. Saying yes takes ten minutes. Saying no takes an hour of explaining and might lose the account.

One counterfeit to watch: “we only do paid search” is positioning, not a no, because that work was never on the table. Ask for an occasion when money was offered and refused. An agency that has never turned down revenue will not start with you.

Reporting a founder can act on

The praised reports share four properties, and none is design quality.

They lead with pipeline and revenue. Qualified opportunities by source, cost per qualified opportunity, pipeline created and revenue closed against spend. Impressions, reach and follower growth sit near the bottom if at all, as diagnostics for why a number moved rather than the number itself.

They contain failures. This is what separates a report from a sales asset: a report with nothing that went wrong in it is measuring nothing, because a large share of everything tried in marketing does not work. What good looks like: we built the comparison pages, impressions came and demo requests did not, we read the intent as research rather than purchase, so we are cutting them and moving the budget to the integration pages.

They fit on one page and read in two minutes with nobody explaining them. If you need the agency on a call to understand the report, it is a presentation.

They match their window to the sales cycle: leading indicators monthly, revenue by cohort quarterly.

The same CMO Survey rated marketers’ own performance at “demonstrating ROI from marketing technologies” at 4.4 on a seven-point scale, and 86.3 percent named stronger performance tracking as a way to show marketing’s value. Read that from the agency’s side. Your buyer is already being asked upstairs to prove impact, so an agency whose reporting does that job is solving a political problem rather than filing an admin document. That is why those agencies get renewed.

They own what happens after the click

Most agencies stop at the click. Most problems live after it. That mismatch produces a large share of failed retainers, and both sides usually misdiagnose it as a traffic problem for months. The clearest evidence is from e-commerce, where the funnel is measurable end to end. Baymard Institute’s aggregate of 50 studies puts the documented average cart abandonment rate at 70.22 percent, and among reasons given for abandoning during checkout, extra costs appearing late accounts for 40 percent, forced account creation 18 percent, a checkout that is too long or complicated 17 percent. Almost none of that is a media problem. Almost all of it is fixable in design.

That is retail, not software, so read it structurally rather than literally. The B2B and SaaS equivalents are familiar: no pricing page, so the visitor leaves to ask a competitor; a demo form wanting company size and phone number before showing any value; a trial that drops a new user into an empty workspace; a demo request sitting unrouted overnight. Each wrecks paid media performance, and none is fixable by better bidding.

So the first-conversation test is direct: what would you want to change that isn’t advertising? An agency answering with the signup flow, the pricing page, the first onboarding email and the speed of lead routing intends to be accountable for revenue. One answering only with channels and creative has told you politely that it supplies traffic and the conversion problem stays yours. That is a legitimate thing to buy. Do not pay for it on a revenue promise, because an agency with no opinion about your onboarding holds no lever on most of what determines that revenue.

The operating rhythm that makes clients feel in control

Notice how much of the praise is procedural rather than creative. Someone was reachable. Nothing was a surprise. I always knew what was happening. Buyers are describing control, and control comes from predictability, not volume of contact. Four things produce it.

A named person who does the work, not only an account manager who relays it. The strong version is the practitioner writing your pages or running your ad account on the call, answering a hard question without checking first.

A predictable cadence, weekly rather than monthly. Same day, same 25 minutes, standing agenda: what shipped, what is blocked, what needs a decision from you and by when. A short weekly beats a long monthly, because a month is long enough for one wrong assumption to become three weeks of wasted work.

Decisions written down where both sides can read them. A running log: date, decision, who made it, what it means. Bureaucratic until a new person joins on either side, or until you and the agency remember a scope conversation differently. Google’s hiring guidance suggests asking a provider whether they “will share with me all the changes you make to my site” and explain the reasoning behind their recommendations. Reasonable to want in writing from any agency, not only a search one.

No surprises at invoice time. Scope changes priced before the work, not appearing as a line item after it. Cheap to do, startlingly rare, and one unexplained invoice undoes six good months.

They know when to fire themselves

The agencies with the strongest reputations are disproportionately the ones clients remember for ending the arrangement. Not a paradox. Reputation here travels by referral, and a founder told “you do not need us any more, here is the handover” repeats that story to everyone. A founder who paid for eighteen months of maintenance tells a different story, quietly, in the same rooms.

The signals are recognisable from either side:

• The work has become maintenance that a competent internal generalist could hold.
• The binding constraint has moved somewhere the agency cannot reach: the product, the pricing, the sales team’s capacity.
• The next $5,000 of retainer now returns less than the same money spent on a salesperson, a developer or ad budget.

Done properly it looks like a written handover, every account in the client’s own name, a playbook someone junior can follow, and an offer of something small instead of something large: a quarterly review, a monthly hour, a project when one is genuinely needed. The CMO Survey found companies outsource around a third of their digital marketing activities, so where the in-house line sits is a live question in most companies. An agency willing to help redraw it is worth more than one defending its own side.

Which produces the question I would rank above anything else in a first call: under what circumstances would you tell us to stop paying you? A specific, unhesitant answer is the strongest signal available. A non-answer is also information.

Half of this behaviour is created by the client

The part buyers enjoy less: these behaviours are partly produced by the client, and the same agency performs differently across two accounts. Four things move it.

Give honest numbers, including the embarrassing ones. Real margin, churn, close rate by source, sales capacity. Hide a poor close rate and the agency does the rational thing with what it has, optimises for volume, and then you both spend six months arguing about lead quality. If you cannot share a number, say so, so they know they are working blind.

Decide promptly and name one decider. One person with authority, a five-working-day approval window, and a rule that work not approved inside it ships as drafted. Approval queues are the commonest cause of an engagement that produced nothing, and they are invisible in the agency’s report.

Give real feedback. “I don’t like it” is a preference. “This doesn’t match how our buyers describe the problem, here is a recording of one describing it” is feedback, and it upgrades the next six deliverables. Nearly free and rarely offered: 45 minutes with your two best salespeople, and three recorded customer calls.

Pay on time. Unglamorous and true. Agencies staff their best people onto accounts that do not require chasing, and nobody writes that in a proposal.

The first-call questions that reveal how an agency operates

• “What do you need to know about our economics before you’d put a number on this?” Listen for gross margin, contract value, sales cycle, close rate by source. A pitch deck in reply is the answer.
• “Tell me about a client you said no to, and what happened next.” A real occasion with a cost attached, not a positioning statement.
• “Who does the work, and can they be on our next call?”
• “Show me a report you sent a client last month, numbers redacted.” Checking two things: is there a failure section, and could a founder read it in two minutes.
• “What would you want to change that isn’t advertising?”
• “Under what circumstances would you tell us to stop paying you?”
• “What is the smallest engagement you would take, and would you recommend it to us?”

Notice what is absent. Not one question is about channel expertise, certifications, tool stack or case studies, which mirrors what clients who rave about an agency actually talk about. Channel competence is the entry requirement, and a paid two-week project verifies it better than any pitch. What you cannot verify cheaply is whether these particular people will tell you the truth about your positioning, put a failure in writing, and hand the work back when it stops being worth your money. Interview for that. If you are building an agency, build it.

If this has convinced you that what you need is the conversion and measurement half rather than another traffic retainer, that is work AB7 Solutions does. We run SEO, AEO and GEO programmes where the reporting starts with pipeline rather than impressions; we build the websites, applications and APIs plus the CRM and automation plumbing that decide what happens after the click, which is where most marketing spend is genuinely lost; and our recruitment and contract staffing side can resource an in-house specialist instead when that is cheaper. We will tell you which of the three you need, including the fairly common case where you need no retainer at all and should put the money into sales capacity or onboarding. Call +1 321 341 7733, email ab@ab7solutions.com or director@ab7solutions.com, or start at www.ab7solutions.com.

Sources: The CMO Survey, Spring 2026 Highlights and Insights Report, Duke University Fuqua School of Business (cmosurvey.org): 308 responses from 2,111 invited US marketing leaders, fielded 7-29 January 2026, 97% VP level or above, self-reported. Baymard Institute, cart abandonment rate: 70.22% documented average across 50 aggregated studies, plus abandonment reasons from Baymard’s own survey; e-commerce data, used structurally rather than as a B2B benchmark. Google Search Central, “Do you need an SEO?”. No published statistic exists for agency retention by behaviour, so those sections describe mechanisms and are opinion.