What to look for when hiring a digital marketing agency

A VP of marketing at a mid-market B2B software company opened three agency decks last Tuesday and found the same slide in all of them: channel mix, a sample dashboard, a promise of “full-funnel growth.” None of them named who would sit in the account, how negatives would get written, or what happens when Meta CPMs climb for four days before lead volume even twitches. That is the actual buying problem. You are not shopping for a logo. You are deciding who owns traffic quality, conversion, and the weekly judgment calls when Google Search CPC sits near $2.96 and 20-30% of SEM spend still leaks into junk queries.

This piece is for the person who owns the marketing number. It is not a catalog of services. It is how digital marketing agencies actually get evaluated, scoped, and run in 2026 when last-click is noisy, AI Overviews sit on more than 20% of searches, and creative dies in days instead of weeks.

What you should demand before a retainer starts, not after month three

Most pitches still sell platforms. You need operating rules. If the team cannot answer “who specifically will manage my account?” in the kickoff, you already know juniors will inherit the work. Ask for the named lead, the backup, and the cadence of insights. Monthly PDFs bury issues. Weekly notes that name wasted search terms, tired creatives, and landing mismatches do not.

Pricing is all over the map. Median monthly retainers sit around $3,000. Paid media management often lands $8K-$25K per month. Full-stack work can run $20K-$75K. PPC fees of 10-20% of spend are still common. SaaS companies under $1M ARR often sit at $3K-$8K; at $5M-$20M ARR, $15K-$30K is typical. None of that tells you if the scope includes negatives, landing tests, or conversion plumbing. A cheaper retainer that only “sends traffic to the homepage” is the expensive one.

Guaranteed ROAS or CPL before anyone has seen your account, pages, or unit economics is a red flag. Google Ads averages about 4.2x ROAS across industries in 2026 benchmarks. Meta sits nearer 2.8x, with verticals in a 2.79x-3.61x band. Performance Max in cleaned-up accounts can hit 4x-8x. Those are directional, not a contract clause. Honest shops will say realistic lift takes 3-6 months, and they will not promise a number on a cold call.

Account ownership matters. If the agency insists on holding the Google Ads login, you risk losing history when you leave. You should own the accounts. They should have access. Same for GA4, GTM, Meta, and Search Console. If they cannot work inside your properties, they are building a hostage situation, not a program.

If your current partner only talks features and never pipeline contribution from a channel, treat that as a scope problem, not a reporting preference. A paid search engagement that cannot show qualified lead cost next to wasted query share is not finished work.

Where agency programs actually leak: queries, creative, and the page after the click

You will hear “Performance Max drives the bulk now.” Fine. PMax still needs query hygiene, asset groups that match offers, and landing pages that continue the ad, not a homepage carousel. Twenty to thirty percent of SEM budgets still get burned on irrelevant queries, neglected negatives, or misaligned pages. Mobile is 63%+ of paid search clicks, yet conversion often lags desktop by 30-40%. If reporting is blended, you will never see that.

On Meta, creative is your targeting. Tight lookalikes and stacked audiences are a 2020 playbook. Broad plus a testing engine outperforms when the funnel is strong. Fatigue shows in CPM first, then in leads. UGC that looked great on day one can be dead by day five or six. If the shop only refreshes monthly, you are paying for decaying auctions. “Run broad” without a systematic creative pipeline is how budget finds low-quality traffic.

Signal quality is the other leak. Ad blockers and privacy changes still wipe 20%+ of conversion data. Smart bidding then optimizes on a broken sample. Enhanced Conversions, Conversions API, and first-party events are not IT chores. They are how you keep Event Match Quality and bidding from drifting. Meta’s in-platform CPA will disagree with your CRM. Last-click will double-count. Cross-channel deduplication between Google Ads and CM360 became less reliable after Google’s 2025 measurement changes restricted third-party signal sharing. You need a model you trust enough to reallocate, not a screenshot war.

Organic is not immune. 68.01% of US Google searches ended without a click in early 2026. AI Overviews reach about 2 billion monthly users and can cut CTR by nearly 60% on affected queries. Industry estimates suggest the majority of top-ranking pages now contain some AI-assisted text - the differentiator is the part a model cannot fake: your proof, your numbers, your point of view. 55.9% of CrUX origins pass all three Core Web Vitals as of May 2026. If your pages fail INP, paid and organic both pay for it.

Watch session recordings when paid traffic “does not convert.” Form abandon, proof buried below the fold, and CTA mismatch are usually the story. HeyLead Insights is built for that post-click evidence: where people stall, which blocks they skip, which fields they never finish. Guessing at copy is slower than watching the sessions.

A working scorecard for digital marketing agencies you already pay

Use this as an internal audit, whether you are hiring, renewing, or rebuilding in-house plus a specialist. Score each item honestly. Anything you cannot evidence in the last 30 days is a gap, not a maybe.

How a full program actually runs when channels, costs, and AI results keep moving

Think in loops, not launches. Paid search: intent clusters, AI Max where it truly cuts CPC 12-18% versus standard Search without wrecking lead quality, negatives as a living list, and PMax only after conversion events are trustworthy. Do not yank structure every week. Learning resets are expensive. High CPC on a new account may force a short stretch of manual bidding until you have enough clean conversions. That is judgment, not a religion about automation.

Paid social: fewer campaigns, Advantage+ or broad, and a creative engine. Stop trying to out-segment Meta’s delivery. Protect user experience in the auction: hooks that interrupt the doom scroll without bait. When CPMs rise and CTR holds, you are usually early in fatigue. Replace the ad. Do not wait for CPL to confirm what the auction already told you. Diagnostics like EMQ help with plumbing. They do not predict profit by themselves.

SEO plus AEO plus GEO: you need pages that earn citations, not another 800-word roundup. Human perspective, original data, and internal links beat generic AI. Rankings screenshots can look fine while qualified pipeline dies. Search Console’s generative reports matter more than a vanity keyword list. Industry estimates suggest the majority of top-ranking pages now contain some AI-assisted text - the differentiator is the part a model cannot fake: your proof, your numbers, your point of view.

After the click, CRO is not a side project. One operator moved landing conversion from 33.22% to 43.52% by putting social proof next to the form, not by adding another ad platform. Another saw a 70% lift in PPC conversion after rebuilding the page to match the query. Traffic to a slow, vague homepage will never show “agency magic.”

Two accounts, two mechanisms, same lesson

A Head of Growth at a $12M ARR security SaaS inherited a Google Ads account with 41 campaigns and a 19% wasted-query share. Automated bidding looked “on,” but conversions included newsletter signups. They collapsed to six campaigns, qualified demo as the only conversion, and rebuilt two landing pages so the headline matched the exact commercial cluster. CPC did not magically fall. Accepted demo rate did. That was the mechanism: signal quality, not another campaign rename.

A founder running a multi-location services brand watched Meta CPL look stable while sales complained about tire-kickers. In-platform CPA looked fine; sales accepted 37% of those leads. CPMs had been climbing for 11 days. The same three UGC cuts had been in market for three weeks. They did not add TikTok “because everyone should.” They cut the dead ads, wrote new hooks that named the emergency job (same-day dispatch, not “learn more”), and sent traffic to a page with a click-to-call and a short proof block. Lead volume dipped for four days, then paid jobs recovered. Creative was the targeting. The funnel made broad delivery safe.

Neither story required a new platform. Tighter positioning, clearer offers, and decent attribution beat stacking channels. Clients who churn after one reel that did not “rocket sales” were never scoped for a real budget. Digital marketing agencies that tell you that up front are doing you a favor.

FAQs

How much should digital marketing agencies cost for a company our size?

Expect a wide band. Median retainers near $3,000 hide a lot of thin scope. Serious paid plus landing plus measurement often sits in the $8K-$25K range for media management, higher for full-stack. Compare what is owned: negatives, creative production, pages, tracking, and weekly insight, not the slide count.

Should we trust an agency that guarantees ROAS or CPL upfront?

No. Anyone who does has not looked at your pages, conversion quality, or unit economics. Use industry bands (Google ~4.2x, Meta ~2.8x) as context, then run 3-6 months of clean data in your account.

Should we keep Google Ads in-house and only hire for Meta, or the reverse?

Split ownership usually splits the landing page and the conversion definition. If you split channels, keep one owner for tracking, offers, and the page after the click. Channel specialists without a shared funnel will argue about attribution forever.

Is automated bidding just inflating budget?

It will if events are dirty or volume is tiny. It can cut CPC (AI Max has shown 12-18% lower CPCs versus standard Search in some accounts) when conversions mean revenue. New accounts with weak data may need a manual stretch first. That is normal.

Do we still need keywords if everyone says “signals over keywords”?

Yes. Google still needs intent clusters, negatives, and landing match. Signals (first-party data, Enhanced Conversions) protect bidding. They do not replace knowing which queries you refuse to buy.

Putting it to work

This week

• Pull 30 days of search terms and mark wasted spend in one spreadsheet column.

• List every live paid URL. Flag homepage dumps.

• Compare Meta CPM trend to CPL for the last 14 days. Note ads older than five days.

• Verify one qualified conversion in GA4, the ad platforms, and the CRM. Write the mismatch.

• Run Core Web Vitals and H1 checks on the two highest-spend landers.

• Ask your current partner (or internal lead) who changes the account this week, by name.

Next 30 days

• Cut campaigns that reset learning without a reason. Prefer fewer, broader structures.

• Ship a dedicated landing page for the top commercial cluster.

• Stand up a creative test cadence so Meta is not running on three aging cuts.

• Rewrite or prune organic pages that rank without information gain.

• Report cost per accepted lead, not cost per form, in the next leadership pack.

If the work that is stalling you is the loop between query waste, dying creative, and landers that do not match the ad, HeyLead can own that operating loop so you are not reconciling three dashboards every Friday. Start with the free tools above, or grab a free audit if you want a second set of eyes. When you want the recurring execution off your plate, Chat with us on WhatsApp.