If you diagnose “sales” but fire the agency, you will spend the next two quarters buying the same leads for a different vendor.

Step 2: Agency-side red flags that justify leaving

• No named strategist. You’re told “our team” handles your account. At $110/hour, “our team” is often three hours of a junior’s week.
• No search-term reporting, ever. Activity is reported. Waste isn’t.
• Ownership friction. The billing profile isn’t in your name, admin access was never granted, or your account lives under theirs.
• “Wasted spend we cut” quoted in percentage, not dollars. A percentage of an unknown denominator is a sentence, not a metric.
• Lock-in with exit penalties. Especially “you can’t take the historical data.”
• Performance reported to you in the same period the account is changing. Without lags acknowledged, results are noise. In B2B, the click-to-conversion lag can run 7–21 days; in energy and industrial longer. A shop that reports daily ROAS on a 21-day cycle is telling you what they want you to hear.

Step 3: Client-side realities — the firing-side that never gets written

• You want more leads but won’t raise the budget. At $8–$12 CPCs, more leads at flat spend means lower-intent leads. That’s arithmetic, not effort.
• You won’t fix the landing page. Message match is not a nicety; it is the conversion mechanism. Bad destination = expensive clicks wasted, regardless of the copy.
• Your sales team doesn’t follow up, and you have no closed-won data. Smart Bidding then learns from form-fills instead of revenue — and scales the wrong thing. The “quality dropped” complaint is usually a follow-up problem.
• Your approval chain is a week long in a channel with hurricane-weekend demand. With a Category 4 in the Gulf, an approval chain is a competitive disadvantage. Seasonality-ready assets need delegation.

Step 4: The no-fault failure

Sometimes the offer is the problem, not the ads. When your offer is what your clients have already rejected, no bidding strategy reaches the auction. A competing offer you didn’t know about can starve conversion.

Step 5: The exit structure that keeps you in control

There’s a wrong way to remain in a bad relationship and a wrong way to leave. A preferable exit is month-to-month or 30-day notice, with the account, the data, and the CRM being yours all along.

One final note: an agency that tells you a lead is not the right fit for a limited budget (below the $3,000–$5,000 per-month floor we discussed) or for a reconciliation scope, and that makes no attempt to sign you anyway, is the best signal of trustworthiness you will get. Take the next meeting.

How We Run Google Ads in 2026: The Operating System

This is where the methodology gets concrete. Here’s the operating system behind our PPC management services, and the same logic applies whether you hire us or someone else — hold any agency to it.

Step 1: Rebuild around intent clusters, not keyword lists

Campaigns map to buying stages and sub-markets: brand defense, high-intent local (“emergency plumber katy”), B2B capability terms, and competitor capture — each with its own budget, bidding, and landing destination. Brand and non-brand never share a campaign, so the truth stays visible.

Step 2: Let Smart Bidding do what it’s good at — with clean fuel

Per Google’s March 2026 guidance, you no longer need to “warm up” an account with manual CPC before switching to automated bidding; the system trains across account-level conversion data from the first impression. What it does need is fuel: enough conversion volume on actions that genuinely indicate value. Their product managers describe finding the “conversion sweet spot” — the lowest action in the journey with enough frequency to be a high-quality signal. For a long-cycle B2B sale, that might be a qualified consultation, not a signed contract.

From there, the 2026 toolkit for Houston accounts:

• tROAS over tCPA wherever margins vary. If your leads differ in value — and in Houston they do, wildly — pass conversion value through enhanced conversions or offline import and optimize toward return, not flat cost.
• Portfolio bidding to pool conversion data across related campaigns, which stabilizes learning for smaller accounts.
• Smart Bidding Exploration (currently for Search with tROAS) — Google reports an average 10% conversion lift versus simply lowering your ROAS target, because it protects core traffic while exploring new query territory. Worth enabling once your base is stable; watch the new “traffic diversity” metric to confirm it’s working.
• Journey Aware Bidding, previewed at Think Week in September 2025 and rolling out through 2026, feeds intermediate funnel events into the bid model — earlier signal for the long sales cycles that define Houston’s energy and industrial sectors.
• “Limited by target” as a signal, not an error. When the interface flags it, your target is too restrictive and you’re leaving auctions on the table. Relax the tCPA or lower the tROAS — and wait a full conversion cycle before adjusting again.
• Conversion delay accounting. In B2B, lag between click and attributed conversion can run 7–21 days. Evaluate campaigns on projected conversions, not raw daily numbers, or you’ll kill winners during their learning window.

Step 3: Engineer the geographic layer

ZIP-level targeting and exclusions, sub-market bid adjustments, “in or regularly in” only, Local Services Ads separated from standard Search where eligible, and ad copy that names the territory. This is the layer national agencies skip and Houston loses on.

Step 4: Build the landing experience, not just the ad

Message match between ad and page, mobile-first load speed, one primary CTA, trust signals appropriate to the vertical (licensing, credentials, local presence). A great ad pointed at a weak page is a donation to your competitor.

Step 5: Optimize on a cadence, report on economics

Weekly search-term and budget review; biweekly creative tests; monthly reporting that leads with wasted-spend dollars recovered, qualified leads, cost per acquisition, and ROAS — with the deltas explained. If a month’s report doesn’t tell you what changed and why, request a better one.

The Para-Google Ecosystem: LSA, Performance Max, and the Cannibalization Wars Nobody Warns You About

Your account doesn’t exist in isolation. Local Services Ads may sit above your Search ads. Performance Max may bid on the same term. Choice (Demand Gen) may sprinkle a conversion into the funnel right before your Search click. By the end, three different parts of Google’s system can claim credit for the person who filled out one form. Which is a problem: it means you can literally bill two budgets for one lead.

Local Services Ads: pay-per-lead is not pay-per-click

LSA operates on a different billing model — you pay per lead, not per click — and it requires passing Google’s screening and verification process, which varies by category and can include license and insurance checks and a background check. That verification unlocks badges like Google Guaranteed or Google Screened. The practical effect: for many home-services queries, LSA can deliver the lead at a lower effective cost than a Search click, because you’re not paying for the people who browse and bounce.

What to watch for: LSA doesn’t disqualify aggressively on geography the way a well-architected Search account does, so a Houston-only business can pay for leads outside its service area. Vet the service-area settings the same way you’d vet ZIP exclusions in Search.

The LSA-versus-Search overlap

LSA frequently captures the click on the core local-service query; Search then pays for the residual. If you report both channels’ conversions at face value, you’ve counted the same lead twice and both channels look better than they are. The honest test is incrementality, not attribution: pause LSA in a matched set of service areas for two weeks and measure how much volume Search actually absorbs. Whatever Search absorbs tells you LSA’s true contribution — not the number in the LSA dashboard.

Performance Max: the “black box” is now only mostly dark

PMax has a bad reputation, some of it earned. Historically it hid where the money went. That’s changed: Google now gives advertisers a search terms report for Performance Max, and you can apply brand exclusions to keep PMax from bidding on your own name. Both are real improvements.

But visibility is not control. PMax still bids across the same auctions as Search; PMax can still consume generic and branded traffic; PMax still reports conversions at face value. The honest assessment is that PMax helps when you have a feed (ecommerce, retail, strong product catalogs) — it’s a data-hungry format — and hurts when you’re a B2B lead-gen business with a long sales cycle and one qualified conversion action, because it will find the cheapest “conversion” available and declare victory.

Demand Gen and the double-counting problem

Demand Gen (and Discovery before it) is top-of-funnel. It generates awareness touchpoints. But it’s also built to report conversions, and if a user sees a Demand Gen ad on Monday and converts via Search on Thursday, both get a star. The result: your awareness channel looks like a performance channel, your Search channel looks twice as good as it is, and your blended ROAS inflates in a way that doesn’t match your bank account.

The account-level reality check

Here’s the simplest diagnostic available, and almost nobody runs it: compare total platform-reported conversions to total closed deals (or total qualified leads) for the same period. If platform-reporting says you generated 200 leads and your CRM shows 120, you don’t have 200 leads — you have a double-counting problem and an inflated ROAS target.

Cannibalization diagnostic

• Pull the PMax search terms report. Are branded queries present? If so, add brand exclusions and re-measure.
• Compare LSA leads and Search leads by ZIP. Overlap means double-counting.
• Check Demand Gen conversions. Do any also appear as Search conversions in the same session window? Label them “assisted” internally.
• Set one channel as your source of truth for each buying stage. BOFU = Search. MOFU = PMax/Demand Gen. Everything else is assist-only.
• Run the totals-versus-CRM reconciliation every month. No exceptions.

The insight: you are not running a set of campaigns. You are arbitraging overlapping auctions inside one machine. Most accounts are quietly paying twice for the same lead — and the reporting will never, ever show it, because each product is doing exactly what it was built to do.

The Dark Funnel: Measuring the Demand That Never Lands in Your Attribution Report

Every dashboard you own has a blind spot, and in 2026 that blind spot is structural, not fixable with a better tool. A large share of the way Houston buyers actually find you never generates an attributable click: the referrer was stripped. A spouse mentioned you to a colleague at a Memorial-area dinner. A client got your name from a friend who found you on a map pin and never visited your site. Someone clicked, converted, then called from a personal cell the system can never match.

This is the dark funnel. It’s not a problem to solve — it’s a condition to measure around. The agencies that pretend it doesn’t exist report flattering numbers. The ones that acknowledge it make better budget decisions.

What’s actually in the dark funnel

• Dark social. Slack, WhatsApp, group texts, Teams messages. The link gets shared; the click gets attributed to “direct/none.” B2B referrals in Houston’s energy and industrial ecosystem move this way constantly.
• Unattributed direct traffic. People who typed your URL from memory or navigated to your brand after an unlinked exposure.
• Google Business Profile and map interactions. Saved profile, call from the profile, route request — the profile gets traffic that never lands in Ads.
• Offline word of mouth and referral. The highest-value lead source in most professional services, and the least attributable.

The honest feedback loop: self-reported attribution

Add “How did you hear about us?” to every lead form and every qualification call. It is crude, it is biased, and it is still the single most useful reality check you have. Read it correctly:

• Social-desirability bias: people say “Google” because it sounds more credible than “a friend told me.”
• Recall lag: attribution by memory decays fast for anything older than a few weeks.
• Small-sample noise: in a low-volume B2B account, fifty responses are directional, not statistical.

Use it as a reality check on your dashboard, not as a replacement for one. When self-reported numbers and platform numbers agree on trajectory, you can relax. When they disagree wildly, you have a measurement problem that matters more than any bid adjustment.

Call-tracking gaps nobody audits

Phone-first verticals — legal, home services, healthcare — live and die here:

• Dynamic number insertion breakage. DNI fails on some page types, some mobile browsers, some third-party scripts. The call happens; the attribution doesn’t.
• Google forwarding numbers masking channel. When Google’s forwarding number is used, the real originating source can be obscured. Cross-check GFB reports against your call platform.
• Missed calls counted as leads. A missed call is not a qualified lead, but many attribution stacks treat it as a conversion. That trains Smart Bidding on ghosts.
• Recording and QA. If nobody listens to the calls, you’re optimizing toward “someone dialed a number,” not “someone had a service need.” Fifteen minutes of weekly call QA beats another dashboard.

Offline conversion import: the bridge from click to revenue

This is the mechanism that fixes the biggest B2B blind spot. By importing offline conversions, you feed what happens after the click back into Google — so Smart Bidding learns the outcome, not just the form-fill. In practice:

• Pass the value at the stage that reflects real business economics. A qualified consultation for a commercial HVAC contractor, a retained engagement for a Houston law firm, a purchase order for an upstream supplier.
• Match on a stable identifier (enhanced conversions for hashed first-party data, or GCLID-based offline import for CRM-driven events).
• Accept the “match” limit: you’ll never reconcile 100%. Seventy to ninety percent match is a working pipeline.

Consent mode and the reporting floor

When consent signals are absent, Google models conversions. That modeling shifts your reported CPA away from your real CPA — sometimes in your favor, sometimes not. This isn’t a reason to distrust everything; it’s a reason to triangulate. Treat modeled conversions as an estimate, then confirm against three imperfect sources: CRM closed-won, call logs, and self-reported attribution. When all three point the same direction, act. When they disagree, investigate before you cut budget.

The insight: budgets don’t get cut because campaigns fail. They get cut because the measurement failed first, and nobody noticed until the CFO asked why leads looked great and revenue didn’t. Restraint beats a dashboard.

The Brand-Bidding Paradox: The One “Obvious” Defensive Move That’s Usually a Tax

“Always bid on your own brand.” You’ve heard it from every agency, including plenty that profit from managing that spend. It’s stated as a law of nature. It isn’t. It’s a conditional decision, and in a meaningful share of Houston accounts, brand bidding is a self-funded tax — you paying Google for clicks you would have received for free.

That doesn’t mean never bid on brand. It means the decision depends on variables almost nobody audits.

When brand bidding genuinely earns its keep

• A competitor is actively bidding on your name. If they’d otherwise take the top slot on your own brand query, defensive bidding is a real cost of doing business.
• Your organic presence is weak or absent. New firm, new location, a brand name nobody recognizes — paid brand gets you the click your organic listing can’t.
• You’re entering a new sub-market. A Katy dental practice with zero organic footprint in Sugar Land should buy brand there while organic builds.
• Brand queries with commercial modifiers (“[your firm] houston commercial litigation”) — those aren’t pure brand; they’re brand plus intent. Bid those.

When it’s a tax

• You rank #1 organically and no competitor bids. You are paying for a click that was already yours. This is the most common brand-spend waste in Houston SMB accounts.
• Your brand term is your only strong performer. That’s a warning sign, not a win. It means non-brand — the part that actually grows the business — is struggling, and brand is hiding it.

The measurement artifact that hides a sinking account

Here’s the trap. Brand terms convert at high rates and low CPCs — because the person already knows you. Blend them with non-brand and your account-level ROAS looks excellent. Your agency reports a heroic number. Meanwhile the non-brand campaign that generates new demand is quietly failing, and because brand is covering for it, nobody notices until brand saturates and growth flatlines.

This is why brand and non-brand must never share a campaign. It’s not organizational neatness. It’s the only way to see whether you’re actually building a business or just harvesting people who already found you.

The trademark nuance

You can often bid on a competitor’s trademark as a keyword, but you cannot use it in your ad copy — Google restricts trademark use in ads from a direct competitor. The reverse trap is worse: pouring legal threats at a competitor who bids on your name usually costs more than the auction itself, and rarely produces the outcome you want.

The incrementality test: how to actually price the defense

The only honest way to know what brand bidding is worth is to measure what organic absorbs when you stop paying. Run a two-week geo holdout or a matched-market pause: turn brand ads off in a defined set of ZIPs, keep them on in a comparable set, and compare total conversions (organic + paid) between them.

• If organic absorbs most of it, brand bidding was largely a tax.
• If conversions drop meaningfully, brand bidding was doing defensive work worth paying for.
• If you can’t run the test, don’t guess — at minimum, check organic rank and competitor presence before extending the spend another year.

Decision tree

1. Rank #1 organic on the brand term? No → bid defensively. Yes → continue.
2. A competitor bidding on your name? Yes → bid defensively. No → continue.
3. Brand recognition low or new sub-market? Yes → bid while organic builds. No → continue.
4. Blended ROAS masks weak non-brand? Yes → separate brand-out, fix non-brand, and consider holding or reducing brand spend. No → run the geo holdout and let the data decide.

The insight: brand campaigns are primarily a measurement instrument — the clean control group that tells you how non-brand really performs — not a performance driver you maximize. Manage the brand line to see the truth, not to claim credit.