What automation changed about the job
For a decade the Google Ads job was largely manual: set keyword bids, adjust them by device and time, prune by hand. Smart Bidding, broad match, and Performance Max have taken most of that work. The platform now sets bids per auction, finds queries you never added, and assembles ads across its whole inventory.
That does not make the account manager redundant. It moves the leverage. Your job in 2026 is to give the automation the three things it cannot generate for itself: accurate conversion signals to learn from, a structure that reflects real intent, and guardrails that stop it wandering into waste. Get those right and automation compounds them. Get them wrong and it scales the error faster than any human ever could.
You no longer beat the auction by bidding better. You beat it by feeding the machine better signals and fencing it in.
Step 1: verify conversion tracking before anything else
Nothing else in this playbook matters if the conversion data is wrong, because Smart Bidding optimises toward whatever it is told to value. The first work on any account is to verify tracking, not to touch campaigns.
That means confirming conversion actions fire once and only once, that purchase values match the backend or Shopify, that duplicate and junk conversions are removed, and that Enhanced Conversions and, where relevant, server-side tracking are recovering the events browsers now block. An account bidding confidently on broken tracking loses money with precision. iClick will not scale spend or enable an aggressive bid strategy until the conversion numbers reconcile with the source of truth.
Step 2: structure the account around intent
Automation still needs a skeleton, and the skeleton is intent. Group campaigns and ad groups so that everything inside one shares a searcher's goal, because that is what lets the bidding model, the ad copy, and the landing page all point the same direction.
The practical rules: keep brand traffic in its own campaign so you can see and control it separately, separate high-intent bottom-funnel terms from broader research queries, and split by margin tier or product priority where budgets need to move independently. Tight, intent-coherent structure is what makes Quality Score climb and what gives Smart Bidding a clean signal. Sprawling ad groups that mix intents confuse everything downstream of them.
The campaign types and when to use them
Search remains the backbone for capturing intent, run with a disciplined mix of exact and phrase match and, where the account has good conversion data, fenced broad match to find new queries.
Performance Max covers Shopping and cross-inventory reach for eCommerce and many lead-gen accounts, but only with guardrails, covered below. Shopping, where run as standalone, lives or dies on the product feed. Demand Gen handles upper-funnel and visual prospecting. Brand campaigns protect your name cheaply and should almost always run, with the caveat that PMax must be prevented from claiming that same branded traffic. The art is not picking one type; it is coordinating them so they capture different demand instead of bidding against each other.
Step 3: set bidding from margin, and ramp it
Bid strategy follows the economics, not the other way around. Derive the target from gross margin: break-even ROAS is 100 divided by your margin percent, then add a contribution cushion for overhead. That gives the number Smart Bidding should aim for.
The execution rule is to never hand the algorithm the ambitious target cold. Anchor the first target ROAS or target CPA to what the campaign already proves, then tighten toward the margin-derived goal in 10 to 15 percent steps, one conversion cycle apart. Use value-based strategies where order values vary and cost-based strategies where conversions are roughly equal. The margin math sets the destination; the ramp is how you arrive without stalling volume.
Step 4: run Performance Max with the fences on
Performance Max is powerful and, left on defaults, quietly expensive. Its opacity lets it cannibalise brand search, lean on Shopping while starving other formats, and report a flattering number while absorbing demand you already owned.
iClick runs it inside a fence: brand exclusions applied from day one so it cannot harvest cheap branded clicks and call them new, a clean and well-structured product feed since the feed is most of its targeting, several themed asset groups rather than one catch-all, and scripts plus placement and search-term insights to see where conversions actually come from. Performance Max earns its budget with these controls. Without them it is a convenient way to overpay for demand you had already captured.
Step 5: search-term and negative-keyword hygiene
Broad match and Performance Max are only as safe as the negatives fencing them. The search terms report is where an automated account tells you what it is actually doing, and reading it is not optional.
On active accounts iClick harvests negatives from the search terms report weekly, excluding off-intent queries before they train the bidding model on the wrong audience, and occasionally promoting a strong performer into its own keyword. Negatives are layered at the right level so a campaign-wide block does not kill one ad group's best term, and over-broad negatives are audited for the valuable searches they might silently exclude. In an automated account the negative list is the steering wheel, not a set-and-forget setting.
The weekly cadence that keeps it honest
An automated account does not run itself; it drifts if unwatched. The weekly cadence is what catches drift early.
Each week: read the search terms report and update negatives, check that Smart Bidding is in target and not stuck in a learning period, reconcile conversions against the backend, review PMax placements and asset performance, and watch the leading indicators, Quality Score movement, impression share lost to budget or rank, and any widening gap between channel ROAS and blended MER. The point is to change one thing deliberately and measure it, not to fiddle daily. Discipline beats activity: an account reviewed properly once a week outperforms one poked at every day.
Change one lever per review, give it a full conversion cycle, then read the result. Daily poking resets learning and hides cause from effect.
What not to do in 2026
1. Do not enable aggressive Smart Bidding on unverified tracking. The model will scale whatever error is in the data.
2. Do not run Performance Max without brand exclusions. It will claim your cheapest branded conversions and inflate its own numbers.
3. Do not judge the account on platform-reported ROAS alone. Validate against Shopify or GA4 and steer larger accounts by MER.
4. Do not chase a low CPC by buying broad, cheap, low-intent traffic. The metric that pays the bills is cost per acquisition, not cost per click.
5. Do not change targets and budgets every few days. Frequent changes keep the bidding model in perpetual learning and throw away what it learned.
