Singapore

SEM agency: the buying brain moves, your signals stay

As of November 2026, the useful question to ask any sem agency is not how fast its dashboard moves, but who owns the signals its AI decides on and what you keep if the vendor behind it is bought. Digiday published Ronan Shields' report on the debate at LUMA Partners' Digital Marketing Summit on November 13, 2026, and the argument inside it is a procurement question before it is a technology one. This piece sets out three kinds of buying brain you will be offered, how to tell them apart from a proposal document, and a checklist to run before signing.

Key takeaways

LUMA's Terry Kawaja framed efficiency as a 10% incremental improvement, against a 10x difference for effectiveness, and the summit's speakers put the competitive edge on decisioning and unique signals rather than campaign workflow.
Platform bidding systems decide impressions inside their own walls; the choice of which platform gets the month's budget has to be made outside them.
An agency dashboard and a system built in your own tenant can look identical in a pitch and differ completely on the day you leave.
SEM is the paid side and SEO is the organic side, and the AI shift being discussed happens on the paid side, in how the decision gets made.

What changed in the buying layer

Two partners at LUMA opened the summit with a claim about how AI is reshaping ad tech: it automates work that used to be complicated, which moves the advantage away from running the delivery process and towards real-time decisions. Conor McKenna put it as the workflow moat evaporating, in comments at the November 9 event in New York City.

The statements are short and the inference is not small. If building and pacing a campaign becomes cheap, the agencies and platforms that charge for those mechanics lose the reason for their margin. What is left to sell is the judgement: which signals to feed the model, which creative to show, where the placement goes and how the result gets improved.

The buyer's version of the same shift is blunter. You are no longer paying somebody to operate a machine. You are paying somebody to decide what the machine should be told.

Why effectiveness, not workflow, sets the price

Kawaja's line at the event was that efficiency is the 10% incremental improvement while effectiveness could make a 10x difference. Read it as a pricing rule rather than a slogan.

Workflow savings are easy to quote and easy to copy, so they compete on price until the saving disappears. Effectiveness, measured on your own revenue, is harder to compare and much harder to copy, because it depends on signals nobody else has. A pitch built on hours saved competes with every other agency in the market. A pitch built on a decision you cannot make yourself does not.

When you sit in front of a proposal, the split usually shows up as the shape of the reporting. Workflow vendors report activity: campaigns launched, assets produced, hours logged. Effectiveness vendors report a funnel and where it leaks. Only the second kind of report can change what you spend next month, which is why sem google ads reporting that stops at click and lead counts leaves the buyer with nothing to decide on.

The second shift in the report sits on the consumer side: AI assistants can pull research, discovery and comparison into a single interface, compressing the path to purchase. The speakers expect people to stay in the loop for now, delegating tasks to agents rather than handing over the decision. If ads end up inside those assistants, trust decides whether the placement is worth anything. An earlier write-up on this site covered a related thread in How AI Slop and ChatGPT Search Share Reshape the SEM Brief.

Two planners study a funnel diagram on a wall screen, one tracing a leak midway down the funnel.

The comparison the RFP should force

Ask every shortlisted agency to fill in one table row per stream, using a real campaign you already run.

Stream	Who sets the budget, and where	What decides each impression	What leaves with you if you exit
Google Ads, Smart Bidding	You set the daily and monthly budget in the account	Smart Bidding allocates within Google's own traffic	The account and its history stay with the account owner
Meta and other platforms	Platform-level budgets, set per campaign	Platform algorithm, within its own system	The account, but attribution history built inside the platform
An agency's own dashboard	The agency's tool, often bundled into a retainer	Agency rules layered over platform bids	Usually the tool; the history is the disputed part
A system built in your tenant	Your account, your data layer	Your rules, with the platform doing the placement	Everything in your tenant, by construction

Ravi Patel of SWYM.ai, speaking separately to Digiday, made the point that belongs in the last column: a unified media platform can manage the what and the how much, but not the which impression, because Google, Meta and Amazon still make that call inside their systems. The row that matters most in a pitch is the last two. Everything above them is platform behaviour you cannot change, however the meeting goes.

Making the table into a test, not a question

You cannot get a straight answer to "who owns the data" in a sales meeting, because the answer is structural rather than a promise. Test the structure instead.

The tenant name that will hold your click, conversion and sales data, and whether you can log into it without the agency.
Which conversion events the platform actually receives, and whether your sales-qualified outcome is one of them or only form fills.
The last change made to your account without prior approval, and the approval record for it.
The benchmark the agency will be measured on, and whose accounts that benchmark comes from.
What happens to the attribution history on the day the contract ends, in writing.

Each question has a checkable answer. An agency operating a rented dashboard will answer the tenant and exit questions carefully and change the subject on the conversion feeds. A platform-first pitch usually covers the conversion feed and the benchmark well and has nothing to say about the tenant.

What the open web is being asked to prove

The same report notes that the open web's share of budgets is described by LUMA as fragmented against the scale and simplicity of large platforms. The speakers' answer is that the open internet has to deliver easier workflows and stronger outcomes together, because it cannot compete on either scale or simplicity.

For an advertiser, that reassessment lands on the split between platforms, and it puts pressure on any sem agency whose pitch is a cheaper way to run the same placements. The largest may be the easiest to buy, but "easy to buy" is the efficiency half of Kawaja's sentence. If the effectiveness half shows up elsewhere in your funnel, the easier buy is also the more expensive one per signed deal.

A buyer pauses at a fork between a crowded gated platform entrance and an open road of small billboards.

What is different between SEO and SEM here

The paid side of search is where this argument bites, so it is worth separating the two plainly.

SEM, search engine marketing, is the paid side: you buy placements per click or per impression and the platform's auction decides who is shown. SEO is the organic side: pages earn visibility through relevance and links, and no auction price is involved. They share a channel and a set of user intentions, which is why people mix them up in a budget meeting.

The split the summit discussed belongs to SEM, and more precisely to the decision layer under bidding, which is the part a sem agency is really selling. The model that decides which ad a searcher sees is the platforms' own; the choice of what to tell that model, how to structure the account around it and what counts as a result, still belongs to whoever runs the account. That is the part an SEM agency sells, whether or not it says so. Ad tech, for the sentence this piece owes the other common question, is the technology that handles the buying and delivery of ads, and the martech stack is the marketing systems on your side of the fence.

How Topkee approaches the brain behind the buy

Topkee's position on this is structural, and it starts from a boundary in the platform's own automation. Google Smart Bidding allocates traffic only inside Google's system, and a platform cannot set the split between platform A and platform B for the month. Google accepts CRM conversion data for training but fixes the attribution window, so conversions outside that window never reach the model. Those are constraints on the platform side of the fence, and Topkee's answer is a system on the advertiser's side of it.

TTO Funnel Tuning Agent is the part that decides. Each campaign gets its own agent instance, which stitches fragmented data into one funnel and produces an eight-chapter full-funnel diagnosis covering core KPIs and trends, leak mapping, spend structure, creative and audience waste, landing pages and lead capture, account health, and the action list with its reasoning. Account health is scored on six dimensions: creative diversity, ROAS performance, audience composition, budget efficiency, retention operations and lead-conversion experience. Every action in that list defaults to "pending confirmation", unauthorised account changes stay at zero, and each human choice is recorded so the next cycle can be back-tested. The same notes describe a closed-loop model in which qualified conversions are passed back to Google and Meta, and Smart Bidding then works from revenue rather than clicks. In that description the funnel runs from click through conversation and opportunity to revenue inside the client's own tenant, with a ttoclid tying cross-device and cross-channel touches into one identity. Leadbox logs prospect behaviour and pushes sales status back to each ad platform, which is the sales signal the buying brain runs on, and the same role a sem google ads account needs somebody to fill. The ownership claim is the one to test with the questions above: the client holds the system and the accounts, and therefore the optimisation history, rather than a dashboard that leaves with a retainer.

A team reviews a funnel model on their own office monitor, cables running towards distant platform buildings outside.

FAQ

Is SEM the same as Google Ads?

No. Google Ads is one place to run SEM, and SEM also covers other search engines and paid search placements. The distinction that matters for budgeting is between paid search and organic search, not between one ad platform and another.

If the platforms already optimise with AI, why pay an agency?

Because the platform optimises inside its own system. It can decide which impression to show, but not how your month's budget splits across platforms, and its attribution window sets which conversions it ever learns from. Someone has to make those decisions.

What is the difference between ad tech and martech?

Ad tech is the technology behind buying and delivering advertising, and martech is the marketing systems that run on your side: customer data, CRM, communications and measurement. When the summit talked about the buying brain, it was talking about the ad tech side.

We already run Google Ads ourselves. Does this argument change anything for us?

It changes what you should keep. A self-managed account already holds the asset that matters, the account and its history. What a system on your side adds is the cross-platform budget decision, the view of where the funnel leaks and the sales signal sent back to the bidder.

How do we know whether an agency's AI is ours or theirs?

Ask for the tenant name your data sits in and whether you can log into it without the agency. If the answer is a dashboard behind their login, the intelligence is the agency's. If it is a system under your ownership, it is yours.