For each of the 12 dimensions, in a robust spreadsheet, I’ve identified: Area of Work. Old Model (what an agency did). New Model (powered by platform AI). New Company Role. New Agency Role. Change in Agency Scope. Estimated Contract Savings.
Since this will take a dozen newsletters to detail, I’ll compress 12 dimensions into 5 clusters and summarize your actions.
If enough Premium Subscribers are interested, I’ll share the robust spreadsheet.
1. Agency Activity Army.
Account and campaign architecture, “keyword” research and match-type sculpting, audience segmentation and targeting, radio & tv ad tactic structures. This was the beautiful old world of a hundred thinly sliced campaigns “for control.”
The machines devoured this work. PMax, Advantage+ (A+), similar collapse structure into a few asset groups that the algorithm allocates internally. Intelligence reads intent, using additional tens of thousands of data points beyond the keyword or post content. A+ audiences treat your segments as hints, and then finds converts you never listed.
The Agency's job is not building, it is deciding. Advertiser (you) sets the reward function hierarchy, brand/non-brand priorities, brand safety red lines, margin protection bands. The Agency’s real remaining role: One-time architecture design, and occasional restructure IF strategy changes. It is NOT monthly rebuild billed as progress.
Approx. 22% contract cost weight, work that can be reduced by approx. 78%.
2. The Bid & Pace Dancers.
Manual bids, budget adjustments, day parting, device modifiers, daily pacing, spend checks, disapproval fixes, “anomaly” checks, “hygiene.” Hours and hours and hours.
AI can do narrow intelligence at a scale and impact that surpassed us in late 2024.
Smart bidding to your reward function (see: TMAI #432: AI Unlock: Value Based Bidding), campaign budget optimization, automated alerts, auto-recovery, now all this work continuously, and better. You just need to let the AI learn. When it starts, results will dip a bit. Your Agency jumps in to rescue by cutting bids/something. The AI learning resets. The dip returns. Agency… You know the rest. No stable learning. Company keeps sucking.
Reminder: Over-touching does immense damage in an AI world. Your Agency is causing this harm and it is your fault - for not understanding that every “rescue” is a sabotage.
The Agency job shrinks to setting the true reward function (aka kpi, target), occasional configuration of guardrails IF business strategy changes, and, hardest of all, operate at the rhythm of the AI’s learning cycle.
Approx. 14% contract cost weight, work that can be reduced by approx. 73%.
3. The Assembly Line.
Trafficking, ad builds, combination making, ad formats \/ creatives \/ variations, torture of tagging, QA, shopping-feed babysitting.
Ad platforms – except TV, Radio, press – can do this at a scale AND relevance that will shock you. Assembling bits from your ads, text, site, building creative, generating variants with GenAI (try this on paid Amazon!) mixing image video audio in the best version per person. All they need: often just your site link (or core assets) and a reward function.
Note: One reason they can do this spectacularly well is that your creative is judged by your VP of Creative/HiPPO, while the creative Ad Platforms deliver is judged by the business outcome delivered. 🤯
The Agency’s role narrows to the genuinely useful: Asset preparation, clean taxonomy, and spot-QA. Feed management is the one line that remains meaningfully skilled for now – sadly, data quality is still human.
Approx. 12% contract cost weight, work that can be reduced by approx. 45%.
4. The Optimization Theater.
Daily rituals of pausing “losers,” shifting budgets by feel, endless small A/B tweaks, the deeply irritating unfocused “learning agenda” items that will never amount to much.
Often the single biggest cost, and now the most destructive thing your agency is doing.
Modern AI-led platforms run their own continuous explore-exploit; they are already testing combos of creatives, bids, targeting to find high-value customers – at scale and cray cray speeds. They can also account for learning cycles that can span weeks/months, try that with a human.
The Agency’s role is cut down to: Few, really big (as in 15%+ increase in revenue), clean experiments, with the right altitude (creative concept – not execution -, audience attribute).
Approx. 16% contract cost weight, work that can be reduced by approx. 75%.
[Premium Subscribers: The role of what an Analyst does is quickly changing. Your current role will be gone by Jan 2028, replaced by a better one. If you are willing to learn and change, here's how to save your career: TMAI #495: Analyst 2028: S.H.I.F.T For Relevance.]
5. The Reporting & Servicing Factory.
Manually pulled weekly decks/spreadsheets/update emails. 2x Weekly “update” “check-ins” with 6-36 Agency attendees. Exhaustive reports on placements and “brand safety.” Agency “account management” meetings with other vendors and partners. Detailed hand written commentary of the data already in dashboards. Urgent need it by EOD cuts of existing decks/reports. “Client Training Sessions,” with content already easily available to all anyway. Dinners, outings, offsites, to ensure client happiness.
AI cannot solve for all these items, certainly not dinners and outings.
But, dashboards and reports are disappearing with Claude-fronted data lakes that can do both what and why to such a great extent that you don’t have to be held hostage by the Agency’s 17-tab spreadsheet. Platforms for automated placement filters, and brand safety tooling that gets better by the hour. Since there is ever increasing automation in Meta and Google, you need drastically less “account management,” and certainly not as many meetings.
The Agency’s work is limited to: Aforementioned dinners. Automating in-flight optimization. Shifting focus from Care (99% today) to Do & Impact (TMAI #391: Data Storytelling Framework: Care, Do, Impact).
Approx. 30% contract cost weight, work that can be reduced by approx. 60%.
Punchline: Contract Cost Weight * Reduction in Work and in my case equals 65% savings. For you, it might be a little less or a little more. Use the table and math above to discover.
The work is not going away, a lot of it is moving to machines. What’s left is more useful, higher impact work. Agencies don’t sell motion anymore. They sell judgement.
