Drift Media — Agency consolidated Meta + Google reporting across 14 clients

The situation

Drift Media is a six-person performance-marketing agency focused on DTC brands in the $50k–$500k/mo revenue range. They manage Meta and Google for 14 retainer clients across skincare, apparel, food and beverage, and home goods. Before Admaxxer, the operational stack looked like most mid-size agencies in 2026: Meta Business Manager with 14 client assets, Google Ads Manager Accounts (MCC) with 14 linked accounts, AgencyAnalytics and ReportsMate for white-label PDFs, Google Sheets for custom metrics the off-the-shelf tools couldn't compute, Slack for client comms, and Notion for internal SOPs. Total tool cost at baseline was about $1,400/mo across reporting platforms plus individual seats.

Each client got a monthly PDF report, a weekly Slack update on pacing, and ad-hoc responses whenever a metric moved. The weekly pacing update was the expensive one: a media buyer would log into Meta Ads Manager, pull spend and ROAS, log into Google Ads, pull spend and conversions, cross-check against the client's Shopify or WooCommerce revenue, format a Slack message, and hit send. Six hours per client, fourteen clients, eighty-four hours total — a fully loaded head of delivery every single week consumed on reporting instead of optimization. The work was repetitive, error-prone at 1 AM on a Friday, and completely invisible to clients who only saw the polished Monday morning output.

This pattern is well-documented across agency subreddits and operator communities: "every report becomes a manual scavenger hunt involving logging into each platform separately, exporting CSVs, reconciling date ranges, and hoping nothing is missed." Multi-platform agencies routinely "run campaigns across five to ten different platforms for each client — Meta, Google, LinkedIn, TikTok, plus email platforms, CRMs, and analytics tools," and the time cost compounds linearly with client count. Drift Media had two failure modes as a result: client churn when performance dips weren't flagged quickly enough, and senior-buyer burnout from repetitive reporting work that never moved a campaign.

The problem

Two measurable problems came out of the reporting drag. First, response time on performance issues was averaging 36–48 hours. A client would DM on a Monday about a weekend ROAS drop; the buyer wouldn't have the context to answer until Tuesday or Wednesday. In the last 12 months, Drift had lost three clients where the stated reason was "slow to catch performance issues." Second, blended performance across Meta and Google was never the number clients actually asked about — clients wanted to know blended MER (total revenue / total paid spend), but the agency could only reliably report platform-level ROAS. The monthly PDF had a "blended" section, but it was recomputed manually and nobody trusted it past the second decimal place.

The math was straightforward. Six hours per week per client, $150/hour fully loaded, 14 clients, 52 weeks = roughly $655,200 of annual delivery capacity consumed on reporting. That is roughly the revenue of three mid-tier retainers paid out as internal cost.

What they did with Admaxxer

• Step 1 — Provision 14 workspaces. Upgraded the agency owner to the Scale plan ($199/mo) which includes unlimited workspaces. Each client got their own isolated workspace with their own ad-platform connections, their own pixel, and their own user invites. The data model enforces workspace isolation at the database level — there is no way for a query from Client A's workspace to accidentally pull Client B's data.
• Step 2 — Connect Meta and Google per workspace. Used the Meta paste-token flow and Google OAuth for each client. Tokens are encrypted AES-256-GCM at rest, per-workspace, never written to logs. Total onboarding time across all 14 clients was about 90 minutes — the bottleneck was reaching out to each client to confirm their Meta Business Manager permissions and Google MCC access were set correctly before the paste-token step.
• Step 3 — Install the pixel on each client storefront. 11 of 14 clients were on Shopify; 3 were on custom or headless stacks (one BigCommerce, two custom React storefronts). The pixel went onto all of them — no platform lock-in, same event schema across every stack.
• Step 4 — Build the standard weekly report template. Created a single dashboard view per workspace: blended MER, CAPI match rate, top 5 ad sets by true contribution, 7-day LTV by acquisition channel, pacing against monthly budget, and week-over-week deltas. Each client's dashboard shared its URL (read-only) with the client so they could self-serve when curious. The agency also white-labeled the dashboard URL under a reports.driftmedia.co subdomain for a more agency-branded presentation.
• Step 5 — Wire the Claude agent for "why did X" questions. Every client workspace got a shared analytics chat powered by the same Claude Sonnet 4.6 agent with prompt caching on the system block and tools array so repeat questions stay fast and cheap. A buyer could ask "why did ROAS drop on March 14" and the Claude agent would check get_campaign_insights, query_metrics, and return a diagnosis in 30 seconds — budget hit cap, CAPI match rate dipped, a creative fatigued, a new competitor auction, iOS 14 signal loss that day, etc. The agent cites which pipe and which date range it used, so the buyer can verify the answer against the underlying dashboard before responding to the client.
• Step 6 — Schedule a weekly "weekend performance summary" per client. Using the agent's query_metrics tool on a BullMQ cron (Sundays at 10 PM), the system auto-generates a summary message per workspace flagging any ROAS drops >15%, CAPI match rate drops >5 points, or pacing anomalies. The buyer reads 14 auto-generated summaries in 20 minutes Monday morning instead of rebuilding them from scratch.

The results

After 90 days:

The agency redirected 63 freed hours per week into testing new creatives, running geo-lift tests, and taking on two additional retainer clients without hiring — roughly $18k/mo in incremental revenue.

Why this worked

The Admaxxer Agency plan is designed for exactly this shape of operator. Unlimited workspaces with isolated ad-platform connections mean each client is a clean tenant; no risk of cross-account data bleed. Ad-platform tokens are encrypted AES-256-GCM at rest per-workspace, so even if one client's token were compromised the blast radius stops at that workspace. The same Claude agent runs per-workspace with the same 6 tools (list_campaigns, get_campaign_insights, get_account_insights, update_campaign, pause_all_low_roas, query_metrics), so a buyer's muscle memory transfers across every client.

Crucially, because the agent reads directly from the same Meta Marketing API v21.0 and Google Ads GAQL data plus 33+ Tinybird pipes that power the UI, there is no "does the agent see the same thing the dashboard sees" question — it is literally the same query path. When the agent says "ROAS dropped 18% on March 14," that number came from the same pipe the client's dashboard rendered. No agency has to reconcile two systems of record because there is only one.

The destructive-action guardrails matter for agencies specifically. update_campaign and pause_all_low_roas require explicit confirmed: true — a junior buyer cannot accidentally pause a client's top campaign from a chat misfire. That single detail made the agent safe to hand to every buyer on the team without a layer of senior review. Agencies have historically been risk-averse about AI in client accounts precisely because the blast radius of an accidental pause is reputational, not just financial; the destructive-action confirmation model removes the scariest failure mode.

The 63 freed hours a week is the number that matters to the agency P&L, but the 30-minute response time is the number that matters to client retention. When a client DMs on a Monday morning asking why Saturday's ROAS looked off, the buyer can now answer before they've finished their first coffee — not on Wednesday, not after a spreadsheet rebuild. That responsiveness is the single biggest retention factor for performance retainers, and it is a direct output of the agent-plus-unified-data architecture.

What other performance agencies can learn

• One workspace per client is the right tenancy model. Shared connections across clients create audit headaches and blast-radius risk — one compromised client Slack can expose every client's ad data if the data model lets it.
• Reporting time is silent margin. Every hour on a weekly update is an hour not spent on optimization — solve that first, before adding headcount. Agencies that scale by hiring junior buyers to do reports are scaling the wrong cost center.
• Blended MER should be the default client-facing metric, not platform ROAS. Clients don't care whether Meta or Google "won" — they care what happened to revenue per dollar of spend.
• A Claude-style agent earns its keep on "why did X" questions, not automation. Buyers don't want ads paused for them; they want a 30-second answer so they can decide.
• The cheaper tool stack matters less than the time saved. Even without the $1,400/mo consolidation, the 63 freed buyer-hours would have paid for the plan 50x over.
• Give clients read-only access to their own workspace. Self-service kills 40% of inbound "quick question" Slacks and reframes the agency-client relationship from "gatekeeper" to "strategist."
• Response time on performance dips is the single most important retention signal. Agencies that answer Monday by Monday churn; agencies that answer Monday by Tuesday-morning renew.
• Standardize a weekly dashboard template across every client before you try to customize per client. Consistency is a force multiplier when a buyer moves between 14 accounts an hour.