Attribution, honestly: a 2026 guide for performance agencies.

Attribution is a religious war inside performance marketing. Every agency we've worked with has at least one person who thinks the pixel is fine and one who swears by a $3,000/mo third-party tool. They're usually both half-right and half-wrong, and the disagreement between them is costing the agency money.

This post is the version of the attribution conversation we wish somebody had given us before we spent a year trialing tools. It's opinionated but grounded in public pricing pages, platform documentation, and the kind of thing you can verify yourself in a weekend. No industry-report statistics, no “70% of CMOs agree.” Just what the four options are, when each one pays for itself, and where the money is disappearing for agencies that haven't sorted this out.

The short version

• There are four attribution schools, not two.Most agency conversations collapse into “pixel vs. third-party tool.” The real options are: platform pixel, third-party click-stitching (Hyros, Triple Whale), marketing mix modeling (MMM), and post-purchase survey attribution. Each one answers a different question. Picking wrong means paying for answers to questions you don't have.
• iOS14 broke less than you think, and more than vendors admit. Conversion rate tracking survived. Audience size estimation did not. Most agencies over-index on the first and under-invest in mitigations for the second.
• Attribution tool spend correlates poorly with decision quality.The agencies we've seen get the most out of attribution are not the ones paying the most for it. They're the ones with a clear rule for which attribution view answers which decision.
• The decision layer is where the real ROI lives. Attribution tells you what happened. A decision layer tells you what to do about it. Most agencies buy a $500–$3,000/mo attribution tool and then still spend two hours a morning deciding what to do with it.

The four attribution schools

1. Platform pixel attribution

Meta pixel + CAPI, Google Ads conversion tracking, TikTok pixel. The default. Every agency already has it. Post-iOS14 it's noisier than it was in 2019 but it still works for conversion rate comparisons at a given moment. What it's good at: telling you whether ad A converts better than ad B within the same platform, right now. What it's bad at: cross-platform attribution, long-window attribution, and anything involving phone calls or offline conversions.

Cost: free, but you need CAPI configured cleanly. Agencies that skip server-side CAPI are the ones complaining loudest about iOS14. Agencies that wired CAPI well still lost signal, but they lost a fraction of what the pixel-only agencies lost.

When it's enough:lead-gen verticals with short windows (same-day to 7-day), single-platform dominance (Meta or Google, not both), and no meaningful offline conversion flow. If a dental practice gets leads on Meta and books them within 48 hours, you don't need a third-party attribution stack. The pixel is fine.

2. Third-party click attribution (Hyros, Triple Whale)

Click-level attribution with cross-session stitching. Works by wrapping every outbound link in a tracking domain, reading first-party cookies, and reconciling against pixel data. Hyros and Triple Whale both do this; their differentiators are the UI, the reporting layer, and the vertical fit.

Hyros has historically been strongest for high-ticket info-marketing and coaching businesses — long purchase windows, phone-heavy sales flows, and high AOVs that justify the price. Triple Whale has been strongest for Shopify ecommerce — post-iOS14 server-side recovery, cohort LTV, and creative-level attribution inside a single store.

Cost: Hyros starts around $499/mo and scales with ad spend, often $2,000–$10,000/mo at volume. Triple Whale starts at $129/mo and scales similarly but less aggressively. Both price in enterprise tiers above. Triple Whale offers a 14-day trial; Hyros does custom quotes.

When it's worth it: high AOV ($2,000+), long purchase windows (30–90 days), multi-channel funnels where Meta sparks and Google closes, or phone-heavy sales flows where call stitching is a core KPI. For agencies managing lead-gen verticals with $50–$200 CPLs and fast close windows, these tools are usually overpowered.

3. Marketing mix modeling (MMM)

Regression-based attribution that models the effect of each channel on a topline metric (revenue, leads, appointments) across time. Google's Meridian, Meta's Robyn, and Recast are the open-source / semi-open options. Enterprise MMM vendors charge meaningful fees; Recast is the most agency-accessible.

MMM answers a different question than the first two schools. It's not “which ad drove this lead.” It's “if I pull $10K from Meta and put it into Google next quarter, what happens to total revenue?” That's a question click-attribution tools can't answer cleanly, because they don't model incrementality — they model attribution.

Cost: open-source is free-ish (your time to run it); Recast is in the low thousands per month per brand. Enterprise MMM is six figures annually.

When it's worth it:agencies managing $100K+/mo ad spend per client where channel-mix decisions are the primary lever. Below that spend level, the math doesn't stabilize. You'll get noisy confidence intervals that won't survive a client review.

4. Post-purchase survey attribution

Ask the customer. “Where did you hear about us?” on the post-purchase page (KnoCommerce, Fairing). Cheapest option; most honest in one dimension and misleading in another.

Surveys are honest about which channel gets creditin the customer's head. They're not reliable for which channel actually drove the click. Customers who saw three Meta ads and then clicked a Google brand search will tell you “Google.” Useful, but not for optimizing Meta spend.

Cost: $50–$300/mo for the survey tool, plus the minor conversion-rate drag of adding a page to checkout.

When it's worth it:ecommerce brands where brand-awareness channels (podcast ads, influencer sponsorships) are in the mix and click-attribution can't see them. Pair with click-attribution, don't use as a replacement.

The iOS14 story, corrected

iOS14's App Tracking Transparency (ATT) launched in April 2021. The story every attribution vendor has told since is some version of “you lost visibility; buy our tool to get it back.” That's half-right. Here's the other half.

What actually broke:deterministic 1:1 matching between a specific ad impression and a specific conversion, for users on iOS who opted out of tracking. Meta's reporting became modeled rather than measured for a meaningful fraction of traffic (~20–30%, depending on the vertical). Audience-size estimation for lookalikes lost precision.

What did not break:conversion rate comparisons within the same audience, creative performance comparisons, within-platform optimization, and the fundamental shape of a healthy funnel. The platforms' internal optimizers still optimize. iOS14 didn't turn off Meta's ability to allocate budget to likely converters; it reduced Meta's ability to tell you which specific impression caused which specific purchase.

What this means for tool selection:if you're optimizing within Meta, the pixel + CAPI combination is still sufficient. If you're comparing Meta to Google to TikTok at the level of budget allocation decisions, click-attribution tools help but MMM helps more. Agencies that bought a click-attribution tool expecting it to “fix iOS14” often describe the tool as disappointing six months in. The tool did what it was supposed to do. The expectation was wrong.

A practical playbook for a 10–50 client agency book

1. Wire CAPI cleanly on every client. Today.Meta's Conversions API is free and it's the single highest-ROI attribution move available. If you're still relying on pixel alone for clients with iOS traffic, you're leaking 20–30% of your reporting signal for no reason.
2. Use platform attribution as the default. Layer up only when the shape requires it. For most lead-gen clients with CPLs under $200 and close windows under 14 days, Meta pixel + CAPI + Google Ads conversion tracking is enough. Layering a $500/mo attribution tool on top of a $3K/mo ad account is often economically upside-down.
3. Reserve third-party click-attribution for clients with the shape.High AOV, long window, phone-heavy, multi-channel funnels. If a client's book doesn't match, the tool is an expense the client is paying for your comfort, not their outcome.
4. Consider MMM at the agency level, not per-client. If you manage many clients in the same vertical, an MMM view across the book can surface incrementality patterns that no per-client click-attribution tool can. One Recast instance running across fifteen dental practices beats fifteen separate Triple Whale accounts.
5. Budget attribution tooling at 2–5% of managed spend, not fixed dollars.Clients with $5K/mo ad spend shouldn't carry $500/mo attribution tooling; clients with $50K/mo easily should. Price the attribution investment to the book, not to the tool's marketing.
6. Separate the attribution decision from the action decision.This is the mistake we see most. Agencies buy attribution and then still spend two hours a morning deciding what to do with the signal. The attribution tool answered “what happened.” Nobody answered “what to do.” That's a separate layer and it's where the margin is.

Where we fit (and where we don't)

Full disclosure: we built Zeke AI and sell it. Here's the honest version of where we fit in the attribution conversation.

We don't compete with Hyros or Triple Whale on attribution math. They're better at cross-session click stitching than we are, and we don't plan to catch up — it's not our lane. What we do is the decision layer that sits on top of whatever attribution you're already running. You keep your Meta pixel, your Google Ads conversion tracking, your Hyros or your Triple Whale if you have it, your post-purchase survey. We read those signals — plus your internal playbooks, SOWs, and house rules — and produce a morning queue of 9 evidence-backed recommended actions across your book.

If your book is lead-gen verticals with CPLs under $200 and the pixel + CAPI are sufficient for attribution, our Starter tier at $197/mo for 10 clients is a fraction of what you'd pay for Hyros and arguably more useful, because the output is decisions, not data. If your book genuinely requires Hyros or Triple Whale for the attribution shape, keep them — and use us above them. We're designed to complement, not replace.

Further reading inside Zeke

• Zeke vs. Hyros — where the lanes differ and where a Hyros-using agency still benefits from a decision layer above it.
• Zeke vs. Triple Whale — why Triple Whale is built for one ecom store and Zeke is built for an agency book, and when each is the right call.
• The state of agency operations, 2026 — the broader context this post lives inside: where the agency workweek goes, why reporting ate so much of it, and what the next generation of tools is going to look like.