Measuring True Marketing ROI With AI Attribution

The ROI problem no dashboard solves

Ask ten marketers what their ROI was last quarter and you'll get ten different math problems. One counts last-click conversions from Google Ads. Another trusts the number Meta reports inside Ads Manager. A third pulls closed-won from the CRM and divides by ad spend. None of them agree, and none of them are right.

The gap comes from three broken assumptions:

• That a click is a lead.
• That a lead is a customer.
• That the last touch before the sale deserves the credit.

AI attribution exists to fix these assumptions with data, not opinion. Here's what that actually looks like in practice.

What "true ROI" actually requires

To measure real return on a marketing dollar, you need four things stitched together:

1. Every touch a person had with your brand, across paid, organic, email, and direct.

2. The identity resolution to know that the anonymous website visitor on Tuesday is the same person who filled out a form on Friday and closed a deal six weeks later.

3. The revenue outcome — not a form fill, not an MQL, but signed contract value or cash collected.

4. A weighting model that assigns fractional credit to touches based on their actual influence.

Most small and mid-sized teams have pieces one and three living in different tools that don't talk to each other. That's why the numbers never reconcile.

Where legacy attribution breaks

Last-click attribution rewards the channel closest to the finish line. That's usually branded search or a retargeting ad — channels that harvest demand rather than create it. If you optimize spend based on last-click, you'll defund the top-of-funnel work that's actually driving pipeline, then wonder why growth stalls two quarters later.

First-touch has the opposite problem. It over-credits discovery and ignores the nurture work that closes deals.

Rules-based multi-touch models (40/20/40, U-shape, time decay) are better, but they're still guesses. The weights are picked by a human, not learned from your data.

What AI attribution changes

An AI attribution model learns credit assignment from the patterns in your own conversion data. Instead of assuming the middle touches are worth 20%, it looks at thousands of paths — some that converted, some that didn't — and calculates how much each touchpoint actually moved the probability of a close.

A few things this unlocks that rules-based models can't:

• Channel interaction effects. The model can detect that your LinkedIn ads only convert well when paired with a follow-up email sequence within four days.
• Diminishing returns curves. It flags the point where another $1,000 into a campaign stops producing incremental revenue.
• Negative contributors. Some touches actively hurt conversion — a poorly timed retargeting ad, a nurture email that triggers unsubscribes. AI models surface these; rules-based ones can't.
• Deal-size weighting. A touch that leads to a $50K contract shouldn't get the same credit as one that leads to a $500 order.

The data plumbing you need

AI attribution is only as good as the data feeding it. Before you trust any model's output, make sure you have:

• Server-side tracking for web events. Browser-based pixels lose 20-40% of events to ad blockers and iOS privacy changes.
• Deterministic identity stitching — matching email, phone, and user ID across sessions and devices, not just cookies.
• Ad platform cost data pulled via API, not exported from CSVs. You want daily spend by campaign, ad set, and creative.
• Closed-revenue data flowing back from your CRM into the same system that holds the touch data.
• Offline conversions for anything that closes over a call or in person.

If any one of these is missing, your ROI number is a fiction dressed up as a report.

Metrics worth watching

Once the pipes are connected, stop reporting on CPL and CTR as if they're outcomes. The metrics that reflect real ROI:

• Marketing-sourced revenue by channel, with fractional credit assigned by the model.
• Payback period — how many days between spend and recouped revenue.
• Incremental CAC — the cost of acquiring the *next* customer at current spend levels, not the average across all past customers.
• Contribution margin per channel, accounting for the fully loaded cost of the campaign, not just media spend.
• Pipeline velocity by first-touch source. Some channels bring in leads that close in 14 days. Others take 90. Both can be profitable, but you need to plan cash flow accordingly.

How Plyto approaches this

Plyto runs the CRM, email, ads, and site tracking inside one system, which means the touch data and the revenue data live in the same place from day one. There's no CSV reconciliation, no UTM archaeology, no arguing with sales about which lead came from where.

The attribution model updates continuously as deals close, so the credit assigned to a June ad campaign can shift in September when a long-cycle deal signs. Ad budget recommendations are tied to the same model — if a channel's contribution drops, spend gets reallocated automatically, with a full audit trail of why.

The point isn't to produce a prettier dashboard. It's to make every dollar of marketing spend traceable to a dollar of closed revenue, so you stop guessing which half of your budget is working.

Where to start this quarter

If you're not ready to overhaul your stack, start here:

1. Pick one channel where you suspect the reported ROI is wrong. Usually it's branded search or retargeting.

2. Pull the last 90 days of closed deals and manually trace the first three touchpoints for each.

3. Compare that to what your ad platforms claim they contributed.

The gap you find is the reason AI attribution exists.