Social Media ROI: The Honest Version for Agencies

In 2026, only about 37% of marketers say it is somewhat or very easy to tie social media activity directly to business outcomes, out of more than 1,100 surveyed (HubSpot, 2026). The other 63% are still asked for the number every quarter.

That is the tension this post resolves. Your client's finance lead wants one figure with a dollar sign on it, and the honest answer is that the figure cannot be isolated cleanly. Pretending otherwise is what loses retainers, because a number you cannot defend under a follow-up question is worse than no number at all.

Key Takeaways

• Click-based reporting missed 19% of the search visits paid social actually caused across 17 Meta lift tests (Brainlabs, figures current as of 2026).
• The ROI formula is arithmetic. The fight is always over fully loaded cost, attributed revenue and the time window.
• Report a floor and a ceiling, not a single fake point estimate.
• When revenue attribution is impossible, branded search lift and cost per qualified inbound survive scrutiny. Engagement rate does not.

What can social actually be attributed to in 2026?

Less than your analytics dashboard implies. SparkToro tracked 1,113 real visits across 11 networks and found that 100% of visits from TikTok, WhatsApp, Slack, Discord and Mastodon arrived with no referrer at all, landing in reports as direct traffic (SparkToro, 2023).

That is the dark social problem stated plainly. Someone sees a client's Reel, screenshots it, sends it to a colleague on WhatsApp, and the colleague buys three weeks later after a Google search. Nothing in that chain is recorded as social. It is recorded as direct, or as branded organic search, and both of those look like credit the client already owned.

Sharing surface	Share of visits arriving with no referrer
TikTok, WhatsApp, Slack, Discord	100%
Facebook Messenger	75%
Instagram DMs	30%
LinkedIn public posts	14%
Pinterest	12%

Source: SparkToro, 2023.

The browser layer removes more. Safari caps the lifetime of any cookie written by JavaScript at seven days (WebKit, 2020), so a client-side attribution window longer than a week is fiction on roughly a quarter of traffic. App Tracking Transparency did the same to mobile signal, and cross-domain journeys between a client's site, a booking tool and a payment processor now break by default rather than by accident.

Here is the part most agencies get backwards: consent banners are not the main culprit in Europe. Didomi's 2026 benchmark, drawn from hundreds of millions of consent interactions during 2025, puts European consent rates between 75.1% in Western Europe and 89.3% in Eastern Europe (Didomi, 2026). Most people say yes. The signal still evaporates, because the break is in referrers and cookie lifetimes, not in the banner.

How badly does last-click undercount social?

By a measurable margin, and the direction is always the same. As of 2026, across 17 Meta conversion lift tests spanning 12 advertisers, paid social drove 19% more search visits than the holdout control, and none of those visits appeared in any report as social-driven (Brainlabs, 2026).

Read the split inside that number, because it is the useful part. Only 31% of the incremental search traffic was branded. The other 69% was non-branded, meaning exposure to social changed which result people clicked on a generic query weeks later. Last-click hands all of that to search.

Now apply the logic to organic social, which is what most retainers actually deliver. Paid social at least has a pixel, a click ID and a platform-side view-through count. Organic has none of those. If the measured undercount on the instrumented channel is 19%, the undercount on the uninstrumented one is not smaller. Nobody has run the equivalent holdout study on organic because you cannot buy a control group for a brand's own feed.

So the defensible statement to a client is not "social drove X". It is "last-click gives social X, and every controlled study we have says the real figure is higher by an amount we cannot measure on your account".

The formula and the three inputs everyone fudges

The formula is one line: ROI equals attributed revenue minus fully loaded cost, divided by fully loaded cost. The reason it produces nonsense is input quality, not maths. Only 57% of marketers prioritise revenue metrics when measuring social ROI at all (Sprout Social, 2026), which means nearly half are dividing something by something else and calling it return.

Fully loaded cost. Most calculations use ad spend and stop. That is the single biggest reason published social ROI figures look implausible. Fully loaded means your retainer, the ad budget, content production billed outside the retainer, tools, and the client-side hours their own marketing manager spends on reviews, approvals and asset gathering. With a US social media manager averaging $71,410 in 2026 (Glassdoor, 2026), those internal hours are not rounding errors.

Attributed revenue. Pick a model and name it in the report. Last non-direct click is the conservative floor. Any-touch assisted conversion is the ceiling. Reporting the ceiling as if it were the floor is the thing that gets an agency fired six months later when the client's own analyst checks.

Time window. A 30 day window on a considered purchase measures nothing. Set the window from the client's actual sales cycle, and get that number from their CRM before you write the first report, not after they challenge the fourth one.

A worked example a finance team can follow

Sprout Social puts the standard baseline across industries at a 3:1 return, with strong paid campaigns nearer 5:1 (Sprout Social, 2026). Those figures are almost always calculated on ad spend alone. Watch what happens to them under a fully loaded denominator.

Take a DTC client on a $6,000 monthly retainer over twelve months.

Costs: $72,000 retainer, plus $30,000 in paid amplification at $2,500 a month, plus $9,600 for a freelance photographer and editor outside the retainer, plus $1,800 in tools. Then the client's own time: their marketing manager spends four hours a week on social, which is 208 hours a year, and at a fully loaded internal rate of roughly $45 an hour that is $9,283.

Total cost: $122,683.

Return, floor case. GA4 last non-direct click credits social with $148,000 of revenue over the same twelve months. ROI is (148,000 minus 122,683) divided by 122,683, which is 20.6%. Revenue to cost, that is 1.21 to 1.

Return, ceiling case. Sessions where social appears anywhere in the path add another $96,000. Total $244,000, ROI 98.9%, revenue to cost 1.99 to 1.

Both numbers are real and neither is true. The truth sits between them, and it sits there because of the referrer gaps and cookie caps above, not because you did the arithmetic badly. Notice also that this client, on either figure, is nowhere near the 3:1 baseline everyone quotes. That comparison is not apples to apples, and saying so before the client notices is worth more than a flattering slide.

Which proxy measures hold up under scrutiny?

Four of them, and engagement rate is not one. Some 68% of marketing leaders currently define social ROI through engagement metrics (Sprout Social, 2026), which is exactly why a finance-minded client dismisses the whole report the moment they see a likes chart.

Branded search lift. Pull branded query volume from the client's Search Console and plot it against your posting and paid calendar. The Meta lift work above showed social causing search that search then takes credit for. This is the cheapest way to make that visible without running a holdout.

Direct traffic correlation. Given that dark social lands in direct, direct traffic is a partial social proxy rather than a mystery bucket. Correlate weekly direct sessions against weekly social reach. A relationship that holds over two quarters is evidence. One good week is not.

Assisted conversions. Report the count and value of conversions where social appears in the path but not last. Present it as the ceiling, explicitly labelled.

Cost per qualified inbound. The strongest of the four, and the least used. Add one required field to the client's lead form asking how the person first heard of them, then divide fully loaded social cost by qualified inbounds naming social. Self-reported attribution is noisy, but it is the only method that captures the WhatsApp conversation your pixel will never see. Track it alongside the metrics in our guide to social media KPIs that survive client scrutiny.

What do you say when you cannot isolate the number?

You say it in the first meeting, not the fourth. The average B2B buying cycle ran 10.1 months in 2025, down from 11.3, with buyers first contacting a seller around 61% of the way through their own process, across nearly 4,000 respondents (6sense, 2025). If most of the journey happens before anyone identifies themselves, no attribution model can see it.

The script that works is three sentences. We cannot isolate social's contribution to revenue, and neither can anyone who tells you they can. Here is the floor, here is the ceiling, and here is why the gap exists. Here is what we can prove moved. Then you fill that last sentence with real movement on the proxies you chose: branded query volume, cost per qualified inbound, share of voice against the competitors you track.

That last part matters. An agency that names its measurement limits and then produces movement on things it can actually measure reads as competent. One that produces a confident single number reads as either naive or evasive, and the finance person in the room will decide which. Which impression do you want in the transcript when the renewal comes up?

Set the frame quarterly rather than monthly. Monthly reports are for operations, and belong in your client social media report. The ROI conversation belongs in the quarterly business review, where the window is long enough for the answer to mean anything. Tools like OwlStat keep daily snapshots of public accounts past Instagram's own 90 day history limit, which is what makes a twelve month baseline possible at all when the client changes agencies or platforms mid-contract.

Agorapulse's session on translating social results for different stakeholders is a decent companion to that script, particularly the part on adjusting the same data for a founder and a CFO.

If you want the competitive half of the picture, benchmarking against rival accounts gives you a denominator that does not depend on any tracking at all. Our walkthrough of Instagram competitor analysis covers the method, and reporting cadence for agencies covers how often to show it.

Where to go next

• Social media KPIs that survive client scrutiny for the metric set that feeds this calculation.
• Running a quarterly business review for the meeting where the ROI conversation actually belongs.
• Building a client social media report for the monthly document that sits underneath it.