There is a scene that repeats itself in almost every results meeting: someone projects a slide with the quarter’s reach and impressions, the client nods, and then asks, “OK, and how much money has this brought us?” The silence that follows explains why so many teams measure a lot and understand little. I’ve seen it in ecommerce accounts, in local businesses with two active networks, and in B2B brands with sales cycles lasting months. The practical conclusion: measuring results on social media isn’t about choosing the flashiest metric, but about building a system that connects what happens in the feed with a business decision.
An ecommerce case: the problem wasn’t ROAS, it was attribution
A fashion project with a small catalog (fewer than 200 SKUs) measured Instagram performance using the ROAS reported by the platform itself. The numbers looked good, but the business wasn’t growing at the pace that dashboard indicated. When cross-referencing data with web analytics, the classic pattern emerged: many users saw the product in the feed, didn’t buy, and came back days later searching for the brand directly on Google or typing in the URL. That sale was recorded as “direct” or “organic,” not as social.
The solution wasn’t magic. A more honest attribution window was implemented, campaigns were tagged with consistent parameters, and social performance began to be viewed through two lenses: the platform’s (useful for optimizing creatives) and the business’s (useful for deciding budget). The lesson that can only be learned by managing real accounts: a social network’s native dashboard is never neutral, because its goal is for you to keep investing in it. Use it to optimize within the platform, but not as a single source of truth.
What to measure in that scenario
- Conversion rate of sessions coming from social versus the site average.
- New users versus returning users: if social only captures those who already know you, reach is a mirage.
- Brand searches on search engines after campaign peaks (an indirect signal of impact that the network’s analytics doesn’t capture).
- Average order value of users arriving from social, which usually differs from the rest.
Local business: when the goal isn’t a click but a call
At a clinic with two locations, the metric that mattered wasn’t engagement but how many people requested an appointment. They published content on Instagram and Facebook, and tracking was done through a link in bio to a form. The flaw: most potential patients called by phone, and those calls weren’t recorded in any system associated with the campaign.
Here it’s worth being transparent about the limitations. Measuring calls attributed to social media is imperfect: dynamic tracking numbers work well on websites with high traffic, but in small businesses they can cause confusion if the customer sees a number different from the usual one. In that case, a reasonable alternative is to use discount codes or mentions like “tell me you’re coming from Instagram” for a limited period. It isn’t precise, but it’s measurable and cheap. According to documentation from the call-tracking platforms themselves, click attribution on a number has configurable time windows that change the result significantly: the same month can appear 40% better just by widening the attribution window.
A frequent mistake in local
Counting followers as a health indicator. I’ve seen profiles with thousands of purchased followers that generated zero real interactions and, worse, distorted content distribution because the algorithm showed posts to an inactive audience. Before measuring results, audit audience quality: proportion of real accounts, comments with substance, direct messages that arrive. If that base is contaminated, any subsequent metric is noise.
B2B: the long funnel breaks vanity metrics
At a software company with long sales cycles, a post with 300 impressions can be worth more than one with 30,000. The reason: it’s read by the operations manager of the target account. Measuring that requires abandoning reach as the main KPI and working with quality signals.
An approach that works in practice: target account lists, manual tracking of interactions from those accounts (likes, comments, profile visits), and a simple score that prioritizes whom the sales team contacts. It isn’t scalable to thousands of accounts, and it should be said: this method doesn’t work for mass-consumer brands or for teams without the capacity for manual tracking. It’s a precision tool, not a volume tool.
In this context, the metrics that provide information are:
- Interactions from accounts included in the target list (not the total).
- Visits to product or case study pages from social, filtered by company when possible.
- Inbound contacts that mention a specific post.
- Time until the sales team’s first response after a signal of interest.
How to build a dashboard that doesn’t mislead
After going through several projects, a pattern repeats: useful dashboards have few rows. If a metric doesn’t change any decision, it’s superfluous. I propose a three-layer structure that I’ve applied with reasonable results:
Layer 1: account health
It serves to detect operational problems, not to justify budget. It includes posting frequency, interaction rate relative to account size, net growth of qualified followers, and response time to messages. If something here collapses, it needs to be reviewed before scaling investment.
Layer 2: content performance
It compares formats and topics against each other, not against an abstract ideal. A useful practice: group posts by topic and calculate the median interactions per group. Medians resist outliers better than averages, and on social media viral peaks distort any average. According to industry studies, the distribution of interactions follows a very uneven curve: a small percentage of posts concentrates most of the engagement. That’s why judging a month by its best post is a measurement error.
Layer 3: business impact
Here come conversions, revenue attributed with the aforementioned caveats, and cost per result. If the business is local, replace it with calls, reservations, or store visits. If it’s B2B, with opportunities created. This is the layer that survives algorithm changes and platform trends.
What can’t be measured (and should be admitted)
There’s a part of social media work that resists quantification. Brand positioning, accumulated trust, or the effect of a post that a customer remembers weeks later don’t appear cleanly on any dashboard. The temptation is to invent a composite metric that looks rigorous, but that’s usually worse than accepting uncertainty. An honest alternative: complement the data with periodic qualitative research, asking customers how they found you. It’s slow and doesn’t scale, but it corrects the biases of digital attribution.
It’s also worth acknowledging an important technical limitation: privacy changes in operating systems and browsers have reduced the ability to track users across sites. Any comparison of social results between different years must be made with caution, because today’s measurement isn’t comparable to that of three years ago even if the tool is the same.
A measurement routine that works in small teams
For teams with limited resources, a realistic cadence:
- Weekly: review unanswered messages and comments, and note brand mentions outside your own accounts. It’s the cheapest source of information and the most ignored.
- Biweekly: compare performance by topic and decide what to repeat and what to discard.
- Monthly: cross-reference platform data with web analytics and close the report with a single question: what are we going to do differently next month?
- Quarterly: review whether the objectives are still valid. A common mistake is keeping the same KPI for years even though the business has changed priorities.
Closing: the metric that matters is the one that changes a decision
Measuring results on social media is less sophisticated than the tool industry sells and more demanding than it seems. It requires knowing the limitations of attribution, distrusting dashboards that only show what’s favorable, and accepting that some things can’t be quantified precisely. The cases that work share one trait: the team knows what decision it will make with each number before looking at it. If a data point doesn’t have an associated decision, it’s probably just decorating the report.