REVIEW
Trustpilot · Google · Yelp
The volume sources, where an operational failure shows up as a rating before it shows up in a KPI.
Signal
What customers say in public, read in the same pipeline as what they say to you. Eighteen platforms, scored for authenticity, joined to the causes your own conversations already show.
[01 / THE PROBLEM]
The complaint in your queue and the review about it are two different systems.
Public feedback is usually owned by marketing, private conversation by operations, and neither number moves the other. So the same failure gets discovered twice, weeks apart, and gets fixed once.
SOCIAL LISTENING AS SOLD
- A rating average with no idea which operational cause moved it.
- Sentiment scored by keyword, so sarcasm and cancellations read the same.
- Fake reviews counted as customer opinion because nothing checks them.
- A monthly report that arrives after the spike it describes.
- No path from a public complaint to the call it came from.
SMI AS A MODULE
- Reviews classified by the same driver taxonomy as your calls.
- Sentiment read from the text with the passage attached, like every other module.
- Authenticity scored per review, from text, behaviour and network signal.
- A daily pull, with anomaly alerts on the day the pattern breaks.
- Public and private signal on one axis, correlated and measurable.
SMI is the eighth L1 module, not a separate product. It writes to the same log, signs its results the same way, and its output is available to every layer above it.
[02 / SOURCES]
Eighteen platforms, pulled daily, kept for two years.
Review sites, app stores, employer sites and the social platforms where a complaint travels fastest. Which ones matter depends on your market, so coverage is configured rather than assumed.
SOFTWARE
G2 · Capterra · TrustRadius
Longer, more specific and written by buyers, which makes them the best source for claim contradictions.
APP
App Store · Play Store · ProductHunt
Release-linked feedback. A version number turns a rating drop into a specific thing to fix.
EMPLOYER
Kununu · Glassdoor · Indeed
The operation seen from inside. Attrition and service quality move together more often than anyone likes.
SOCIAL
X · Reddit · LinkedIn
Where an incident becomes public. Velocity here is the earliest warning the platform can give you.
TRAVEL
Tripadvisor · Facebook · Instagram
Location-linked feedback, which maps cleanly onto a site or a team rather than onto the brand.
CADENCE
Daily pull, weekly re-score
Webhooks where a platform offers them, a daily pull where it does not. Authenticity is recalculated weekly against model drift.
RETENTION
24 months raw, 36 aggregated
Raw reviews are held anonymised for two years under GDPR terms. Aggregates run three, so a year-on-year comparison is possible.
[03 / AUTHENTICITY]
A review nobody wrote should not move your decisions.
Every review carries an authenticity score between 0 and 1, built from three independent signals. Below 0.2 it is treated as likely fake and pulled out of the averages instead of quietly weighting them.
01 · TEXT
How the review is written.
Specificity, internal consistency, whether it names things a real customer would name, and whether the same phrasing appears across accounts that should not know each other.
02 · BEHAVIOUR
How the account behaves.
Posting history, timing, rating distribution and the gap between account age and first review. Patterns, not identities: the account is a signal, never a person we profile.
03 · NETWORK
What happened around it.
Clusters of reviews arriving together, coordinated wording, and bursts that do not match any event in your own operational data.
A fake-review detection fires as an event, so a suspected cluster reaches a human for review rather than being deleted by a model. Constitution A5 applies here as everywhere: the platform recommends, a person decides.
[04 / DASHBOARD]
Eighteen indicators, each with a target and a drill-through.
Rating, sentiment split, authenticity, source coverage, velocity, an NPS proxy, the response gap against your own knowledge base, and the correlation between public sentiment and internal satisfaction.
Those two views are the point of the module. On the left is what the public says; on the right is what your own conversations already showed. The correlation between them is an indicator in its own right, and where it is weak, one of the two is not telling you the truth.
Two more of the eighteen end up carrying most of the weight. Response gap counts the topics customers raise in public that your knowledge base has no answer for. Time to detect measures the days between a signal appearing socially and the same thing moving an internal KPI, which is the number that says whether listening is worth anything.
[05 / NUMBERS]
Two ways to run it.
A one-time audit answers whether a brand has a problem it cannot see. Continuous monitoring answers whether the fix worked, every day, for as long as it matters.
18
PLATFORMS
Review, software, app, employer, social and travel sources in one pipeline.
18
INDICATORS
Each with a target, a threshold and a path back to the reviews behind it.
200 → 500
REVIEWS PER AUDIT
The one-time read, across every source that carries the brand. Under 24 hours to process.
Flat pilot
MONTHLY · BEFORE ANY LICENCE
Continuous monitoring as a flat monthly pilot, before any per-agent licence. Priced on the call.
As a module inside the platform, SMI is included from Professional upward on the standard per-agent models, cloud and Local alike. The model table lives on the platform page · pricing happens on a call with your own numbers.
[06 / CONSOLE]
Open the console.Read your own reviews back.
The fastest version of this conversation is an audit of your brand. Send us the name and we will read what is already public about it.
quality visible.