Turning 45 million readership events into a handful of sales signals
Objective
Turn BlueMatrix’s readership data into an AI product for its sell-side clients, one that detects meaningful shifts in buy-side demand and explains, what each shift means and what to do next.
Obstacle
Every month, 45 million readership observations flow through BlueMatrix. Hidden in that volume are early signals of shifting buy-side intent.
Outcome
An anomaly detection layer screens 1.5 million account-topic relationships to find the moves that matter. An LLM layer then turns each one into a recommendation.
Background
BlueMatrix provides the research distribution infrastructure that sell-side firms rely on to publish research and track how their buy-side clients engage with it. Every shift in what the buy side reads more or less of is not just an engagement metric. It is an early signal about where interest, and money, is moving next.
The opportunity was to turn that signal into an AI product BlueMatrix could offer its sell-side clients: not another dashboard, but decision support that surfaces who to contact and why.
Challenges
Every Client Reads Differently: Some buy-side accounts consume research constantly, others dip in occasionally, so the same jump in readership can mean opposite things depending on who it comes from. Detecting a real change meant learning each account’s own pattern of normal behaviour at a topic level across thousands of accounts, so the system flags a genuine shift rather than routine activity, and does not drown the desk in noise.
Which Signals Lead To Sales: Whether a spike in readership ever turns into a sale or a meeting is not something BlueMatrix’s data can see; that happens downstream, off-platform. That meant the modelling had to focus on something the data could establish confidently: what behaviour should normally be expected for each client, and when what actually happened was meaningfully different. Every alert also had to remain simple enough for a person to understand exactly why it fired and trust it on its own terms.
Solution
Layer One, Anomaly Detection: We model normal reading behaviour for each account and topic, then flag the deviations that matter: spikes, drop-offs, and gaps where expected coverage is missing. Each flag weighs the size of the change and how unusual it is for that account, so a genuine shift is separated from routine variation, rather than judged against one threshold for everyone.
Layer 2: Interpretation and Action: The detected signals are scored, ranked, and de-duplicated so no account is flagged twice for the same shift. An LLM then turns each one into a plain-language signal that says what changed, for which account, and what it means, mapped to a recommended action: a drop-off means re-engage, a spike means follow up, widening readership can signal a growing buying group. The desk receives a ranked list of signals with a next step attached, not raw data to interpret.
Impact
A New Product Line, Not Just A Better Report: BlueMatrix was sitting on valuable buy-side engagement data it could not package for sale. The AI product turns it into something sell-side clients will pay for: signals that tell their desks which account to contact and why, before the change is buried in a monthly report.
Every Signal Is A Sales Lead, Not A Statistic: A shift in buy-side reading is one of the earliest indicators of where demand is heading. Because each signal arrives with a recommended action, sell-side teams can be first to the conversation, reaching the account as interest builds rather than chasing it once competitors already have.
Trust That Protects The Value, Long Term: None of this works if desks stop opening the signals. By learning each account’s normal behaviour and explaining every signal clearly, the system surfaces what matters without the noise, earning a place in the daily workflow rather than becoming another dashboard people tune out. And as outcomes are captured and fed back into the models, the system learns which signals convert, sharpening its judgement over time.
Creating Value For BlueMatrix...
45 million reading events screened every month,
1.5 million account-topic combinations modelled continuously,
A handful of ranked, actionable, plain-language signals delivered every day.
Get In Touch
Our friendly team are always on hand to answer questions, troubleshoot problems and point you in the right direction.
