KeelShift turns acquisition, billing and behavioural data into commercial decisions: where to spend, what a customer is worth, when they pay back, and where revenue is leaking.
We are not trying to replace commercial judgment. We give it better inputs: one reconciled set of definitions, evidence with the reasoning attached, and an honest account of confidence.
Who We Are
A small team with senior, hands-on experience across telecom, value-added services and iGaming, not a generalist analytics shop.
Between us: years spent building and scaling CRM, churn and lifetime-value models inside a major telecom’s value-added services business, and leading data science functions from ground-up builds to production features running at scale today.
We have sat on both sides of the table: the commercial team asking for a number it can trust, and the team responsible for producing one honestly.
We use AI in our own work every day, and it makes us faster. It is not what makes the answer right. Knowing what a number means in a carrier-billed business, and what it cannot mean, is.
Who We Serve
Plenty of vendors will tell you they serve any business with a subscription. We would rather be useful to a narrower set whose problems we have already lived with.
Mobile VAS
A subscription is a few euros a month. The economics only work in aggregate, and only if the aggregate is built correctly.
Carrier billing
The operator, the aggregator and the merchant each hold a different version of the truth.
Digital content
Engagement decays long before the billing stops. By the time revenue moves, the decision was needed months ago.
Subscription content
Retention and acquisition compete for the same euro, usually with no shared number to settle it.
Telecommunications
The same recurring-revenue mechanics, at a scale where one point of churn is real money.
iGaming
Value sits in very few players and deposit behaviour moves fast, so averages actively mislead.
Why We Picked This Lane
Because we know where the bodies are buried. Failed charges that look like churn. Aggregator reports that disagree with settlement files. Cohorts attributed to a campaign that did not acquire them. An operator-side change that moves the whole base and looks like a behavioural signal to anyone who was not there.
The standard subscription-analytics playbook assumes a customer you can identify, contact and interview. Carrier-billed customers offer none of those, and the data has pathologies of its own.
Billing is attempted, not guaranteed
A failed charge is not a cancellation. Treat them alike and every churn number downstream is wrong.
Acquisition passes through intermediaries
The conversion you paid for and the subscription that started are not automatically the same record.
Settlement arrives late, and revised
Revenue recognised this month may belong to a cohort you acquired two months ago.
Refunds land after the fact
Revenue counted in month one can disappear in month three, long after the campaign was judged.
Identity is thin
No email, no login. Often nothing but a hashed MSISDN and a timestamp.
The base can move at once
An operator-side change looks exactly like a behavioural signal to anyone who was not there.
How The Work Runs
We start with the decision, not the technique. Every step below exists to make one commercial call safer, and to hold up when someone senior pushes back on it.
Frame the decision Start from the commercial call the work has to serve: a budget reallocation, a scaling decision, a retention investment. Everything after it is judged on whether it moves that call.
Reconcile the sources Put acquisition, billing and behavioural data onto one definition of a customer, and report where they disagree instead of quietly picking one. This is the step we call Flybridge, and it is where most of the unglamorous work happens. If the data cannot carry the decision, you hear that here rather than at the end.
Measure, then project Compare like-for-like cohorts to establish what actually happened, and where a forward-looking answer is needed, base it on what customers have already done. What comes out is a ranked list of causes, not one number you are asked to believe.
Test it, then recommend Check the findings against customers and periods deliberately held out of the work, say where the evidence is thin, and connect what survives to the decision from step one.
Decisions, Not Dashboards
Most revenue analysis stops at a figure: a blended CAC, a churn rate, a lifetime value quoted to the last euro with nothing said about how confident anyone is in it. Nobody can act on a number they cannot interrogate.
So every figure we hand over arrives with the reasons behind it ranked and evidenced. When the aggregator report and the settlement file disagree, the discrepancy is quantified and shown rather than resolved quietly in favour of the more convenient number.
We Work With What You Already Have
The work runs on one extract of the acquisition, billing and behavioural data you already keep for other reasons. No integration project, no access to production systems, no change to your stack, and nothing new to start collecting.
It does not replace your BI stack or your CRM. Those systems report what happened and act on it. This sits above them: one reconciled definition across sources that were never built to agree, and the decisions it makes possible. The output is a document and a readout, and the segments go back into the systems your team already runs campaigns from.
Measured, Then Checked
The work is meant to produce a commercial outcome you can point at afterwards: budget moved off a channel that was not paying back, retention spend aimed at the cohorts that carry the revenue, offers going to the fifth of the base that converts.
So the findings are tested rather than asserted: results are checked against customers and periods held out of the work, chosen before the work starts rather than after the answer is known. Where a question cannot be settled by the evidence you have, we say so instead of borrowing confidence from the findings that can.
Nothing is quietly dropped
Records that fail our checks are reported, not discarded to make the analysis tidier.
Nothing is graded on its own homework
Findings are tested against customers and periods held back from the work, so a result that only holds where it was built does not reach you.
Nothing is smoothed over
Where two of your systems disagree, the size of the gap is reported rather than resolved in favour of whichever number reads better.
The techniques are standard and available to anyone. Choosing the right definitions, reconciling sources that were never built to agree, and knowing when the data will not carry the conclusion: that is what decides whether any of it is worth acting on.
Work With KeelShift
We work with recurring-revenue businesses that want this run on their own data. Most start with the Revenue Diagnostic: fixed scope, ten business days, and a clear account of what your data can and cannot carry.
