Methodology
Not model assumptions. Not cohort averages. The customer base as it actually is.
Observed behaviour, not assumed models
Keystone IQ's methodology is built on a single principle: customer behaviour is observable, measurable, and interpretable directly from transaction data — without imposing probability distributions, survey-based proxies, or cohort averages that flatten the signal.
Five dimensions. No averages.
The dimensions are not arbitrary measurement variables. Each one captures a distinct building block of how customer value and engagement are actually constructed.
| Dimension | What it measures | Why it matters in diligence |
|---|---|---|
| Frequency | How often a customer transacts within the observation period | Distinguishes habitual buyers from occasional ones. Frequency collapse is often the earliest indicator of customer base erosion. |
| Total spend | Cumulative revenue contribution per customer | Shows where the revenue is actually concentrated. The 21/66 pattern (21% of customers, 66% of revenue) is common and rarely visible in headline metrics. |
| Breadth of purchasing | Number of distinct product categories or services accessed | Breadth is a strong predictor of retention. Multi-category customers churn at lower rates and respond less to single-category pricing pressure. |
| Recency | Time elapsed since most recent transaction | The leading indicator of likely departure. Recency deterioration in high-value segments precedes revenue decline by 6 to 12 months. |
| Order value | Average transaction value over the observation period | Shows whether customers are growing their spend, maintaining it, or quietly trading down. |
From five dimensions to cohorts
Each customer's five-dimension score is their behavioural fingerprint. Cohorts are formed by finding customers whose fingerprints cluster naturally together. This is not assignment. It's discovery.
From cohorts to movement
The same process is run at each point in time across 12 to 48 months of transaction history. The movement of customers between cohorts period on period — who stayed, who improved, who deteriorated — is the transition matrix. That movement is where the methodology produces its most commercially significant output.
Headline revenue can be stable or growing while the cohort composition underneath is shifting in ways that will matter for the investment thesis.
What averages hide, distributions expose
A retention rate tells you how many customers stayed. It does not tell you which ones, what they are worth, or where they are heading.
Take a 70% retention rate. Inside it:
- 25% increased their frequency and spend — strengthening
- 30% maintained similar behaviour — stable
- 15% reduced frequency but maintained spend — early migration signal
- The 30% who churned were disproportionately the highest-frequency buyers
The customers you lost were worth more than the customers you kept. That is invisible in the retention rate. It is visible in the cohort movement data underneath it.
Methodology FAQs
Probability models assume a distribution shape before fitting the data. If the real customer base matches that assumed shape, the model works well. If it doesn't — and in our experience, most don't — the model smooths over exactly the compositional movement that matters to an investment thesis. We prefer to observe the actual distribution and track how it moves over time.
Individual-level transaction data covering 12 to 48 months. This typically comes from CRM systems, ecommerce platforms, billing systems, or point-of-sale data. Data is totally anonymised — we need transaction patterns, not customer identities.
We assess data quality at the scoping stage, before any fees are committed. If the data is insufficient to support a credible analysis, we say so. If it is messy but recoverable, we resolve quality issues during the ingestion phase rather than passing uncertainty through to the output.
The analytical framework — distributional segmentation across five behavioural metrics with operator interpretation — is Keystone IQ's proprietary methodology. We publish this summary because transparency builds trust. The competitive advantage is not in the framework description; it is in the twenty years of operator experience that interpret the output.
Theta fits a statistical model to transaction data and projects forward from a small number of parameters. It treats the customer base as one population. Keystone IQ identifies the natural clusters that already exist in the data. Segments emerge from observed behaviour, not from model assumptions.