The banking & financial services value tree
In banking the tree runs on spread and risk rather than units. Revenue is what you earn on balances and fees; cost is what it takes to serve, control and prove it; and the cash branch is capital, which is scarcer than either.
01 The same tree, this industry
Where the money is made and lost here
The structure does not change: value is profit plus how well that profit becomes cash, profit is revenue minus cost, and revenue is price times quantity. What changes is which drivers sit underneath each branch, and which system holds them. If you have not read the general version, start with the Enterprise Value Tree and come back.
Two things make this tree different. Revenue is a rate applied to a balance, so price and quantity are the same transaction seen twice. And the cost branch contains a line no other sector carries — the cost of proving to a regulator that the other branches are true.
Every driver below names the metric it lands on. Follow one and you get its formula, the system the number lives in, and the ways it is commonly misread.
02 Price
What decides the price you actually get
Spread on lending
The rate you charge against your cost of funds. Small, and multiplied by a very large balance, which is why basis points are a governance question.
Price realisation net_revenue ÷ list_revenuePriceDeposit pricing and beta
How much of a rate move you pass to depositors, and how quickly. The single largest swing factor in a rate cycle.
Average net price net_revenue ÷ unitsPriceFee schedule and waivers
The published schedule is not the realised one. Waivers are granted at the front line and accumulate where nobody reports them.
Price realisation net_revenue ÷ list_revenuePriceRisk-based pricing
Whether the rate reflects the risk taken. Under-pricing risk grows the book and shows up two years later as provisions.
Contribution margin (net_revenue − variable_cost) ÷ net_revenuePriceFX and interchange margin
Per-transaction spread that is invisible individually and material in aggregate.
Average net price net_revenue ÷ unitsPriceRelationship and bundle pricing
A discount given for a relationship that is never measured at relationship level.
Share of wallet your_revenue_from_the_account ÷ the_account_total_category_spend03 Quantity
What decides how much you sell
Balances, not customers
The volume driver is money held, so growth is balance growth — and a customer count that rises while balances fall is not growth.
Active customers count(distinct customers with at least one purchase in the window)VolumeNew-to-bank acquisition
What it costs to bring a relationship in, against what that relationship is worth over its life.
Customer acquisition cost (sales_cost + marketing_cost) ÷ new_customers_wonVolumeAttrition and dormancy
Accounts rarely close; they go quiet. Dormancy is churn that has not filed the paperwork.
Customer retention rate customers_retained ÷ customers_at_startVolumeProducts per relationship
The cheapest growth available, and the hardest to measure without one resolved customer record.
Attach rate orders_containing_the_attached_item ÷ eligible_ordersVolumeShare of wallet
Retention says they stayed. Share says whether their main banking relationship is with you or with someone else.
Share of wallet your_revenue_from_the_account ÷ the_account_total_category_spendVolumeApproval and conversion rates
Applications that fail at credit or KYC are demand you paid to create and then declined.
Win rate opportunities_won ÷ opportunities_closed04 Cost
What it takes to operate
Cost to income
The sector's headline efficiency measure, and the one that hides which channel or product is actually expensive.
Operating margin operating_profit ÷ net_revenueCostCost to serve by channel
A branch interaction, a call and an app session differ by an order of magnitude, and the mix moves faster than the cost base.
Cost to serve (delivery + service + returns + order_handling) ÷ customers or ordersCostCredit losses and provisions
The cost of risk taken earlier, recognised now. Expected-loss models make this a data quality question as much as a credit one.
Contribution margin (net_revenue − variable_cost) ÷ net_revenueCostRegulatory and reporting effort
The reports, the reconciliations and the people who produce them — a cost line that exists because the data cannot answer for itself.
Time to insight elapsed time from a business question being asked to a trusted answerCostTechnology and core platform
Run cost plus the change portfolio, in a sector where the core system is usually older than the people maintaining it.
Capex intensity capital_expenditure ÷ net_revenueCostFraud and operational losses
Detected late, absorbed centrally, and rarely attributed to the product or channel that generated it.
Data quality score weighted pass rate across completeness, validity, uniqueness, timeliness and consistency rules05 Cash
Where the cash actually sits
Regulatory capital
The genuinely scarce resource. Every asset consumes it, and return has to be measured against what it consumed.
Return on capital employed EBIT ÷ (total_assets − current_liabilities)CashFunding mix and cost
Deposits against wholesale funding: the stability and price of the money you lend.
Net debt to EBITDA (total_debt − cash_and_equivalents) ÷ EBITDACashLiquidity buffers
Assets held because a regulator requires them, earning less than the alternative.
Free cash flow operating_cash_flow − capital_expenditureCashProvision timing
When a loss is recognised is a modelling and data decision before it is an accounting one.
Forecast accuracy 1 − ( Σ|actual − forecast| ÷ Σactual )CashRisk-weighted asset density
The same book can consume very different capital depending on how well it is classified and evidenced.
Return on capital employed EBIT ÷ (total_assets − current_liabilities)CashBudget against plan
In a cost-to-income culture, variance discipline is the operational form of the capital constraint.
Budget variance (actual − budget) ÷ budget06 Where the numbers live
The systems behind the branches
A value tree is only usable once each box maps to a system and a record. These are the six that matter most in banking & financial services: what each one is actually for, the records inside it the tree depends on, and which branch it feeds.
| System | What it holds | Key records | Feeds |
|---|---|---|---|
| Core banking | Accounts, balances and every posting against them — the ledger the rest of the bank reconciles to. | Account, Customer, Posting, Product, Balance | Price, quantity, cash |
| Loan origination | The application journey: what was applied for, what was offered, what was declined and why. | Application, Decision, Offer, Collateral | Quantity, price |
| Payments and cards switch | Transaction-level flow, interchange and the fee events that never reach a statement line. | Transaction, Scheme, Merchant, Interchange record | Price, cost |
| CRM | The relationship view: households, groups, interactions and the waivers granted inside them. | Customer, Household, Interaction, Waiver | Price, quantity |
| Risk and finance (GL, IFRS 9) | Exposure, staging, expected credit loss and the general ledger those feed. | Exposure, Stage, ECL model run, GL account | Cost, cash |
| Regulatory reporting | The submissions themselves, and the lineage that has to survive a regulator asking where a figure came from. | Return, Data point, Lineage node, Attestation | Cost, cash |
Almost every hard question in banking & financial services needs two of these joined. That join — not the calculation — is the work.
07 Where it leaks
Value lost between two systems
These are the losses that no single system can see, because the evidence is split across two of them. Each one is a real number that stays invisible until the join exists — which is why the tree is an integration exercise before it is an analysis one.
| Where value leaks | Why it happens | The join that finds it |
|---|---|---|
| One customer, several identities | The same person is a current account in core, an applicant in origination and a household in CRM. Every relationship metric divides by a number nobody agrees on. | Resolve customer and household to one golden record across all three |
| Risk and finance disagree about the same exposure | Risk stages an exposure on one model run; finance books it on another extract. Both are defensible and they do not reconcile. | Join exposure to GL posting through one governed definition and one run identifier |
| Waivers granted locally, margin lost centrally | Fee income is budgeted from the published schedule while the front line waives against relationship arguments no system records. | Join the waiver event in CRM to the fee line in core banking |
| The regulatory return is built from a copy | A submission assembled in spreadsheets from extracts is provable only by the person who built it, which is the exposure the regulator is actually testing. | Lineage from the reported data point back to the source posting |
| Acquisition cost by channel, value by product | Marketing measures cost per acquisition; the P&L measures product margin. Neither can say which channel brings profitable relationships. | Join campaign and channel attribution to the resolved customer's lifetime contribution |
08 Worked example
A growing book that earned less
Balances grew 7% and net interest income fell. The tree separates the causes: deposit beta ran ahead of asset repricing, a retention campaign discounted the renewal book, and provisions rose on a segment that was priced before its risk was understood. Three different owners, one number, and no report that showed them together.
| Component | Effect | What sits behind it |
|---|---|---|
| Balance growth | +€12.4m | 7% growth at the prior spread |
| Deposit repricing | −€9.8m | Beta passed through faster than assets repriced |
| Retention discounting | −€3.1m | Renewal rates below book average |
| Fee waivers | −€1.4m | Granted at the front line, not in plan |
| Cost of risk | −€2.2m | Provisions on the segment priced two years ago |
| Net | −€4.1m | A larger book, earning less than the smaller one |
Illustrative figures, shown to demonstrate the split. The point is the shape of the walk, not the numbers — on your own data the same bridge is built from your ledger.
09 Diagnostics
Six questions to ask in banking & financial services
Ask them of your own team before anyone asks them of you. In most organisations at least two of these cannot be answered without a manual exercise, and those two are the plan.
- Can you produce the full relationship view of your largest twenty customers, including every product and entity, without a manual exercise?
- When risk and finance report the same portfolio, do the numbers tie — and is the reconciliation automated or a person?
- What proportion of fee income was waived last quarter, and who granted it?
- Can a reported regulatory figure be traced to source postings without the analyst who built the return?
- Do you price for risk at origination using the same data you provision with later?
- Which channel produces relationships that are still profitable in year three?
10 The metrics behind it
Definitions for every box
11 Questions
Frequently asked
Why does entity resolution matter more in banking than elsewhere?
Because almost every regulatory obligation and every relationship metric is defined per customer or per group. If the same party exists three times, exposure is understated, share of wallet is meaningless, and a single-customer view is impossible to attest to.
Is the cash branch really capital?
For a bank, yes. Cash in the corporate sense is inventory of the trade; the scarce resource that constrains growth is regulatory capital, so return on capital employed is the honest top of the tree.
How does this connect to the regulatory agenda?
A regulator asks the same question the tree asks: where did this number come from. Lineage coverage and governed metric coverage are the two metrics that decide whether either question can be answered quickly.
Where do most banks start?
With the customer golden record, because every other branch of the tree divides by it. The second domain is usually the reconciliation between risk and finance, which is where the largest manual effort sits.
See this tree on your own data
Connect the systems above, define each box once, and the tree stops being a slide.