IEI logoIntelligence EconomyInstitute

Tokenization Economics Lab

Make the economics of tokenization defensible.

The technical question is largely answered. The economic one is not: where value is created, what it costs to get there, how much of it the ledger is actually responsible for, and which participant ends up keeping it.

The Lab is a shared vocabulary for those questions and a disciplined way to build one institution’s case with it — against a real baseline, with the evidence attached, and with a result a board can interrogate line by line rather than take on trust.

The method

Four columns, one boundary, and the honest question in the middle.

A case holds the same initiative four ways. The columns share one system boundary and one cast of participants, so a benefit cannot appear because somebody moved the line.

How it works today

The current process, measured or reconstructed from actuals. The starting point, and deliberately not the benchmark.

Best conventional alternative

What the same outcome would cost with an API, a workflow redesign and better data. This is the column that decides whether tokenization earned its place, and it is the one almost every business case leaves out.

Tokenized, as planned

The design at its intended operating point. A plan, and labelled as one — a target cannot carry a realised claim however confident it is.

Tokenized, as measured

What it actually does once it runs. The only column that can support a realised number, and the one that usually arrives smaller than the plan.

Discipline

The useful part is what it will not let you say.

A fee is not value created

Every economic flow has a payer and a recipient. A fee between two parties inside the boundary cancels — one party's revenue is the other's cost, and the system as a whole is no better off. The consolidated result is calculated twice by independent routes, and the two must agree before anything is shown.

A benefit with no bearer cannot be entered

Not validated afterwards — inexpressible. There are exactly two things you can record: a transfer, which always has both ends, and a resource effect, which always has exactly one bearer. There is no third shape for a benefit that nobody pays for.

You cannot promote your own baseline

The baseline class is assigned from what the case can actually evidence: measured side by side, your own history adjusted, a validated process model, or a general estimate. It is not a setting, and a class D baseline never reaches a peer comparison at any sample size.

Faster settlement costs you netting

Released prefunding and destroyed netting happen together, so they are one calculation. A liquidity benefit reported without the netting it gave up is half an answer, and the engine says so on the figure rather than in a footnote.

One exposure cannot fund three benefits

The same reduction in settlement exposure can plausibly be told as a capital benefit, a liquidity benefit and a fall in expected loss. Each is defensible alone; together they triple-count one fact. The case has to declare whether they share an underlying exposure.

Weak evidence does not move the number

It changes what the number may be used for. Pulling an uncertain figure toward a neutral middle would make a bad case look average, which is the opposite of the point. Quality sits beside the economics and governs use, never value.

Attribution

How much of it was the ledger?

Almost every tokenization programme also digitises something, redesigns a workflow, standardises a data format and changes its volume mix — all at once. Each of those would have produced some of the same improvement on its own.

So the observed difference is allocated across eight mutually exclusive drivers that must sum to a hundred, and the result is reported three ways: everything that changed, what the transformation caused by any route, and what needed the ledger. The third is nearly always the smallest, and publishing it honestly is the most useful thing this exercise does. Anything nobody can allocate is carried as unallocated rather than scaled away — a case that has attributed sixty percent should look like one.

The eight drivers

  • The ledger itself

    Effects that need a shared ledger or programmability and could not be obtained any other way.

  • Ordinary digitisation

    Replacing paper and manual keying. Real, valuable, and available without any of this.

  • Redesigning the process

    Removing steps and approvals. Often the largest single driver, and frequently credited to technology that merely provided the occasion.

  • Standardising data

    A common format or message. Most of the interoperability benefit usually lives here rather than in the ledger.

  • A change in the rules

    A regulation moved, and would have had this effect regardless of the platform.

  • Volume, product or client mix

    You are doing more, or a different mix. Unit economics move without anything improving.

  • Rates and the wider economy

    Liquidity and funding benefits are especially sensitive to this.

  • Something else, named

    Naming it is the point. 'Other' with no explanation is not an allocation.

The dictionary

14 domains, 47 metrics, 23 of them required.

Open in full, free, no account.

Anything expressed per unit of something carries its denominator, because “cost per transaction” is not a number until it says whether it counts transactions initiated, accepted or actually settled. Two institutions using different answers cannot be compared, and usually do not notice.

47 of 47

  • Completed transactions

    TE.VOL.TXN.COUNT

    Transactions that finished successfully in the period, on the stated denominator basis.

    The denominator for most unit economics. Whether it counts what was attempted or what actually settled changes every per-unit figure downstream.

    RequiredDepends on the caseper transaction settled

  • Transaction notional

    TE.VOL.NOTIONAL

    Total value transacted in the period.

    Scales fee income and settlement exposure. Rising notional is not adoption if it rose because prices did.

    RequiredDepends on the case

  • Assets outstanding

    TE.VOL.ASSETS_OUTSTANDING

    Value held on the platform at the measurement date.

    The base most servicing and custody fees are charged on.

    RequiredHigher is better

  • Active clients

    TE.VOL.ACTIVE_CLIENTS

    Clients meeting a stated minimum activity in the period. The threshold is part of the number, not a note beside it.

    Onboarded is not active. Most disappointing adoption figures are the gap between the two, and the gap is the finding.

    RequiredHigher is betterper active client

  • Repeat transaction rate

    TE.VOL.REPEAT_RATE

    Share of active clients who transacted more than once in the period.

    The cheapest test of whether the thing works. One transaction can be a favour; the second is a decision.

    Higher is better

  • Secondary turnover

    TE.VOL.SECONDARY_TURNOVER

    Secondary volume as a share of assets outstanding, per period.

    Where the liquidity claim is tested. Primary issuance efficiency implies nothing at all about this.

    Higher is better

  • Incremental revenue

    TE.REV.TOTAL_INCREMENTAL

    Revenue that would not have been earned without this initiative.

    The only revenue that belongs in the business case. Revenue that moved across from the legacy product belongs in the line below.

    RequiredHigher is better

  • Cannibalised revenue

    TE.REV.CANNIBALIZED

    Existing revenue lost or displaced because this initiative exists.

    Subtracted, always. A case that reports incremental revenue without it is reporting gross where it means net.

    RequiredLower is better

  • Fee yield

    TE.REV.FEE_YIELD

    Fees earned divided by the relevant notional or assets outstanding.

    Makes fee levels comparable across very different sizes. Retrospective only, and never shared identifiably.

    Calculated, never typedDepends on the caseper million of notionalper million of AUM

    TE.REV.TOTAL_INCREMENTAL / TE.VOL.NOTIONAL * 10000

  • Contribution margin

    TE.REV.CONTRIBUTION_MARGIN

    Net revenue less the variable cost of producing it.

    The numerator of breakeven. Without it, the volume threshold cannot be calculated at all.

    RequiredCalculated, never typedHigher is better

    (TE.REV.TOTAL_INCREMENTAL - TE.REV.CANNIBALIZED) - TE.COST.RECURRING.TOTAL

  • One-off implementation cost

    TE.COST.FIXED.INITIAL

    Everything spent to reach production: build, integration, legal, licensing, assurance, change.

    The number payback is measured against, and the one most often quoted before integration is finished.

    RequiredLower is betterOne-off, not recurring

  • Legacy coexistence cost

    TE.COST.FIXED.DUAL_RUN

    The cost of running the old process alongside the new one, per period, until the old one is switched off.

    Frequently the largest number in the model and the one most often missing. A saving that depends on decommissioning is not a saving until decommissioning is funded and scheduled.

    RequiredLower is better

  • Recurring operating cost

    TE.COST.RECURRING.TOTAL

    Everything it costs to keep running for a period: people, technology, vendors, network, compliance, support.

    The run rate. Pilots understate it whenever exceptional manual support is doing part of the work.

    RequiredLower is better

  • Cost per completed transaction

    TE.COST.UNIT.TRANSACTION

    Recurring operating cost divided by transactions on the declared denominator basis.

    Meaningless without the denominator and without utilisation beside it: a low unit cost at 10 percent of capacity is an artefact of the divisor.

    RequiredCalculated, never typedLower is betterper transaction settledper transaction acceptedper transaction initiated

    TE.COST.RECURRING.TOTAL / TE.VOL.TXN.COUNT

  • Cost per active client

    TE.COST.UNIT.CLIENT

    Recurring operating cost divided by active clients on the declared activity threshold.

    The right denominator when the business is a relationship rather than a flow.

    Calculated, never typedLower is betterper active client

    TE.COST.RECURRING.TOTAL / TE.VOL.ACTIVE_CLIENTS

  • End-to-end cycle time

    TE.OPS.END_TO_END_TIME

    Elapsed time from trigger to completion, across the whole in-scope process.

    Only becomes economic once it changes funding, exposure or headcount. Speed on its own is a fact, not a benefit.

    RequiredLower is better

  • Straight-through-processing rate

    TE.OPS.STP_RATE

    Share of items completing with no manual intervention.

    The cleanest predictor of unit cost, and the number a well-automated legacy process already scores well on.

    RequiredHigher is better

  • Manual touches

    TE.OPS.MANUAL_TOUCHES

    Human interventions per transaction.

    Where pilot economics quietly diverge from production economics.

    Lower is betterper transaction settled

  • Exception rate

    TE.OPS.EXCEPTION_RATE

    Share of items requiring investigation or repair.

    Exceptions cost far more per item than the happy path, and they rarely disappear on a new rail.

    RequiredLower is better

  • Failed settlement or process rate

    TE.OPS.FAIL_RATE

    Share of items that did not complete.

    Drives both cost and counterparty exposure, and is the number counterparties ask for first.

    Lower is better

  • Mean recovery time

    TE.OPS.RECOVERY_TIME

    Average time to restore normal service after a failure.

    What a resilience claim means in practice, and what a client actually experiences.

    Lower is better

  • Required prefunding

    TE.LIQ.PREFUNDING

    Cash that must be in place before the process can run.

    The headline liquidity benefit — and the one that must always be reported alongside the netting it gave up.

    RequiredLower is better

  • Peak intraday liquidity

    TE.LIQ.INTRADAY_PEAK

    The largest funded balance required at any point in the day.

    What the treasury actually has to provide for. Averages hide the constraint that binds.

    Lower is better

  • Average idle balance

    TE.LIQ.IDLE_BALANCE

    Cash sitting available but unused across the period.

    Carries an opportunity cost every day it sits there.

    Lower is better

  • Liquidity funding cost

    TE.LIQ.FUNDING_COST

    The cost of funding the balances the in-scope process requires, over the period.

    Converts a balance into a number the business case can use. Rate-sensitive, so the basis must be stated.

    RequiredLower is better

  • Value of netting gained or lost

    TE.LIQ.NETTING_CHANGE

    The change in netting benefit, signed: positive when netting improves, negative when it is given up.

    Gross settlement releases prefunding and destroys netting. A liquidity benefit reported without this line is half an answer, and the engine treats it as one.

    RequiredHigher is better

  • Risk-weighted-asset change

    TE.CAP.RWA_CHANGE

    Change in risk-weighted assets attributable to the in-scope process.

    Real, and dependent on a supervisory view somebody has to actually hold. Store it as modelled until they do.

    Lower is better

  • Capital charge

    TE.CAP.CAPITAL_CHARGE

    Cost of the capital the in-scope process consumes, per period.

    Must not double count against liquidity or expected loss — the engine checks the overlap explicitly.

    RequiredLower is better

  • Collateral posted

    TE.COL.POSTED

    Value of collateral posted against in-scope obligations.

    Encumbered assets are unavailable for anything else, which is a cost even when it is not a charge.

    Depends on the case

  • Effective haircut

    TE.COL.HAIRCUT

    Average discount applied to posted collateral.

    Decides how much has to be posted for a given exposure.

    Lower is better

  • Time to mobilise collateral

    TE.COL.MOBILIZATION_TIME

    Elapsed time from decision to collateral being usable where it is needed.

    One of the few places a shared ledger can pay for itself outright, because the delay is the entire product.

    Lower is better

  • Eligible collateral reuse rate

    TE.COL.REUSE_RATE

    Share of eligible collateral reused within permitted limits.

    Higher is better only inside the controls. Outside them it is leverage wearing an efficiency label.

    Higher is better

  • Expected loss

    TE.RISK.EXPECTED_LOSS

    Probability-weighted loss from in-scope operational, settlement and technology risk, per period.

    Enters the risk-adjusted result exactly once. Where there is no loss history, use a transparent range and accept the lower evidence grade rather than inventing a point.

    RequiredLower is better

  • Service downtime

    TE.RISK.DOWNTIME

    Time the service was unavailable during the period.

    A 24/7 promise is a 24/7 cost. This is where the difference shows.

    Lower is better

  • Key-control coverage

    TE.RISK.CONTROL_COVERAGE

    Share of identified key controls that are implemented and tested.

    Risk that moved to a bridge, an oracle or a key is not risk removed. This is where the relocation becomes visible.

    Higher is better

  • Client onboarding time

    TE.CLIENT.ONBOARD_TIME

    Elapsed time from first contact to a client able to transact.

    The friction that decides how many of the eligible ever become active.

    Lower is better

  • Minimum economical ticket

    TE.CLIENT.MIN_TICKET

    The smallest transaction that is worth processing.

    Where an access claim is settled. If this does not fall, nothing was opened to anybody new.

    Lower is better

  • Service availability window

    TE.CLIENT.AVAILABILITY

    Hours per week the service can actually be used.

    Extending it is a genuine client benefit and a genuine staffing, liquidity and incident-response cost.

    Higher is better

  • Bid-ask spread

    TE.MKT.SPREAD

    Average quoted spread on the in-scope instrument.

    Test whether it survives without a subsidised market maker before treating it as a durable improvement.

    Lower is better

  • Concentration index

    TE.MKT.CONCENTRATION

    How concentrated activity is among participants.

    High concentration flatters volume figures and is a fragility. It also blocks a peer result from being published.

    Depends on the case

  • Point-to-point integrations

    TE.INT.INTEGRATIONS

    Number of bespoke integrations required for the stated coverage.

    Usually the largest single driver of the fixed cost, and it scales with the network rather than with volume.

    RequiredLower is better

  • Reused integration components

    TE.INT.REUSE_RATE

    Share of integration work reused from earlier initiatives.

    Decides whether the second use case is cheap or costs the same as the first. Claimed often, measured rarely.

    Higher is better

  • Time to launch or replicate

    TE.STR.TIME_TO_MARKET

    Elapsed time to bring this, or the next one like it, to market.

    The clearest measure of the option a platform actually bought.

    Lower is better

  • Critical partner concentration

    TE.STR.PARTNER_DEPENDENCY

    How concentrated the initiative is on partners it cannot replace.

    A dependency is a price you have not been quoted yet.

    Lower is better

  • Required-field coverage

    TE.EVD.COVERAGE

    Share of the required metrics that carry a value or a reasoned exclusion.

    A case can be precise and incomplete. This is the number that says which, and it is derived so it cannot be talked up.

    RequiredCalculated, never typedHigher is better

    answered_required / total_required * 100

  • Evidence quality

    TE.EVD.QUALITY

    Weighted score across evidence reliability, comparability, completeness, reproducibility, reconciliation, recency and independence.

    Governs whether an observation can ever enter a peer distribution. It never changes the economic value itself — a weak case must look weak, not average.

    RequiredCalculated, never typedHigher is better

    see quality.ts — weighted components with hard caps

  • Result confidence

    TE.EVD.CONFIDENCE

    How much weight the calculated result can bear, from 0 to 1.

    Reported beside the result, never folded into it. A result can be economically precise and poorly comparable, or comparable and internally uncertain; those are two different problems.

    RequiredCalculated, never typedHigher is better

    f(TE.EVD.COVERAGE, TE.EVD.QUALITY, dispersion)

Taxonomy version 1.0.0. Every calculated result is stamped with the version that produced it, so a figure signed off today can be re-derived under the rules that produced it rather than under a later set.

Scope

What this is not.

Not a benchmark, yet

Confidential peer comparison is coming, and it is the reason the evidence discipline is built the way it is. For now the Lab is worth using with nobody else on it: the case you build is for your own decision, and nothing you enter is pooled, compared or published.

Not a vendor comparison

It measures what an initiative does economically, not which platform is best. Vendor estimates enter as class D evidence — usable for orienting a discussion, never as the basis for a decision.

Not advice

Nothing produced here is legal, regulatory, accounting or investment advice. It is an instrument for making an internal argument legible to the people who have to approve it.

Build your own

One institution, one initiative, one case a board can argue with.

The workspace is part of the Institute’s member programme. Your organisation gets a private workspace, seats for the people who have to fill it in, a review step so nobody approves their own numbers, and an export that carries its working. Nothing you enter is pooled, compared or published — there is nothing to pool it into.