A multi-year program is obsolete before the first milestone. Tokenomics maturity is built in 90-day cycles across two parallel tracks—the tech and the people.
AI capabilities evolve constantly. Any roadmap structured in years is obsolete before the first milestone is reached. A 90-day sprint is the only cadence that survives the pace of change.
Banks must adopt a 90-day sprint for AI as the only viable cadence
The six stages of tokenomics maturity map cleanly to a three-phase, 90-day sprint that produces live results each cycle.
Technology and people activation are not sequential. They are parallel tracks that must move together for the sprint to work.
Stage one builds the workload inventory. Stage two adds instrumentation. Stage three turns on attribution and chargeback.
Stages four through six layer optimization, scaled platform delivery, and continuous governance onto the foundation.
Why bank AI programs need a sprint discipline, not a multi-year plan
In our 4th article, How banks can bend the AI cost curve before their competitors, we established that the platform cost curve is the structural lever that determines whether bank AI economics scale. This article picks up the delivery cadence that turns that lever into live results inside a single quarter. Tokenomics maturity is built in six sequential stages. Each has a defined outcome, a defined set of capabilities, and a defined set of decisions for senior leadership. Maturity in stage ‘n’ requires meaningful capability in stages 1 through n minus 1. Attempting later stages without the earlier foundations produces poor results and erodes credibility with the CFO and the board.
The six stages of tokenomics maturity for banks
Stage 1: Foundation and inventory
The bank maintains a single, authoritative, continuously updated inventory of every AI workload, capturing token consumption, models called, business owners, and regulatory classification. This is the most common gap and the prerequisite for compliance and tokenomics discipline.
Stage 2: Instrumentation and observability
Every workload emits structured telemetry on token consumption, latency, error rates, model mix, and business outcomes across three layers: infrastructure, workflow, and outcome. Inventory tells you what exists; instrumentation tells you what is happening.
Stage 3: Attribution and chargeback
Every dollar of AI spend is attributable to a business unit, product, and outcome owner across three mechanisms: direct attribution, shared-service allocation, and platform infrastructure allocation. Run showback for the first 12–18 months, then transition to chargeback as the platform matures.
Stage 4: Optimization and FinOps
The bank runs an active, continuous practice that bends the unit-cost curve down while preserving or improving outcome density across six levers: model right-sizing, prompt optimization, caching, batching, commitment discipline, and workload sunsetting.
Stage 5: Scaled platform delivery
The bank’s platform crosses the inflection point where the marginal cost of the next workload is meaningfully below the cost of the first. The platform is funded as a platform, with centralized vendor relationships, and becomes the default delivery model for all new AI work.
Stage 6: Continuous governance
The bank operates a continuous governance regime integrating AI economics, model risk, regulatory compliance, and operational risk into a single accountable practice. Four pillars hold it together: integrated reporting, aligned cadences, common ownership, and continuous evaluation.
The 90-day sprint, across two parallel tracks
The six stages are not a sequential program to be handed to a project manager and tracked against a Gantt chart. The right discipline is three phases of 30 days each, repeating, with the technology track and the people track moving in lockstep.
Banks that activate technology without activating people produce dashboards no one acts on. Banks that activate people without activating technology produce committees that meet without data. The sprint works because the two tracks reinforce each other inside each 30-day window.
Why the dual-track sprint works
The 90-day model works because it is ruthlessly scoped. Each cycle focuses on the workloads that matter most right now, produces a live result leadership can see, and generates the evidence base for the next cycle’s investment decisions. The dual-track design adds a second discipline. Every technology milestone has a matched people milestone in the same 30-day window. The inventory ships with named owners. The instrumentation ships with a CFO in the FinOps review. The optimization ships with a Council decision logged against it. There is no waiting for a program to complete before value appears. Value appears in 90 days, or the engagement is not working. The cadence also matches the underlying technology. Model capabilities, pricing, and regulatory expectations are moving in quarterly increments. A program structured in years has its assumptions invalidated before its first phase gate.
What the full article covers
The PDF explains why this one does not. Inside: the six stages of tokenomics maturity walked through in full, including the CXO decisions each stage forces, the nine data points every workload inventory must capture, the three telemetry layers instrumentation must handle, and the six optimization levers a mature FinOps practice runs continuously. Plus the full attribution and chargeback playbook, the platform inflection behaviors that separate mature banks from stalled ones, and the four pillars of continuous governance that keep the sprint durable beyond the first cycle.
The dual-track 90-day sprint in brief
Phase 1: Days 1 to 30
Diagnose the workload estate, name the Tokenomics Council chair, map every AI-adjacent role to its tokenomics responsibility, and surface the accountability gaps.
Phase 2: Days 31 to 60
Bring priority workloads into observability, name business owners before instrumentation goes live, get the CFO into the first FinOps review, and close the accountability gaps.
Phase 3: Days 61 to 90
Run the first optimization cycle, log a Council decision against it, review outcome density alongside token cost, and start cycle two immediately.
Can the 90-day sprint model short-circuit enterprise rigor?
The sprint model reads like consultant theater until you notice what annual programs actually hide. A one-year plan buries its assumptions for four quarters at a time. A 90-day sprint surfaces them every quarter, when correction still costs nothing. The rigor is not lowered. It just arrives sooner.
How to activate the 90-day sprint in your bank now
Frame the next AI initiative as a 90-day sprint with a live deliverable, not a multi-quarter program with a phase gate.
Stand up the workload inventory and the Tokenomics Council inside 30 days, even if both surface uncomfortable gaps.
Pick the optimization lever with the highest expected payback for cycle one and execute it with a named Council decision.
Treat the two tracks as one sprint, not two programs, with technology and people milestones matched in every 30-day window.
A series for the agentic banking era
This is the fifth article in a seven-part series on tokenomics for banks. The next article addresses the ownership question every bank is currently getting wrong: who, by name, is accountable for the economics of intelligence.
The hard questions teams must ask before and after the first sprint
Most banks fund the first sprint from existing FinOps, risk, and transformation budgets. The recoverable spend from cycle-one optimization levers usually offsets the sprint cost inside the same quarter.
If a first sprint produces no live deliverable, the problem is scope. Cut the workload count, shrink to the top spenders, and rerun with a Tokenomics Council decision as the exit criterion for the next cycle.
Instrumenting workloads whose owners are not named. The result is visibility without accountability. The sprint prevents this by making named ownership a pre-condition for entering the instrumentation cycle.
Regulatory reporting fits into the continuous governance stage. Because instrumentation captures workload classification from cycle one, EU AI Act inventory and audit trail requirements are met as a byproduct.
