Quick wins versus strategic projects in finance

Quick wins (high impact, high feasibility) and strategic projects — also called gold nuggets (high impact, lower feasibility) — form the two attractive quadrants from the Impact × Feasibility matrix for AI in finance. For finance teams the mix decides whether you end up with a collection of stand-alone AI tools in a year or a structurally different way of working — not one-or-the-other, but a portfolio with a deliberate ratio.

After scoring on Impact × Feasibility you're left with a matrix featuring two attractive quadrants: quick wins (high impact, high feasibility) and gold nuggets (high impact, lower feasibility). The temptation is to look only upward and merrily build. But what you pick from those two quadrants, and in what order, decides whether your finance function will end up with a collection of stand-alone AI tools in a year, or a structurally different way of working.

The choice between quick wins and strategic projects isn't either/or — it's a portfolio question. For finance specifically: too many quick wins without an underlying track yields fragmentation within a year; too many strategic projects without visible wins drains team confidence before there's any result.

Three horizons — finance edition

A useful split that works in finance:

• Horizon 1 (0-6 months) — Optimize: make existing finance routines faster, better, or cheaper. Quick wins: AI-assisted variance commentary, bank-reconciliation proposals, AR text drafts. Low risk, impact directly measurable, easy adoption.
• Horizon 2 (6-18 months) — Transform: how value is delivered changes. Processes get reorganized, roles shift. Gold nuggets: Close agent, VAT agent, Cash agent — projects that require investment in data and integration, but after go-live deliver a materially different finance function.
• Horizon 3 (>18 months) — Innovate: the proposition level. An embedded finance stack that prompts new behavior from customers, suppliers, or investors. For SME and scale-up finance: usually deliberately later — learn from H1 and H2 first.

A healthy AI portfolio in finance carries weight in all three, unevenly distributed. For a team just starting out: 70-25-5 — lots of quick wins, a few serious strategic projects, one experiment.

Why you should start with quick wins

The first year of AI adoption in finance is not a technology question, it's a change-management question. People need to gain experience, build trust, learn to organize work differently. That doesn't work via an 18-month project that only delivers at the end. It works with a stream of small successes visible in the close cycle, in the week, in standup.

On top of that, every quick win builds capability. The team that has delivered an AI-assisted variance commentary can build the bank-reconciliation flow in a fraction of the time. The infrastructure — authentication, prompt templates, review workflow, audit log — gets built only once. Quick wins are simultaneously deliverable and investment in the learning curve.

For finance this matters extra because the culture of precision and audit awareness makes the team cautious in adoption. A visibly quick win — "last month's variance commentary took two hours, this month 25 minutes at the same quality" — accelerates that adoption more than any beautifully presented roadmap.

When a gold nugget gets priority anyway

Three reasons to start a strategic finance project in parallel with the quick wins.

Strategic necessity. A competitor or sector trend forces it: an investor expects a rolling forecast for the next round that you don't have today; a PE fund wants close within 5 days across the portfolio; a new accounting package or ERP migration is the natural moment to rethink. Waiting on "learn first" is more expensive than starting early.

Data prerequisite for everything. Sometimes feasibility is low across every interesting finance use case for the same reason: the data is poorly organized. The chart of accounts is incoherent across entities, cost centers are used inconsistently, dimensions are missing. In that case the underlying data-cleanup project is a gold nugget that unlocks dozens of quick wins. Don't start with quick wins that won't work; fix the foundation first.

Scale and cash. A use case with large cash impact (an AR flow that reduces DSO by a week on €10M of revenue = ~€190K of cash) is substantial enough to justify a longer track, even at lower feasibility. ROI covers the complexity.

In all other cases: a portfolio weighted to quick wins, one or two gold nuggets in parallel, the rest parked.

What a finance quick win looks like technically

A typical AI quick win in finance has these characteristics:

• Existing tool as the host (Microsoft 365, your accounting package, your reporting environment) rather than a new app.
• Language model via an existing platform (Copilot, Claude, ChatGPT) or a built-in Saldus building block — no custom development.
• Prompt templates and a project space instead of code.
• One role or one team is affected (AP, AR, controller), not the whole finance function.
• A human reviews before output goes out or a booking lands.
• Measurable in minutes-per-task, errors-per-week, or euros of cash effect.

Examples: draft replies on AR mail, summaries of management reports, classification of purchase invoices, first drafts of variance commentary, extracting amounts from scanned invoices.

A quick win that isn't in production after four weeks is probably not a quick win. That's a signal to revisit the feasibility score.

What a finance gold nugget looks like technically

Strategic AI projects in finance have a different profile:

• Custom integrations via MCP or API with the accounting system.
• Data infrastructure that must be brought into order beforehand (chart-of-accounts cleanup, dimensions, lookup tables).
• Optionally a vector database or search index over internal documents (close manual, tax documentation, historical reports).
• Workflow automation across multiple systems, often agentic — see From prompts to agents in finance.
• Changes how an entire process flows (the whole close, the whole AR), not just one step.
• Governance and audit get explicit attention, with an approval layer between agent and ledger.
• Six-to-twelve-month lead time, project structure with a steering committee where finance, IT, and compliance sit at the table.

A finance gold nugget is a regular IT project with AI as a critical component. The project discipline that comes with it — scope, planning, stakeholders, go/no-go — applies in full.

The proof-of-concept trap

A common pitfall: the first AI initiative becomes a proof of concept that never reaches production. The team builds something, the demo impresses, and then it fades because nobody owns the production rollout.

Antidote: every initiative you start, even the smallest, has from day one a named end user, a production owner inside finance, and a decision moment after four to six weeks (go/no-go on rollout). Without those three it isn't a quick win, it's an experiment — fine, but call it that, and time-box and budget-cap it.

Where this lands in practice

• A 60-FTE scale-up starts with three quick wins at once: variance commentary for MT reports (live after three weeks), automatic data extraction from inbound invoices (six weeks), and draft emails for AR (four weeks). Alongside, a gold nugget runs in the background: an integrated Close agent, with chart-of-accounts cleanup as the first phase. The latter only delivers direct results after nine months, but then unlocks multiple follow-up use cases.
• A 90-FTE professional-services firm starts the other way: with a strategic project for an AI-driven knowledge base. After six months there's something technical, but adoption lags and there are no intermediate results carrying the initiative politically. After a reset three quick wins are added; within two months the sentiment changes.
• A 30-FTE family business deliberately picks only quick wins in year one — six to eight in finance — and only picks up a gold nugget in year two (full AR automation with cash impact), with year-one capabilities as the base.

What works: deliberate portfolio, continuous rollout, and strict discipline on "is this a quick win or actually a project." What doesn't: wanting everything at once, or waiting for the perfect strategy before starting.