Most AI portfolio discussions begin with a familiar exercise: score each idea for business impact and ease of delivery. The top-right quadrant becomes the roadmap.
That is a useful way to rank ideas. It is a weak way to sequence an enterprise AI portfolio.
Consider supplier invoice processing. On its own, it may not have the largest headline ROI in the portfolio. A customer-service copilot or revenue use case may score higher. But invoice processing can create something the next applications can reuse: document ingestion, extraction, validation, exception handling, access control and integration patterns.
That changes the investment question. The first use case should not be judged only by what it returns directly. It should also be judged by what it leaves behind for named follower use cases.
This article sets out a practical way to move from ranking AI ideas to sequencing an AI portfolio. The aim is simple: create value now, build only the reusable capability that real demand justifies, and make later solutions cheaper and faster to deliver.
At a glance
- Business value and readiness should qualify a use case before ROI is used to rank it.
- The best first investments are often anchors: they create value and establish capability that named follower use cases will reuse.
- Funding should move in waves. Later use cases should prove that reuse reduces delivery time and cost.
- Portfolio economics matter more than forcing every individual use case to pay for itself.
A ranked list is not an AI portfolio
Traditional prioritisation treats each use case as a separate investment. Enterprise AI rarely behaves that way.
Different applications repeatedly need the same engineering beneath them: access to enterprise data, document processing, retrieval, evaluation, security, system integration, monitoring and approval controls. If every project rebuilds these foundations, the organisation is funding the same engineering several times.
The additional question is therefore: which use case will make the next five faster, safer and less expensive to deliver?
That question turns prioritisation into sequencing. It also prevents the enterprise from building a large platform in advance of proven demand.
The portfolio sequence has three decisions
1. Qualify2. Choose anchors3. Sequence in wavesIs there measurable value, usable data, accountable ownership and a safe delivery path?Which use cases create value and establish capability needed by committed follower use cases?Can later waves reuse what was built and show falling delivery cost or time?Exhibit 1. Move from ranking isolated ideas to sequencing investments that create reusable capability.
1. Qualify before you rank
A high expected return is not enough. Near-term AI investments also need basic readiness.
A use case should have a measurable outcome. “Improve customer experience” is not enough. “Reduce request resolution time from 18 minutes to 12 minutes” creates a baseline that can be tested.
The same applies to productivity. Saving ten minutes on a low-volume activity may be useful but may not release meaningful capacity. Value depends on scale, adoption and whether the saved time changes cost, throughput, risk or service.
Readiness matters just as much. The business needs usable data, an accountable owner, a realistic integration path and an operating model appropriate to the risk. An assistant that drafts an email has a different control requirement from a system recommending a contract position or approving a payment.
Qualification removes attractive but vague ideas before they distort the roadmap. It still does not tell the enterprise what to build first.
2. Choose anchors for direct return and reuse return
An anchor use case does two jobs. It creates a measurable business result and establishes capabilities that committed follower use cases will consume.
Supplier invoice processing is a useful example. Its direct return may come from less manual entry, faster processing and fewer errors. But its engineering may also establish document ingestion, classification, field extraction, table extraction, validation, exception routing and system integration.
Those capabilities become valuable only when there are named followers. Purchase orders, contracts, onboarding forms, claims or compliance documents may reuse them. “Someone may reuse this later” is not enough to support the investment.
Field and table extractionClaims; compliance documents
Validation and exception routingContracts; supplier records
Access control and enterprise integrationOther document-led workflows
Exhibit 2. Reuse return should be linked to committed follower use cases, not to a generic platform promise.
3. Build and fund the portfolio in waves
Wave 1 should prove value and establish only the foundations the anchors genuinely need. Do not build a general-purpose AI platform for imagined demand three years away.
Wave 2 should deliberately select follower use cases that reuse the capabilities from Wave 1. Their delivery time and incremental cost should fall. If they do not, the supposed reuse should be challenged.
Later waves can introduce deeper integration, broader data and greater autonomy after the core patterns have been tested under real operating conditions. Funding should follow the same logic. Further investment should depend on adoption, realised value, reuse, falling delivery cost and acceptable operational performance.
Cost-neutrality belongs at portfolio level
Not every AI use case needs to fund itself independently. The first anchor may carry more of the cost of shared retrieval, document processing, evaluation or governance.
Later applications should benefit from that investment. Over an agreed period, the portfolio should show that business value is covering an increasing share of shared AI build and run cost.
This requires two connected views. At use-case level, track adoption, realised benefit and incremental cost. At portfolio level, track shared cost, reuse and whether each new implementation is getting cheaper or faster.
The ROI of the first application matters. The economics of the fifth may say more about whether the strategy is working.
What Leaders should require
Ask for a sequence, not a ranking
The portfolio should show which use cases create value now, which reusable capabilities they establish and which committed followers will consume them.
Fund shared capability through real demand
Build common engineering through anchor use cases. Avoid creating a large platform programme before the portfolio has proved what genuinely repeats.
Measure whether reuse is real
For follower use cases, track delivery time, duplicated engineering, tested components reused, control consistency and incremental cost. Reuse should appear in operating numbers, not only in architecture diagrams.
The answer is a sequence, not a quadrant
The highest standalone ROI may still be the wrong first investment if it leaves the organisation with nothing reusable for what follows.
A better first use case creates measurable value, establishes capability needed by a committed second wave and helps the portfolio move towards better unit economics.
The practical test is simple: the next successful AI implementation should be easier to deliver because the first one existed.
Are you ranking AI ideas or sequencing an AI portfolio?
Bring us two or three high-value AI opportunities and the follower workflows already expected behind them.
Ampersand can help separate distinctive business logic from reusable engineering, identify viable anchor use cases and define what should be proven before the next funding wave.
Practical output: A capability-led AI portfolio sequence with anchor use cases, named followers, reuse assumptions and measurable gates for the next wave.
