Modernization business cases often count only delivery cost and ignore the operational burden of the current system. AI cases may claim productivity benefits without explaining whether capacity can be used or how performance will be verified.

A credible return-on-investment model connects a defined business workflow to an observable baseline, complete investment cost and benefits the organization can actually capture. It also makes uncertainty visible.

Define the decision and comparison period

State what management is deciding. Examples include rebuilding a bounded order application, introducing an AI-assisted service workflow or modernizing a data platform that enables several use cases.

If the opportunity is still broad, the AI consulting process from use-case discovery to measurable business value can help define the decision before the financial model is built.

Choose a period long enough to include delivery, transition and operations, and use the same period for every option.

Compare:

  • continue with the current state;
  • improve or stabilise the current system;
  • modernize or rebuild;
  • buy and integrate a product where viable;
  • add AI with the minimum required foundations.

For the technology option itself, use a structured build, buy or integrate AI decision framework rather than assuming custom development is always the answer.

The current state still incurs maintenance, incidents, manual work and continuity risk.

Establish a measurable baseline

Select a bounded workflow and collect evidence before estimating improvement.

Labour and process

Measure volume, handling time, waiting, rework and escalation. Separate capacity that can genuinely be redeployed.

Technology run cost

Include infrastructure, licences, support, maintenance and contractors. Do not remove costs shared with other systems incorrectly.

Change cost and delay

Measure time for changes, testing, releases and reconciliation. Slow systems impose a cost when improvements wait.

Quality and service

Record errors, incidents, recovery time and customer or employee impact.

Risk exposure

Document unsupported components, security findings, key-person dependency and recovery weaknesses. Risk belongs in the decision even when it is not a cash benefit.

Name the source and owner of each assumption. Show a range when the baseline is uncertain.

Calculate total investment cost

Include all work required to reach and sustain the target state:

  • discovery, architecture and design;
  • software, data and integration delivery;
  • migration, cleansing and reconciliation;
  • model, API and infrastructure usage;
  • security, privacy, legal and compliance work;
  • testing, evaluation and user acceptance;
  • change management and training;
  • parallel operation and cutover;
  • client employee time;
  • support, monitoring and continuous improvement;
  • contingency for identified uncertainty.

Model the complete route to usable value. AI pilots omit integration and operations; rebuild estimates often omit migration and business transition.

Classify benefits by how they become real

Every benefit needs an owner and capture mechanism.

Avoided cost

Confirm when licences, infrastructure or external support costs can genuinely be removed.

Capacity released

Automation may reduce handling time. Released capacity creates value only if it supports more volume, improves service, avoids future hiring or is redeployed to valuable work. Do not automatically treat every saved minute as a cash saving.

Revenue or margin contribution

Link faster onboarding, conversion or new services to a commercial mechanism. Do not claim the full revenue change when other factors contribute.

Quality and service improvement

Convert reduced errors or faster response to money only where a defensible relationship exists; otherwise report the operational measure separately.

Risk reduction

Modernization can reduce the likelihood or impact of outage, security failure or key-person loss. Use expected value carefully: estimated probability multiplied by estimated impact. Show risk reduction separately when estimates are too uncertain for the central ROI.

Use transparent financial measures

Simple ROI is:

ROI = (Total benefits − Total costs) ÷ Total costs × 100

Payback period is the time until cumulative benefits exceed cumulative costs. Net present value discounts future cash flows using the organization’s approved rate:

NPV = Sum of discounted net cash flows − initial investment

ROI shows proportional return, payback shows timing, and NPV reflects the value of money. Finance should confirm the method, tax treatment and discount rate.

Build monthly or quarterly cash flows rather than placing all benefits in year one. Delivery, adoption and system retirement happen at different times.

Create base, downside and upside cases

AI performance and modernization scope contain uncertainty. Use scenarios rather than one forecast.

  • Downside: slower adoption, extra integration work and delayed retirement.
  • Base: evidence-supported performance and expected delivery.
  • Upside: stronger adoption or volume where a plausible mechanism exists.

Vary the assumptions that matter most: eligible transaction volume, task success, employee adoption, migration complexity, operational cost and time to release. A sensitivity table shows management which evidence deserves investment before approval.

Avoid hiding contingency in a single percentage. Link it to named risks and reduce it as discovery creates evidence.

A hypothetical worked example

Consider an illustrative case workflow with 30,000 cases annually. Analysis suggests 40% are eligible for AI assistance, with a four-minute handling-time reduction subject to quality and adoption.

Potential annual hours released are:

30,000 × 40% × 4 minutes ÷ 60 = 800 hours

This is not automatically a cash saving. The business must show whether capacity avoids hiring, supports growth or is redeployed, then subtract review, exceptions and operations. Confirmed license or support retirement is more directly cashable. This illustration is not a RITE NRG client result.

Include modernization dependencies in AI ROI

An AI use case may depend on data quality, APIs, identity, workflow state and observability that the legacy system does not provide. The business case must decide how to allocate those foundation costs.

If a foundation supports one use case, include it there. If it enables several owned initiatives, model it as a shared platform investment. Do not rely on speculative future uses.

Conventional software may produce most value, with AI handling variable steps. Evaluate the combined workflow rather than forcing every improvement into an AI category.

RITE NRG brings AI consulting and automation together with software consulting and engineering so architecture, data and automation can be assessed as one investment.

Use staged investment gates

Reduce financial risk by purchasing evidence in stages.

Gate 1: opportunity and baseline

Confirm the business owner, process, current cost, data access, risk and candidate options. Stop if the problem is not material or measurable.

Gate 2: architecture and delivery case

Define the target system, integrations, controls, migration, acceptance criteria, cost range and benefit assumptions. Decide whether to build, buy, integrate or retain.

Gate 3: production pilot

Deploy a bounded capability with real users and measure task outcomes, operational cost, adoption and severe failures. A demonstration without workflow integration is not ROI evidence.

The distinction between demonstration and operating evidence is explored further in AI proof of concept vs production system.

Gate 4: scale and retirement

Expand only when measured performance supports the case. Capture benefits by changing work, removing old cost and retiring systems safely.

The RiteWay approach uses frequent production evidence, human approval gates and ownership by design to prevent speed from outrunning the business case.

Assign benefit ownership

Technology teams can deliver capability, but operational leaders capture value. Name an owner for adoption, process change, workforce planning, commercial impact and system retirement. Review forecast versus actual benefits after launch.

Track leading indicators such as eligible volume, completion rate and adoption, alongside financial outcomes. When performance differs, diagnose whether the cause is model quality, data, workflow design, user behavior or an incorrect baseline.

Frequently asked questions

What is a good ROI for an AI project?

There is no universal threshold. Compare the risk-adjusted return, payback and strategic value with the organization’s investment criteria and credible alternatives.

How do we value time saved by AI?

Measure the time on eligible work, then state how capacity will be captured: avoided hiring, greater volume, improved service or redeployment. Do not treat every saved hour as cash.

Should risk reduction be included in ROI?

Include it when probability and impact can be estimated responsibly. Otherwise show risk reduction as a separate decision measure rather than inventing precision.

When should benefits start in the model?

Benefits should begin when the capability is in production, adopted and replacing current cost or improving outcomes. Ramp them according to realistic deployment and change-management timing.

Make the business case testable

A strong ROI model is not a promise; it is a set of assumptions that delivery can test. Connect every benefit to a baseline, owner and capture mechanism, include the full production cost and use staged gates to turn uncertainty into evidence.

To build a defensible business case and roadmap for an AI or software modernization investment, contact RITE NRG.