Insights

AI and automation: how to estimate the savings before you invest

Techtone Group · Aug 24, 2026 · Last reviewed Oct 14, 2026

Do not start with the technology. Start with the processes: map them, measure what each costs today, compare those costs with companies that have already automated the same work, and invest where the gap is largest. That turns "where could AI help?" into a ranked list with a number next to each line.

Why most AI business cases are weak

Many AI initiatives begin with a tool and look for a use. The business case then rests on vendor claims or generic percentages, and finance teams rightly discount it. A stronger case starts from the company’s own processes and costs, and compares them with what peers that have already automated actually spend.

Step 1: audit the processes

• List the end-to-end processes in scope, for example invoice processing, customer onboarding, claims handling or reporting.
• For each, record volumes, the people and time involved, error and rework rates, and the systems used.
• Describe each process in a way that can be compared with external benchmark data: the same start and end points, the same units.
• Mark the automation potential of each step: rules-based, judgment-based, or dependent on unstructured documents and language.

Step 2: benchmark each process

Compare the cost per unit of each high-potential process with good and best practice from companies that have already implemented automation and AI. The difference is the financial gap. A large gap with high automation potential is a candidate; a small gap is not worth the change, however interesting the technology.

Benchmarks also show what is realistic. A process that several peers have already automated at scale is a safer target than one where the best example is still a pilot.

Step 3: plan the investment

• Rank the candidates by financial gap, investment need and implementation risk.
• Estimate the investment for each: licenses, integration, data preparation, change management and the cost of running the solution.
• Phase the plan: quick wins that fund the next wave first, platform decisions second.
• Set the measures before starting: cost per unit, cycle time, error rate and employee time released.

What usually decides the return

• Data quality. Automation magnifies messy data; cleaning it is often the largest hidden cost.
• Process redesign. Automating a broken process makes it fail faster. Simplify first.
• People. Savings appear only when released time is used or roles change; plan that explicitly.
• Governance. AI decisions in regulated industries need human oversight, audit trails and clear accountability.

An example of how the numbers come together

Take supplier invoice processing in a mid-sized company. The audit shows the volume of invoices per month, the minutes each takes to capture, check, match and approve, and the share that needs rework. Multiplying these gives a cost per invoice. The benchmark shows what companies that have automated capture and matching spend per invoice. The difference, multiplied by the volume, is the annual gap. Set against the cost of a capture and matching solution, its integration and the change effort, it gives a payback period that finance can test line by line.

The same logic works for customer onboarding, claims, reporting, contact center queries or document review. What changes is the unit of measure and the benchmark, not the method.

Where AI differs from classic automation

• Rules-based steps suit classic automation: workflow tools and robotic process automation handle them reliably and cheaply.
• Language and documents suit AI models: reading contracts, classifying emails, extracting data from unstructured files and drafting replies.
• Judgment steps suit a combination: AI prepares and recommends, a person decides, and the system learns from the decisions.

Separating these three types inside each process avoids paying for AI where a simple rule would do, and avoids promising full automation where a person must stay in the loop.

Questions to ask before approving the budget

• Which processes are in scope, and what does each cost per unit today?
• What do companies that have automated the same process spend per unit, and how reliable is that comparison?
• What is the full investment, including data preparation, integration and change management?
• When will the savings appear in the budget, and who owns delivering them?
• What will we measure each month to know whether it is working?

From speculation to numbers

The aim is simple: not to guess where AI can help, but to prove it with numbers before the budget is committed. A benchmark-based plan gives management a ranked list, a credible return per item and a measurable target for each project.

Related service: AI and automation benchmarking