Contents
- What intelligent automation is (and isn't)
- Real case: Canon USA, invoice processing
- Real case: a top-25 US commercial bank, mortgage processing
- How to read an automation case study without being sold to
- The patterns that repeat in the deployments that hold up
- Frequently asked questions
- What we do differently
Most “intelligent automation case studies” online follow the same formula: an unnamed company, a round number (“30% cost reduction”), and no source you can check. This isn’t that. These are two real, named, dated cases with a quoted executive each, followed by a guide to telling a real case study apart from a sales deck wearing a data costume.
What intelligent automation is (and isn’t)
Intelligent automation combines robotic process automation (RPA) — software that repeats rule-based tasks, like copying an invoice field into an ERP — with AI capabilities that used to require a person: reading an unstructured document, deciding whether an exception should be approved or escalated, classifying what kind of request just came in. The practical difference from classic automation isn’t the technology stack, it’s that it can handle cases that don’t always follow the exact same pattern.
Real case: Canon USA, invoice processing
Canon USA automated invoice processing in its accounts payable department with UiPath Document Understanding, implemented by Greenlight Consulting. The initial goal was 75% straight-through processing with no human intervention. The result, documented in UiPath’s official case study:
- In under 9 months, Canon processed about 40,000 invoices (roughly 4,500 a month).
- They reached 90% straight-through processing, beating the original 75% target.
- Estimated savings: 6,000 hours of work per year.
- Before the project, manual extraction accuracy was around 35-40%.
“In less than nine months after deploying the UiPath solution, we processed about 40,000 invoices, or about 4,500 monthly. We initially had a goal of processing 75% without human intervention, but achieved about 90% straight-through processing.” — Thomas Earvolino, Director of Financial Systems, Canon USA
Before the project, Canon was carrying a three-to-four-month invoice backlog — exactly the kind of bottleneck we cover in why it still costs $9.40 and 9 days to process one invoice: the problem is almost never reading the paper, it’s everything that happens to the data after.
Real case: a top-25 US commercial bank, mortgage processing
Automation Anywhere documents a case involving one of its banking clients — one of the top 25 US commercial banks by deposit market share; the bank didn’t authorize disclosing its name, which is common in financial services, but the case and the figures are published by the vendor. The bank automated GAAR worksheet updates, appraisal orders and procurement across several business units:
- 100% error reduction in the automated processes.
- 2.6 fewer days in the full mortgage processing cycle.
- 6.3 fewer days managing appraisals.
- 32% faster appraisal ordering, measured just 4 days after the bots went live.
“We were amazed at how fast the benefits accrued. Our bots reduced the days to order and beat the manual process by 32% after just four days in operation.” — Director of Risk Technology and Execution, the bank
How to read an automation case study without being sold to
Vendor case studies (UiPath, Automation Anywhere, Blue Prism and the rest) are marketing material and should be read as such — that doesn’t make them useless, they usually include enough technical detail to be worth something if you know what to look for and what’s almost always missing:
- The denominator. “We saved 6,000 hours” means little without knowing over how many people, over what period, against what baseline.
- The build cost. Savings get reported; licences, integration work and ongoing maintenance rarely do.
- What broke along the way. Every real deployment hits exceptions the process owner doesn’t mention. A case study that reports none has been cleaned up for the brochure.
- How long it lasted. Automations decay when the systems underneath them change. A result measured at three months tells you nothing about year two.
Apply those four questions to the two cases above — and to any vendor deck you’re handed.
The patterns that repeat in the deployments that hold up
- High volume, low variation, stable rules. Invoice matching and claims intake automate well. Anything requiring judgment on unstructured input historically didn’t — which is exactly what’s changed with agent-based approaches, see AI agents vs RPA.
- A named owner. Deployments without an internal owner degrade quietly until someone notices the numbers stopped being right.
- Measurement agreed in advance. If nobody defined success before the build, the result gets argued about afterward instead of measured.
Frequently asked questions
How long until an intelligent automation project shows a return? In the two cases above, the first measurable results landed anywhere from 4 days (the bank, on ordering speed) to 9 months (Canon, on its straight-through processing target). It depends on the scope of the process, not a fixed number.
What processes are the best candidates to automate first? High-volume ones with low variation and rules that already exist in writing: invoice matching, order processing, worksheet updates. Anything that needs judgment on different cases each time needs AI agents, not classic RPA.
Is intelligent automation the same thing as RPA? Not exactly. RPA automates tasks following fixed rules. Intelligent automation adds AI to handle exceptions, read unstructured documents or make small decisions — like the Document Understanding engine in Canon’s case.
Why doesn’t the bank in the second case give its name? It’s common in financial services: the vendor publishes the case with real figures and quotes from the project lead, but without naming the institution, for commercial confidentiality reasons. That doesn’t change the fact that the case and the numbers are documented and published.
What if my company is smaller than Canon USA or a top-25 US bank? The patterns that make a project work (high volume, clear rules, an internal owner, measurement agreed in advance) don’t depend on company size, they depend on the process being automated. The project is smaller; the logic is the same.
What we do differently
At Optima Quantum we don’t show you unnamed case studies to sell a demo — we help you identify which process in your business has the volume and stability to make automation pay for itself, and we measure the result from day one against a real baseline, not a number that gets argued about after the fact.
Want to know which process in your business is a good candidate? Talk to our team or check our AI and automation services.
More on this: AI agents vs RPA, AI document processing for business, AI automation solutions: how to choose and what they cost.
Get our new articles by email
One email per article. Unsubscribe anytime.