Optimising processes with AI: a method, rapid results, measurable value
Experiments with artificial intelligence are on the rise, but few are being permanently integrated into business processes. Starting with a single process – and measuring it before and after the intervention – enables organisations to decide where to invest based on real results.
According to Istat, by 2025 artificial intelligence will have been adopted by 16.4 per cent of Italian companies with at least 10 employees – double the figure from the previous year – whilst the European average, as reported by Eurostat, stands at 20 per cent. The Artificial Intelligence Observatory at the Politecnico di Milano reports that 71 per cent of large Italian companies have launched projects involving the use of AI, but only one in five uses it extensively within the organisation, and many of these still struggle to measure its benefits.
Why do so many initiatives falter halfway through? The limitation is not always the technology. Far more often the mistake lies in not having a clear understanding of which problem one wishes to solve, or in entrusting a probabilistic system such as AI with the task of autonomously managing processes that, for the company, must remain strictly deterministic. Consequently, companies fall into two traps: lengthy programmes that produce no verifiable results for months on end, and Proof of Concept projects that demonstrate the technology works but never truly address the processes to which it is applied.
When, on the other hand, AI is used for a defined purpose and in a specific task, the results are clear to see. This is the case, for example, with a major software company – a Fortune 500 firm – where customer service agents, supported by an AI assistant, resolved on average 15 per cent more enquiries per hour. It matters how AI is used, and it matters where. In an experiment conducted by Harvard researchers involving 758 consultants, those who relied on AI for a task beyond their capabilities were 19 percentage points less likely to arrive at the correct answer. Deciding what to entrust to machines and what to entrust to people is one of the choices that makes all the difference. This division is based on precise rules: the reversibility of the action, the frequency of exceptions and the cost of error. Anything that is standard or easily correctable is automated, whilst rare cases and critical validations always remain the sole responsibility of humans.
For those tasked with authorising an investment, it is not enough to know whether the technology works and has been implemented correctly. Three further answers are needed: 1. whether the process has genuinely become more efficient; 2. how much time the redesigned activity now saves; 3. whether the benefit gained justifies a larger investment.
These answers can only be obtained if the result achieved is measured and compared with the data collected at the start of the analysis, using the same metrics.
It is from this understanding that AI Process Sprint (APS), the method developed by Orvian Innovation to take a process from analysis to a working AI solution, with the impact measured both before and after its implementation. The project approach is based on co-design, with the company’s needs and expertise remaining at the heart of every phase. The current process is analysed together with those who carry it out every day, as they are familiar with the routine exceptions that no manual records. The aim is not to introduce AI, but to improve an operational process, increase its added value and demonstrate the resulting impact.
APS begins by gaining an in-depth understanding of how the process is currently carried out – using which systems and by which people – and establishes a baseline in terms of times, volumes, errors and manual tasks. This discovery phase typically lasts two or three weeks. We work on one process at a time without losing sight of its links to upstream and downstream activities and to the systems that feed into it, because an isolated improvement risks merely shifting the problem elsewhere. Next, we determine what the AI will do and what will remain the responsibility of people, and develop the solution by integrating it into the digital processes and systems already in use. Very often, AI is used to manage the initial touchpoints in the workflow, acting as a pre-processing layer: it reads unstructured inputs (such as emails, PDFs or Excel spreadsheets) and transforms them into structured, verifiable data before feeding them into the company’s management systems, testing everything against real-world cases. Once implementation is complete , the results are measured against the same indicators set at the outset and compared, and the difference between the two represents the benefit achieved. On average, the entire process took around eight weeks.
Implementation examples show, following APS, a simplification of day-to-day operations, process optimisation and a significant reduction in errors typical of tasks involving too many files and excessive manual work. For example: in a tech company, the processing time per invoice fell by 73 per cent; in a large-scale retail company, manual checks on HR files decreased by 77 per cent; in a manufacturing company, 19 per cent of administrative FTEs were freed up. Results of a similar magnitude have emerged across very different processes and sectors, ranging from invoicing and sales to logistics management.
The data collected and the comparison between pre- and post-APS situations also form the basis for informed future decisions. They indicate whether there is a benefit, what challenges remain and whether it makes sense to extend the solution to other processes. And since the method and indicators are already defined, the approach can be replicated without starting from scratch, optimising investments in people, time and money.
To make AI work within a company, solving real problems and delivering a tangible impact, we must start with the problem to be solved and the value to be generated. The figures then tell us which technology to adopt and where to invest next.
