How AI Is Rebuilding Business Processes and What Companies Should Do: Global Practice and Kazakhstan

By Bagdat Zhaxygeldin, founder of JB Solutions

For decades, business processes were automated on one logic: take the existing way of working and run it faster. Electronic document flow sped up approvals, the accounting system sped up postings, a robot sped up data transfer. The process itself stayed the same: a chain of handoffs between functions, each with its own performer and its own control.

AI agents change the terms of the problem. They read a document in any format, work through an exception, prepare a decision and carry out the next step themselves, which makes it possible to rebuild the process itself. In its review of October 5, 2026, McKinsey names the redesign of the work itself as the sign of progress and separates it from speeding up the old model [1].

Tools are being adopted faster than processes are being redesigned. According to McKinsey, 80% of respondents report higher personal productivity from AI, but only 37% see a positive effect on EBIT, and the leaders in returns make up 6%. Almost three quarters of the leaders have substantially redesigned their workflows; among the rest, a quarter have [1]. Below we look at what exactly is being redesigned, how it works, what gets in the way, and what a company in Kazakhstan should do.

Which processes are redesigned first. According to a Gartner survey (183 finance leaders, May and June 2025), 59% of organizations use AI in finance. The most common uses are knowledge management (49% of adopters), accounts payable automation (37%), and detection of errors and anomalies (34%) [8]. The scope is wider: banks run part of their mortgage and compliance processes with "squads" of agents, at one global bank an "agent factory" runs client due diligence with ten such squads, insurers are rebuilding claims settlement, and telecom operators use agents in customer service [2].

How the structure of a process changes.

How such processes are managed. An end-to-end outcome has an owner who is accountable for the goal, the economics and the constraints, and a small team with business, data, risk, compliance and cybersecurity expertise from day one. McKinsey calls this format a "mission factory": it redesigns work as a combination of people and agents, builds the controls, and leaves behind reusable components for the next outcome [1]. Above it sits an executive with the authority to reallocate funding and scarce specialists [1]. The organizational structure adapts to end-to-end processes (more in the article "Who Decides What in a Group of Companies").

In what order to act. According to HFS Research, IBM redesigned workflows in HR, IT, finance, procurement, sales and customer support for about 270 thousand employees: it first removed unnecessary steps and only then automated. According to the company's figures cited by HFS, annual savings over three years reached 4.5 billion dollars against a target of 2 billion [3].

Fewer than 10% of AI pilots reach production [3]. Gartner notes that 91% of finance organizations see low or moderate impact at the start, while those that have advanced are almost three times more likely to get a high one [8]. The Stanford Digital Economy Lab (April 2026, 51 successful deployments in 41 organizations across seven countries) found that 77% of the hardest problems were invisible costs: changes in how work is organized, data quality, rebuilding processes. 61% of successful projects were preceded by a failed attempt, and the common feature of those attempts is that AI was run as a technology project and applied to broken workflows [4]. The sample includes only successful cases. Lawyers, HR, risk and compliance resisted most often (35% of cases), so control functions should be brought into the project from day one [4].

Another reason: work is done differently from how it is written down. SSON cites a ScottMadden estimate under which 30 to 50% of process work is performed outside formal systems. We did not find the primary source, so treat this as an order of magnitude [5]. A published Dreher Consulting case shows the scale: at an equipment manufacturer with 320 employees there turned out to be 41 actual process variants instead of the three assumed, in 11% of orders production started without technical approval, and each such order lost 17 days on average. After a process owner and four metrics were appointed, cycle time fell by 23% in nine months. This is a single case, and the figures come from the company itself [6].

From simple to complex:

Before handing a process to AI, five things are needed (our working framework, there is no single standard): a map of the actual process for at least one end-to-end case; an owner with the authority to make decisions on data and rules across all companies of the group; a list of exceptions stating who resolves each one and how; a common identifier that links an order, a contract and a payment across systems; three to five metrics (cycle time, rework rate, cost per operation).

In its 2026 study of Central Asia and the Caucasus, KPMG notes that companies in the region are adopting AI actively, but a noticeable share struggle to scale solutions and do not measure the effect (as retold by Digital Business) [7]. The picture matches the global one: pilots are launched faster than processes and metrics are built. In Kazakhstani groups of several legal entities, where each company has its own accounting base and its own reference data, the gap between the regulation and reality is usually the widest (our observation from practice, we found no direct studies).

Which processes to start with (our recommendation). The right candidates are processes with a large number of repeating operations, a clear share of exceptions, available data and a limited cost of error. For most mid-sized and large companies these are supplier settlements and the payment calendar, reconciliation of receivables and payables, processing of bank statements and source documents, period close and management reporting.

How to proceed (our position, following the logic of global examples at an achievable scale):

We start with a diagnostic of one or two processes and take it through to a redesigned process that is ready for automation.

What we do. We study how the process runs today and document it (an as-is map with steps outside the systems and exceptions); we design the target model (to-be), removing bottlenecks and duplication; we lock the rules into regulations and a responsibility matrix; for AI we add a scheme of exceptions, controls and metrics, and an implementation plan with control points.

What we need from you. Access to one or two processes and to the people who perform them, system extracts for a sample of completed cases, and a responsible person on the company side with the authority to make decisions on the process.

What you get. The process map as it is, the target model, regulations, a responsibility matrix, a list of exceptions with owners, three to five metrics with baseline values, an implementation plan. After that the process can be handed to your own IT team or a contractor, and for finance tasks you can use ready-made JBS AI solutions.

Timing and cost. In our experience the work takes two to six weeks depending on the number of processes. We set the cost after the diagnostic. More on the service page "Business Processes".

Transformation and shared service centers. JB Solutions has experience of management transformation in large companies in Kazakhstan and Central Asia: in mining and metallurgy, oil and gas, railways, technology and telecom. We created shared service centers and holding structures, allocated functions between the legal entities of groups, and redesigned processes and organizational structure. More on the page Management and Consulting.

AI practice. For us AI is a working tool that we apply in our own projects and implement for clients: