Who Should Take the Lead?
Unlike traditional applications with fixed functionalities, self-developed processes utilizing latest process technologies are inherently dynamic and continuously evolving. This flexibility not only opens the door to value creation but also introduces new challenges in maintaining efficiency and effectiveness. As these processes proliferate throughout the enterprise to challenge existing organisational structures, the complexity of managing them will increase significantly.
With ever growing adoption, within and across functional domains, a broader question arises: who should take the lead in the ongoing management of these enhanced processes and how will organisational structures change in the future, bearing in mind the increased velocity of processing, deeper levels of interconnectivity, demands of compliance, and even more relevantly the ever shrinking demarcation lines between internal and external stakeholders.
Let’s explore this further.
- Introduction: The Evolving Landscape of Process Management
In the contemporary business landscape, organizations are increasingly challenged to create more efficient processes, removing any sources of transactional friction, and evaluating how the promise of artificial intelligence (AI) will further benefit productivity levels.
It is in fact a real time live jigsaw puzzle, as one needs modern day business processes to produce timely, high quality data flows before AI can become a consideration. Many organisations are now finding out that there current data flows will not readily support AI without change.
As a result, in order to achieve a realistic starting point, enterprises and their users must navigate the intricacies of modern process design, implementation, and pay specific attention to their ongoing management. The latter is increasingly important, and should be a special area of critical focus, due to processes becoming more interconnected and mission critical.
This criticality is because front office and back office functions are becoming more intertwined, seamless and fast moving, and can contain real time interconnections between internal and external systems, including Open Banking APIs.
These modern processes are driving powerful data transformations to produce actionable contextual workflows with relevant supporting documentation to the materiality required + simulations. This results in timely, usable high quality data. So one can appreciate that operational risks are therefore increasing due to these interconnections.
AI or no AI. Should that even be a Question?
First and foremost the effectiveness of a process is its ability to generate timely high quality information in a way that is useful and adds value to relevant stakeholders, without overloading them with heaps of peripheral information.
There is a missing, not so obvious, link to grasp here. In many cases, implementation of modern processes, even without AI, can provide timely high quality information. This is because modern technologies can now get around the previous core barriers of compute capability, plus have the ability to define agile data transformational processes at an ultra-granular level. Once data quality has been achieved the introduction of AI becomes a viable next step, but without achieving high data quality progress stalls.
Process Capabilities without AI
Putting this into context, as to the possibilities and power of the technologies today, complex financial consolidations can be automated to very high levels, illustrating that transactional friction existing in the form of spreadsheets, reconciliations, changing entity structures / groupings / segments etc can be automated to the degree required by the users.
Additionally, full compliance capabilities can be built-in into processes, noting also that data sets can be enriched for traceability (for example explanations of calculations and rationale behind tiered allocation amounts, originating source ledgers in complex segmental analysis etc).
This is all before AI introduction can drive even further deeper value. However, it is worth introducing here that there are inherent limitations with using AI working with financial data, which at its lowest form is an exact science. Financial models are built on deterministic principles, where outcomes can be predicted with a high degree of accuracy based on known variables.
Here are some key points to consider:
Process Capabilities with AI
In essence, any subsequent AI introduction will come down to evaluating what one is aiming to drive from it. For example, i) whether processing decision points normally taken by a human can be automated, and to what extent this can be done to an acceptable level (typically this will be a top down AI which is explained below); ii) the level to which automations will be executed to maintain trust in the results ie full or partial automation, or even simply enforcing an eyeball review; and iii) the purpose and anticipated value of the planned end result against the investment to make it happen.
One should always keep assessing, what is the purpose of AI within a process; for example is it:- i) simply presenting you with appropriate facts about your data based on questions that you ask; or ii) is it providing holistic insights that make a difference by taking you further than currently possible today ie by analysing a massive number of data permutations / combinations / operating simulations to specific goals that could not be achieved before this time; or iii) getting the computer to take decisions as if human.
Further below, we explore the different types of AI. This will become important to answering the questions posed above, as we tease out how various types of AI might be leveraged to reach a desired result. We will also address further the extent to which results produced by AI should be trusted by you.
Common Ground
As businesses grapple with these challenges, an important question arises: who should take the lead in managing these enhanced smart processes. This is something that is gaining critical importance, particularly surrounding the depth and breadth of change ie departmental, across functional areas etc.
Unlike before, where vendor upgrades provided changed functionality, what we have now are many more processes that cross multiple applications / systems of record with on line real time connections to other business ecosystems eg banking, logistics, HR related compliance processes etc. With process velocity increasing and shrinking lines between domain functional areas, business risk is increasing with organisational structures potentially challenged.
2. Understanding Processes vs. Applications
To effectively manage processes, it is crucial to understand some fundamentals between processes, and applications. Applications are typically designed by external vendors with specific functionalities, compliance options, and limitations, serving defined purposes within an organization, including being a trusted statement of record.
For instance, a payroll application may automate salary calculations and tax / pension deductions, but its capabilities are largely confined to those tasks. In contrast, processes such as those contained within Employee Self Service (ESS) systems are broader, unique to your business, and more dynamic, encompassing a series of activities that can be continuously refined and adapted to meet your specific changing business needs.
Processes can be designed for multiple scenarios and these will typically bridge across-applications and across-ecosystems to facilitate decision making. Companies can also build their own application using the same technologies. So in essence your corporate systems will be: Applications + Smart Processes.
This flexibility allows organizations to innovate and respond to demands as well as unique aspects to their business more effectively. However, it also presents an evolving management challenge, as processes require ongoing evaluation and adjustment to ensure they remain aligned with organizational goals. Unlike applications, which are typically implemented and left to operate with minimal oversight, resultant processes demand proactive management and continuous improvement.
Interestingly, it also contrasts current realities vs past ones. For example, many organisations continue to be slightly reticent in updating vendor applications for fear of upsetting their current operational status quo or them introducing new challenges (functionality changes, software defects etc). This really contrasts with modern day process functionalities that encourage small changes and tweaks or larger changes that today are more a way of life in our business and personal lives.
This in essence is a broad reflection of what we are discussing here ie deeper levels of agility: after all custom processes are small in size as they contain no excess functionality, other than that required by you. As a result programs are small in size, faster to execute as they are lightweight and contain no excess code, and benefit from being more easily maintained / modified with a small threat vector surface.
The integration of AI into process management, however, will need to be managed to a higher level. While AI can enhance processes, it also requires careful oversight to mitigate risks such as data privacy concerns (ie where and how data is being processed, encryptions etc) and algorithmic bias.
Furthermore, it is also important that i) operational budgets are put into place for future process management changes, noting that vendor low code / no code functionalities help you reduce costs here to some extent (noting that ease of use and depth of functionality are diametrically opposed) and ii) that one thinks through how trusted systems of record / access controls will be maintained by the organisation.
Working with vendors who specialise in business processes, ie software is optimised by them to address typical issues, provides deeper benefits that will further help minimise the investment associated with roll outs.
3. Types of AI for Process Enhancement
There are three broad types of AI approach that can be used to enhance processes: these are i) Generative AI; ii) Bottom up AI; and iii) Top down AI. Each requires a different level of investment to train and run models, recognising that there can be multiple algorithms / algorithmic types in any single process that might be run sequentially or in parallel, and that the approaches described herein also allow you to also leverage algorithmic models from your other trusted third parties (DeepSeek, ERNIE, OpenAI, Google etc).
- Generative AI: often abbreviated as GenAI, is a subset of artificial intelligence that focuses on creating new content across various media, including text, image, audio, and video types by utilising LLM’s (Large Language Models). This technology utilizes “generative” models to learn patterns from existing data and produce new outputs based on user prompts or speech inputs. It can provide immediate results, an easy example being the extension of ESS processes so that employees can use natural language to get relevant information contained within the employee handbook. Circular process management enables the fine tuning of source documents so as to ensure clarity, so processes become more useful
- Bottom Up AI: models start with raw data and learn patterns autonomously, making decisions without predefined rules. This approach, fundamental to machine learning and neural networks, relies on training data (LLM’s) to learn from examples rather than following explicit instructions. A good example would be document recognition for processing incoming invoices
- Top Down AI: This is used by those who prefer applying previously acquired knowledge to focused problem solving. Top-down AI = ‘Data-efficient AI’, because operationally it needs less data. The top-down approach involves starting with a high-level understanding of the problem and then breaking it down into smaller, manageable components. In essence letting the system make decisions based as if it were human. An example might be processing multiple leave applications fairly, based on multiple criteria ie seniority, other staff leave being requested for the same dates, project status, staffing levels etc. Conversely it might be used to improve work:life balance by identifying negative operational characteristics ie billing rates which are over and above planned utilisation rates, and where no holiday has been taken or is planned etc, leading to employee burn out risks
- Hybrid AI: A combination of the above. For example GenAI for the employee handbook, infused with specific corporate policies eg maternity leave guidance over and above statutory allowances etc. Procurement is another example: using vendor packaged Bottom Up AI for document recognition (no need to build all of your own code), but top down AI where there are complex overlapping / prioritised overriding tiers of supplier discounts available
In all cases careful consideration needs to take place around all aspects of employee engagement, privacy and cybersecurity. This will prevent, for example, sensitive information being put into a non-approved public Gen AI model ie your sensitive data being used to train their model.
On a more technical note, systems architectures can be designed granularly so you can leverage edge computing capabilities to keep sensitive information in-house, so you can decide whether compute and storage processes are cloud based, on-premise or hybrid. This might see you undertake cloud bursts where data is processed but not stored in the cloud for scalability considerations etc.
Where does Agentic AI fit?
It fits into Hybrid AI where a final solution will be made up of multiple algorithms which are of Bottom Up / Top Down / or GenAI types. Any one algorithm in totality will be formulated using deterministic and/or probabilistic formulae. The former is an exact result and the latter a probability assigned to a result. Multiple algorithms can work in sequence or in parallel.
What distinguishes agentic AI is that operationally it is all about autonomy. It can learn from feedback and experience. Algorithmic designs to achieve this might include Bayesian algorithms or other methods like reinforcement learning, deep learning, and multi-agent systems. A good macro example, which gives clarity to the concept, surrounds autonomous driving cars that proactively adapt to real time conditions. However, there is a longer way to go using AI with financial data.
This style of agent also means that it will need to be proactively managed over its life until retirement. It will need on-boarding, changing access rights as they self learn and adapt. At retirement, rights will have to be withdrawn plus adjustments “managed” should there be any dependencies.
Organisations are eager to use AI with financial data, but accuracy is paramount in order for results to be trusted by users. With AI suffering from hallucinations and with accuracy not being 100%, there needs to be a way for understanding how AI models are performing in practice with financial data.
Some organisations are therefore leveraging FinanceBench, the first benchmark specifically designed to evaluate LLM performance on financial questions. This benchmark includes a large dataset of over 10,000 question and answer pairs derived from publicly available financial documents, aimed at addressing the limitations of existing LLMs in financial applications.
However today, organizations have to design algorithms carefully for consistent acceptable results and also put into place compliance controls. As a result, and for the foreseeable future, most business processes will have an element of both deterministic and probabilistic formulae. With AI developments moving fast, eventually AI deployments will be easier to implement, scale and deploy.
Here are some key ways AI enhances processes:
Process Automation: AI enables the automation of repetitive and mundane tasks, allowing employees to focus on higher-value activities. For instance, it can handle data entry using AI document recognition technologies in procurement activities; simulations to find the product mix that generates the best cash flow or profit etc.
Data-Driven Insights: AI can analyse vast amounts of data to uncover patterns and trends that may not be immediately apparent. This capability allows organizations to make informed decisions based on real-time data, enhancing strategic planning and operational efficiency.
Worthy of mention is that simulations can be driven on a more practical basis than ever before, but although the technologies are there to be leveraged, there may be a knowledge gap as to how to utilise them to best advantage. In broad terms the same is true when it comes to deploying smart processes with actionable contextual functionality, so there is a bridge to cross to ensure that employees can deploy these new technologies to best advantage.
Enhanced Customer / HR related Employee Experiences: AI-powered chatbots can provide instant support to customers, answering queries and resolving issues around the clock. This not only improves customer satisfaction but also reduces the workload on human customer service representatives.
Thinking more granularly and specifically about one area describe above. For Process Automation: think of internally deployed Virtual Assistants, with Virtual Eyeballs and Virtual Fingers.
- Virtual Assistants: improve end user productivity @anywhere @anytime within the process. They can ensure that high quality information is presented to the right person(s) in a timely manner for proactive decision making. The key here is presenting data to the required materiality level, so as to streamline processing
- Virtual Eyeballs: allow for operational variances to be ranked on the fly to drive directional actionable contextual workflows with supporting content to the materiality required. This can be for numbers in aggregate, by entity, by operating segment etc. For example, it might be instant alerts to product managers to identify fast or slow moving goods based on algorithmic analysis etc
- Virtual Fingers: gives process owners an on-screen only messages that are not printed on reports as a pointer to potential areas of trouble
Selecting AI Variables / Positioning AI within Processes
While AI offers numerous benefits, organizations must also be mindful of the challenges it presents. Remember that AI is easy to describe at high level, but actual practical execution is very specific in nature ie in any algorithmic design you have to select the right number of variables that actually impact your required outcomes, rather than a selected variable simply moving as a consequence of operational circumstance. These decisions will impact both AI training and IT infra-structure costs.
One also has to place the algorithms within processes for their execution, but as processes can be built on a granular basis this should not be an operational issue. They can be deployed where required on premise or in cloud.
Moving on, there is an emerging challenge that companies are now trying to solve.
4. Challenges re Managing Processes
Managing processes can pose significant challenges for organizations. Think about your current operational environments and those in the future. Businesses will have i) Applications + Processes that themselves cross multiple applications and ecosystems, with or without AI.
Presently, many companies are at varying stages relating to their digital enablement strategies. However, one striking statistic is that 75% of digital enablement projects fail to deliver against their original plans. This typically can be due to the lack of access to timely, high quality data (as described above), which could be system or know how related, or simply that expectations were incorrectly set ie the delta that exists between easily describing AI versus the laser focused mathematical execution of algorithms etc.
One additional fact is relevant. Employees might have knowledge gaps or blind spots when it comes to process design and roll out. Often today, some businesses (often not by choice) use month-end reporting as the trigger point for taking corrective actions, and this has become their traditional modus operandi benchmark for reviewing business operations, which due to time lapses of deadlines can hide a host of transactional frictional issues.
More poignantly, there is often no management of i) the number of full time employees (FTE’s) or temporary staff utilised for a process function within or across entities; ii) no quantitative management of transactions within transaction types, eg # of transaction by type, execution timeframes, especially critical ones that impact cashflow etc; iii) no analysis of the # of spreadsheets used and why; iv) no oversight of the actual usage of reports produced (often complex reports are produced with great effort but not used).
Some corporates, who utilise BPO or that have regional service centres, already have these types of analytics baked into processes, but even then there can be surprises and potential licences cost savings as data can be passed thru different vendor sub-systems to a final system (issues here again is often simply down to data transformation capability / cost and connectivity).
Later below, we will explore who should control and manage the process design, but first let’s add into the mix how existing HR structures are being challenged today
5. Emerging Challenges
Holistically speaking, processes are i) increasing in velocity; ii) becoming richer in functionality, including the use of APIs between applications and ecosystems; and iii) as a result, activities continue to shrink demarcation lines between functional areas, including front office and back office systems.
As a result, traditional HR organisation structures are likely to be increasingly challenged, and business processes will have to be more proactively managed on an ongoing basis due to increased mission criticality and dependencies. As a consequence of processes becoming richer, and faster, there will also need to be a higher degree of co-operation between business functional domains, both within and across countries / regions.
This creates two specific challenges with the end to end process design:-
- Planned Depth of Change: This is the most significant area to tackle and will become increasing challenging to do well. Specifically, figuring out project scoping and the depth & breadth of change required to deliver required value. Changes taking place within a domain functional area are relatively easy, but with ever changing process dynamics greater business value creation is going to originate both within functional domains as well as shrinking the gaps between them. This will be regardless of whether they be internal or external functions, particularly noting that businesses today can be organised geographically (ie centrally /regionally); by functional discipline; by hemisphere, eg east : west etc
- Making Change Happen; This also continues to be a challenging area to get right. This is because different functional areas often have different layers of management working to different standards of operational compliance (cut-off and shadow IT being two examples), thereby hindering collaboration. Independent siloed environments often lack trust and open communication, making it difficult to share information and work effectively together. Employees accustomed to working independently often resist changes that require enhanced collaboration. Successful silo elimination requires strong leadership support and commitment to foster a collaborative culture. As noted above, determining the depth and breadth of change and supporting it with strong management is critical for success and is going to be a future differentiator.
6. Employee Augmentation & Trusting AI Generated Results
With AI there are other important ramifications to consider. As described above, there are different types of AI and these deployments can be simple or complex. In very complex algorithms, process outputs from bottom up AI / large language models are not always auditable, simply as there are so many computer neurons being used to reach a result. Additionally, there is the concept of hallucinations where very feasible sounding results are in fact incorrect.
As a result, the degree of trust that will be given to processes will be different, and this will be process dependent. In many transactional processes, this will result in human operators being augmented by AI. This augmentation is particularly relevant when probabilistic formulae are used. For example, in onboarding documents using AI, if the accuracy of a result is evaluated as 90%, then which documents contain errors etc. Over time, user focus can further fine tune efficiency.
7. Process Management – Proactive Engagement
Moving forwards the ongoing management and scoping of both simple and complex processes is going to become critical, particularly where there are multiple algorithms in play.
This is going to positively benefit stakeholders, as more emphasis will have to be placed on efficiency and as a result KPI’s / OKR’s introduced which will provide the currently missing link ie a quantitative approach to process management, a discipline that is only partially executed in most businesses today.
Think of this as an evolutionary change. As businesses move to systems that are more proactive, and that provide actionable contextual information, decision making will change to be based on timely, high quality information. A data driven organisation.
Process Ownership – Short Term
Experience shows, that at the start of process deployment, that process ownership will typically primarily reside within a business domain function, and be supported by different operational teams such as IT, other functional domains etc. For continuity purposes there are likely to be different tiers of process owner (in contrast to many process users) within a domain to ensure process continuity.
The entire process, in more complex scenarios, will likely be sub-divided into smaller individual processes, with individual automated steps taking place sequentially. This approach puts into place the checks and balances that ensures processes are deployed correctly. As time passes, these process steps will be fully or partially automated with eyeball reviews as appropriate.
Process scope can be described in broad terms as i) those that are complex tasks undertaken by a few specialist staff, and ii) complex processes, like FP&A, undertaken across the entire company.
Digital Teams to Support Functional Areas
Creation of self-contained digital teams that facilitate functional areas in making changes for digital enablement is something that many corporates are doing now. Within the digital team all key areas are covered ie security, privacy etc. This approach works to a point but has a potential weakness if not managed at a senior level, in terms of ensuring that the scope of a project is too conservative or too aggressive to make sense. In essence, the key question becomes are digital teams working strategically or tactically.
Process Ownership – Medium and Longer Term Scenario
The longer term will present deeper levels of complexity and continuity concerns due to organisational structures being fuzzier and less discrete with more stakeholders. This is because there will be many inter-connected processes and dependencies across the enterprise + ecosystems, such as logistics and embedded banking processes, not to mention that privacy and security considerations will have to scale.
A few organisations are now starting to grasp that i) a long term phased plan would be highly beneficial so that stakeholders have a road map as to the shape of the digital process organisation, and ii) that organising the development, management, and risk associated with broader scope process development is going to become essential, especially as these processes are going to cross internal functional domain lines.
At the moment this is an emerging scenario, but let’s explore some options that may be put to use later as complexity grows:-
HR Managed (Employee Engagement, Change Management, Performance Management)
HR systems might be a logical place to explore options, and one can quickly see that a process could be positioned as a “digital employee” or as an augmentation to a specific employee. It also provides a proactive conduit to understand the impact of staff changes on automations.
For example, it can be used to store the modus operandi details of each automated process / sub process, recognising that there may not be a 1:1 relationship between numbers of connected push / pull processes, and responsible persons. Each named automation, at the decided level of granularity, would have;
- Job description; sponsor; responsible process owner /manager / participants, each with clear roles and responsibilities
- Other responsible functional management dependents. This may include IT (licences, privacy, security etc)
- Process review dates
- Operational Details: location of execution for compute and storage, overseas data transfer details if applicable, AD considerations as needed, modus operandi details that include policies (including data retention), operating procedures and standards
It also fits well with the currently changing business dynamics, as companies start to focus more at the intersections of domain functional areas for value creation. Fundamentally, there is great value to be released at the intersections where applications meet.
For example, there can increased focus on proactive human capital management efficiencies / risks by functional domain (eg intersection of HR and Financials), where this also becomes a powerful future conduit for handling change management ie new enhanced processes, redeploying and retraining staff. It also facilitates businesses to better understand increased process velocity, shrinking demarcation lines, and compliance risks.
The above scenario is not perfect, and enhanced HR systems might not be the solution, but one can start to appreciate the details and risks associated with increased business velocity and shrinking domain demarcation lines.
IT Managed (Technical Implementation, Data Management, Support /Maintenance)
On the surface this also works to a point re process development and execution, but it does not solve the wider issues associated with change management, plus evolving and more fluid organisational structures.
8. Final Thoughts
The important takeaway is that business velocity is increasing and demarcation lines between domain areas are shrinking. As a result, one can see that one needs to have a short, medium and long term plan into both i) managing and developing processes and ii) thinking through how organisational structure might change over time, with a special consideration into ongoing management of process execution to maintain ongoing continuity.
The irony here, and one potential positive outcome, is that as processes cross demarcation lines, they are likely to foster increased co-operation between functional areas and more transparency, which will reduce data silos and data duplication. Over time as demarcation lines shrink between functional areas this co-operation will expand. It will not remove the need for a long term plan and ongoing management, but it will over the longer term help facilitate change to some extent, with change management still being a critical area.
Conclusion
In conclusion, managing processes enhanced with or without AI presents significant challenges for an organization. The dynamic nature of processes allows for continuous optimization, but this flexibility can lead to complexities in implementation and oversight. As AI technologies become more integrated into business operations, the roles of the C-suite, HR, and IT must evolve to ensure effective management and continuity. Ongoing collaboration between these departments is essential to navigate the intricacies of process enhancement, address potential inefficiencies, and leverage AI’s capabilities. Ultimately, organizations that successfully manage these challenges will be better positioned to thrive in an increasingly competitive landscape.
