by Liam O’Neill, MD – bpmd
Introduction
Process work has always involved a lot of legwork: interviews, notes, models, requirements, workshop packs, test scripts, governance reviews and follow‑up actions. AI is useful because much of that work starts with messy input and needs turning into something structured.
The risk is that people confuse a cleaner first draft with a better answer. AI can help a process professional get to the right answer faster. It can also produce generic rubbish with confidence. The difference comes down to judgement.
The useful use cases tend to show up in three types of process work:
- Supporting a major system transformation, such as an ERP programme
- Improving an end‑to‑end process after transformation or as part of a value realisation review
- Managing process data for quality, control, compliance and audit
1. When you are supporting an ERP transformation
In a major system transformation, process plays a practical role. It helps the business understand how work happens today, define how it should happen tomorrow, align business and IT around that future state, and reuse that understanding through requirements, testing, rollout and adoption. That lifecycle involves a lot of process work where AI can reduce time, especially around research, drafting, structuring and checking.
Build the current‑state baseline
A good starting point is the process architecture. Some programmes rush straight into future‑state design. That can work in a smaller business or a narrower project. In a large transformation, it is usually worth understanding the current way of working properly first. You need a clear view of the process areas, teams, systems, handoffs, major variations and known pain points.
Traditionally, a process team might use something like APQC’s Process Classification Framework as a reference point, then adapt it to the organisation. That still works. AI can speed up the first cut by challenging your process taxonomy, suggesting missing areas, comparing alternative groupings and testing whether the structure would make sense to someone outside the project team. You still need to take it to the business and make it fit the way the organisation actually works.
Use transcripts without outsourcing the interview
Right now, AI does not run complex process interviews well. Its main value is as a note taker. That is still useful. It can transcribe what people say and help turn a messy conversation into structured outputs: roles, systems, decisions, handoffs, controls, risks, exceptions and open questions.
The interview itself is skilled work. A good process interview needs a clear scope, enough structure to stay on track, and enough flexibility to follow an important comment when it comes up. You need to keep people engaged, manage the time, ask the right follow‑up questions and spot when someone is describing the official process rather than how the work actually happens.
AI is starting to move into this space. The Guinness price index example is a useful sign of direction: an AI voice agent called pubs across Ireland to ask for the price of a pint of Guinness. That shows how quickly AI can collect structured information from a large number of people. Complex process discovery is different. A tool will not reliably manage a difficult workshop, control the discussion, read the room, challenge weak answers or know when a side comment has exposed a bigger issue. Today, treat it as an accelerator.
Turn rough content into first‑pass models
BPMN models start with a trigger event, move through tasks and decisions, and end with an outcome. They can also carry extra detail on roles, systems, risks, controls, issues, data and regulations. AI is useful because much of that input starts as text. You can paste in a process description, Word document, Excel view or interview transcript and get a first‑pass model back.
That works best for simple, linear processes. It is much weaker with complex decisions, dependencies, exceptions, handoffs and different ways of working across teams. The strongest use case today is migration or first‑draft creation. If you are moving basic process content from Word, Excel or an old modelling tool into a proper repository, AI can create the initial draft faster. A process professional still needs to check the structure, BPMN quality, missing steps, wrong assumptions and missing metadata.
Shape the future state and requirements
AI is also useful when you start shaping the future state. If the target system has strong documentation, you can use AI to search and summarise that knowledge in a controlled way. For example, you can explore how a new ERP capability could change a process, where manual work might reduce, where standard functionality exists, and where the design may need a decision. That gives architects, SMEs and process leads something better to react to. It will still need validation because the tool will not know all the local constraints, programme decisions, data issues, integration limits or political realities unless those are provided.
The same applies to requirements and user stories. After a future‑state workshop, you usually have a mix of process models, notes, decisions, open points, system steps, exceptions and pain points. Turning that into requirements is time‑consuming and easy to delay. AI can take that raw material and produce a rough first cut. The useful work is then in the review: removing duplicates, spotting vague requirements, checking whether something is really a requirement, and making sure the wording matches the design agreed in the workshop.
Support mining setup, testing and adoption
A lot of the hard work in process mining happens before analysis. You need to understand the source system, work out which tables and fields matter, extract the right data, build the transformation logic and create the data model that makes the process readable. AI can help search system documentation, suggest relevant tables and fields, and draft parts of the transformation model. System experts still need to refine it, especially where the documentation is incomplete, out of date or different from how the client actually uses the system.
The insight side is weaker today. AI can suggest patterns and possible issues, especially where the data is clean and the process is fairly standard. It is less reliable when the data is fragmented, unusual or missing important events. The process professional still needs to judge whether the finding explains the business problem and is worth acting on.
Process work should also carry through into testing, rollout and adoption. Once you have the future‑state process, you can reuse that same understanding to shape test scripts, training content, work instructions and adoption planning. AI can turn process flows, system steps and requirements into a first draft of test scenarios or scripts. Digital adoption platforms and task mining tools can also capture user interaction data: clicks, screens, paths, delays and workarounds. Tools such as WalkMe, Mimica and Soroco use AI to make those patterns easier to interpret, which can help show where users are struggling or where the designed process is being bypassed.
2. When you are improving a broken process
The second story is process improvement. This could be a value realisation review after a transformation, a targeted improvement project, or a review of an end‑to‑end process such as accounts payable, order management, complaints, maintenance or onboarding. After a transformation, the question is usually simple enough: the system went live, but did the value land? Cycle time may still be too long, exception rates may still be high, users may still be working around the system, and the business case may depend on benefits nobody is properly measuring.
Prepare properly before the workshop
AI is useful before you get into the room. You can use it to research the process area, the system, the industry, the function and common failure points. This works better if you create a project or custom GPT loaded with the right context: system documentation, process notes, internal terminology, known issues and relevant market material. Then you are working with something primed around the problem, rather than asking a generic tool a generic question.
Take accounts payable. Before a workshop, you can explore common causes of invoice matching failure, typical controls, ERP configuration issues, supplier master data problems, buying behaviour issues and automation options. Those are prompts for better questions. The real process issues still need to come from the business, the data and the people who know the work. That preparation also helps with workshop material. If you already know the structure, story and content, tools like Claude and PowerPoint can help turn it into a cleaner workshop pack faster. That is useful when you have done the thinking and want to avoid hours of formatting. You still own the argument, structure and message.
Review your own facilitation
AI can also help you improve how you run interviews and workshops. Take a transcript from a process interview, discovery call or workshop and ask for feedback. It can show where you moved on too quickly, missed a follow‑up question, let the scope drift, or defended a point when a better question would have helped. That is useful because process professionals rarely get proper coaching on facilitation. AI will not make you a great interviewer, but it can give you a cheap review loop after each session.
Challenge the first explanation
AI becomes useful again once the business has described the problem. Say the team tells you invoice matching is poor. The first explanation might be badly formatted invoices or difficult suppliers. That may be true, and it may also be incomplete.
AI can help widen the list of things to test: poor PO discipline, weak supplier master data, inconsistent goods receipting, price variance, tolerance settings, contract leakage, duplicate suppliers or manual buying outside the agreed process. The value is that it stops the team accepting the first explanation too quickly. It can also bring in improvement ideas from outside the people already in the room. If the issue is document handling, it may point you towards intelligent document processing. If the issue is customer query handling, it may suggest classification, routing, knowledge retrieval or response drafting. If the issue is audit evidence, it may suggest automated control monitoring or exception‑based review. You still need to decide whether those ideas fit the organisation, system landscape, data quality, risk appetite and budget.
Keep value qualification real
Value qualification still needs to be done properly. If an improvement is worth pursuing, quantify it with real volumes, baselines and assumptions from the business. How much cost comes out? How much revenue is protected or gained? How much risk is reduced? AI can help tidy the wording, research benchmarks or structure the options, but the hard number needs to come from expertise, evidence and the people who own the process.
3. When you are managing process data, governance and audit
The third story is the more traditional process data role: process architecture, enterprise architecture, business architecture, quality, control, compliance and audit. Large organisations often have thousands of process models and even more dictionary entries: roles, systems, data objects, risks, controls, policies, documents, organisational units and applications. The challenge is keeping that content clean enough to be reused. If the repository is messy, people stop trusting it. Transformation teams rebuild knowledge from scratch. Roles and systems are inconsistent. Models are technically approved but out of date. The process data exists, but it cannot be used properly.
Keep the repository clean
Process repositories become messy quickly. Different teams use different names for the same role. Systems are entered inconsistently. Data objects are duplicated. Ownership gets out of date. Controls are linked in one area and missing in another. AI can help review that metadata and suggest fixes. It can identify duplicate dictionary entries, inconsistent naming, missing links between models and systems, similar roles with different names, weak process descriptions, unused objects and gaps in ownership. Some process tools are already adding this kind of capability. SAP Signavio, for example, has a “Dictionary Doctor” style feature that reviews dictionary content and suggests improvements.
Make process knowledge easier to access
A common problem with process repositories is that people do not always use them. Some struggle to read process models or cannot find them in the first place. A process repository is a structured record of how work happens: activities, roles, systems, decisions, controls, risks, data and documents. You can connect approved process content through a controlled retrieval layer, so people can ask natural language questions and get answers based on your own process information. This is still early. AI struggles with complex, non‑linear models. It is usually better at pulling from descriptions, metadata and linked documents, or pointing people to the right process so they can check the source themselves. The value is easier access to trusted process knowledge. The answer should still link back to the source.
Assess regulatory impact faster
Another useful case is regulatory impact analysis. When new regulation appears, the hard part is not just understanding the wording. It is working out which processes, controls, roles, systems, data objects and policies may be affected. A well‑maintained process repository helps. AI can summarise the regulatory change, compare it against available process and control information, and produce a first view of likely impact areas. For example, it may suggest which processes need review, which controls may need updating, which systems handle relevant data, and which teams should be involved in the discussion. That does not remove the need for legal, risk or compliance expertise. The output still needs proper review. The value is speed: AI can help process teams get to a first impact view faster, so the right people can focus their time on the areas most likely to matter.
4. A word of warning
AI is useful, but process professionals need to use it carefully. This article is a decent example. I used AI to research modern blog practice, common AI‑writing tells and newer capabilities in the market. I then worked out the structure myself: three common types of process work, told through major transformation, process improvement and process data. I used ChatGPT’s annotate feature to talk through the content as a rough brain dump. AI helped turn that into a usable first draft. I then went through it section by section, challenged it, rewrote parts, removed things that were wrong, and read the full piece back to check the flow.
The same rule applies to process work. Use AI to research trends, find examples, draft rough material and package the work cleanly. The hard work still sits with the process professional: deciding what matters, what is true, what should change, and what the organisation should do.
A lot of process professionals sit in quality, risk or governance teams. That usually brings sensible caution, but it can also make teams slow to adopt newer tools. You may have tried some of this already and been underwhelmed. Maybe AI‑assisted process modelling was poor. Maybe automated analysis in process mining was shallow. Maybe generic prompts gave you corporate sludge. That may have been a fair judgement at the time. It should not be your final judgement. Something weak six months ago may now be usable. Something unusable a year ago may now be good enough for a first draft.
Some process work will become easier to automate: basic modelling, first‑pass documentation, data preparation support, requirements drafting and pattern finding. That changes where the value sits for process professionals. The useful skills are the ones that need judgement: asking good questions, challenging weak assumptions, understanding the business context, resolving conflicting views, connecting process change to business value, and helping people adopt a better way of working. AI can take on more of the legwork. The process professional still owns the judgement.
5. Five things you can try now
- Use meeting transcripts properly – If your AI policy allows it, record process meetings and use the transcript afterwards. Ask AI to pull out roles, systems, handoffs, decisions, issues, controls and open questions. That can add useful detail to your process and systems architecture.
- Use AI to support process mining setup – If you are setting up process mining, use an approved AI tool to help identify which tables and fields may matter in the next system you are looking at. Treat it as a starting point for the system expert to review, not as the final extraction design.
- Get coaching on your interviews and workshops – Take a transcript from one of your process interviews, discovery calls or workshops and ask AI what you could have done better. It can highlight missed follow‑up questions, unclear sections, points where you moved on too quickly, or areas to probe next time.
- Prime AI before an improvement workshop – Create a project or custom GPT for the process, function, system or industry you are working on. Load it with approved context, then use it to research common issues, trends, newer practices, technology options and improvement ideas before the session.
- Use AI to take the pain out of workshop decks – If you already know the story, structure and content, use a tool like Claude to create a first‑pass PowerPoint deck. Let it handle the formatting and slide production. Keep control of the argument.