2025 process automation predictions: AI, agents and citizen developers

Niamh Lordan00:00

Hello everyone, and welcome to today's webinar, 2025 Process Automation Predictions. My name is Niamh Lordan, I'm head of marketing here at FlowForma, and I'll be your host for today's session.

Niamh Lordan00:17

A few tips before we start. This webinar is being recorded and your lines are muted. If you need any help, for audio support or otherwise, email info@flowforma.com. We want this session to be as interactive as possible, so please take part in our polls and pop your questions for the panelists into the questions box. We'll do our best to answer them in the live Q&A.

Niamh Lordan00:55

I'm delighted to be joined by two fantastic panelists. First, James Corcoran, head of AI and analytics at STAC. Thanks for joining us, James.

James Corcoran01:13

Thanks for having me, Niamh. It's great to be here.

Niamh Lordan01:16

And Gerard Newman is our CTO here at FlowForma. Thanks for joining the panel, Gerard.

Gerard Newman01:26

No problem, Niamh. Good to be here.

Niamh Lordan01:35

We've had loads of interest in today's session. Everything related to AI and automation for the year ahead; I think it's going to be a transformative year for automation. We're here to equip you with actionable insights and tips to help you get ahead of the curve in 2025 and achieve the outcomes you want.

Niamh Lordan02:05

Our agenda today covers five top predictions from our panel. Toward the end we'll share our top three tips for automation success, give a quick intro to FlowForma, as I know a lot of people on the call are new to the product, and then we'll have a live Q&A, so pop your questions into the chat.

Niamh Lordan02:35

Prediction one. A key prediction for 2025 is that AI is no longer going to be on the periphery. It's going to become much more deeply integrated into business operations, with all the outcomes that come with that. James, I'd love to put that prediction to you.

James Corcoran03:09

If you look at 2024, it was undoubtedly the year we all became familiar with AI, and generative AI in particular. Everyone started using ChatGPT or various LLMs, in their personal lives and at work. Most organizations will admit that employees used generative AI organically to become more efficient, but on an individual or small-team basis.

James Corcoran03:46

The prediction for 2025 is that enterprises will get a lot more serious about rolling out generative AI at scale, in a controlled and predictable way. That covers how teams collaborate using generative AI in their day-to-day processes, and enterprise guardrails, so we can use generative AI, LLMs and chatbots securely, in a way senior management is comfortable with. There are platforms that standardize that implementation, but the most important thing is that businesses are starting to look at this holistically across operational areas.

James Corcoran04:46

I'm familiar with financial services, healthcare and telco, and I think every industry is going to implement generative AI. I tend to look at horizontal operational areas, and use cases like customer support and back office will integrate AI in a much more scaled way in 2025. Interestingly, a lot of my conversations are also about middle management, looking for ways to run the business more efficiently. So we'll see more AI and more thoughtful implementation at the enterprise level, and that will differentiate 2025 from 2024.

Niamh Lordan05:42

I love the piece about thoughtful integration; I think that sums it up very well. Gerard, anything you'd like to add?

Gerard Newman05:55

Last year in our predictions webinar we talked about bring your own AI to work, and that's exactly what happened. Whether or not organizations had a strategy around AI, during 2024 people brought their own AI tools to work to help them do their jobs. We also have a lot of AI creeping into organizations through applications: nearly every SaaS application now has an AI element, some of it useful and some of it not, and things like Copilot in Microsoft Office are gaining traction.

Gerard Newman06:37

So AI really is ubiquitous, and the real challenge is to embed it into your organization on your terms, not just let it happen. That's about defining your goals and strategy, then preparing your organization for more embedded AI by educating people, communicating the overall strategy and integrating gradually over time.

Gerard Newman07:02

Over this year we'll see some organizations embedding AI as the next step in their evolution, improving themselves and their processes, and others using it to revolutionize what they do: changing their offerings and processes and bringing different products to market. Organizations will progress at different speeds.

Niamh Lordan07:35

Prediction two is a practical one from Craig Le Clair, a digital process automation analyst at Forrester. He thinks the key to automation success in 2025 will be balancing AI innovation with the scale and reliability of proven tools. Gerard, how do you recommend people achieve that balance?

Gerard Newman08:15

Depending on which predictions you read, opinions vary, but I personally think 2025 is a bit early for completely automated processes running on generative AI in the background. Instead, organizations will look to leverage AI within their existing automation platforms. If they're using DPA or RPA tools, they'll see how they can gain further efficiencies by layering AI into those environments.

Gerard Newman08:52

We may well see rudimentary AI agents making certain types of decisions in 2025, in cases where there's clear data, clear parameters and clear rules. Integrating those agents into a process will bring efficiency and productivity gains, make processes more responsive and remove some of the delays humans introduce.

Gerard Newman09:31

Organizations that are already on their automation journey with DPA are in a better place to layer in AI, because they have an existing automation framework. They've been collecting data for years, they're clear on their processes and where the decision points are, and they have monitoring and controls around the process. All of that makes it easier to introduce AI.

Gerard Newman10:04

We've talked a lot about keeping the human in the loop. When you introduce a generative AI capability into a process, people will start by monitoring it closely, and what the AI agent is effectively doing is preparing work for you. Established DPA tools are constantly adding AI capabilities, not just to make processes more efficient, but to make building them more efficient too.

Gerard Newman10:46

Organizations that haven't started an automation journey might be holding out for further AI advances. But I think in 2025 they'll realize that automating their existing processes puts them in a much better position to take advantage of what AI brings this year and in the years after.

Niamh Lordan11:13

Thanks, Gerard. James, what's your perspective?

James Corcoran11:25

There's a big difference between proving AI can work and getting it into production. I talk to a lot of the global investment banks in London and New York, and one senior executive told me recently that generative AI is more of a software engineering problem than a data science problem, which I thought was a neat way of looking at it.

James Corcoran11:57

To Gerard's point, mature automation tools have figured out how to get into production, how to automate workflows in a scalable and explainable way. There's a gulf between writing a toy AI app and the mature automation solutions out there. So this balance is about right: integrating AI into existing automation frameworks and tools is the fast track to production this year.

Niamh Lordan12:51

Given this prediction is about success for the year ahead, I wanted to run a quick poll asking what automation and AI success looks like for you. Our poll isn't working, so maybe I should have used AI to generate it. We'll skip that one.

Niamh Lordan13:27

Prediction three. It wouldn't be a 2025 predictions webinar without talking about agentic AI. All the analysts and consulting firms say agentic AI will be a significant focus for organizations in 2025. James, how do you see this prediction unfolding?

James Corcoran14:02

You can't talk about AI in 2025 without talking about agents, and there was a lot of hype about them in the second half of last year. Despite all the analysts saying agentic AI will completely revolutionize the way we work, we need to stay grounded in what this type of AI can actually do for us and how to roll it out.

James Corcoran14:40

Agentic AI is a new architecture, a new way of deploying this technology, and as with any new technology the architecture evolves very rapidly. What preceded it, retrieval-augmented generation (RAG), evolved significantly too: what was state of the art one month was suboptimal six months later. Agentic AI will evolve through 2025, and adoption at scale will likely hinge on trustworthiness and efficiency.

James Corcoran15:33

At the moment it's tricky to guarantee trustworthiness, and the software architecture around agentic AI is probably suboptimal. That will evolve quickly through the first half of 2025, but even the big AI players, who have a lot to gain, cautioned late last year that there are things to be aware of on trustworthiness and efficiency.

James Corcoran16:19

That said, I've had several conversations with senior executives at large banks in London about how they plan to roll out agentic AI in 2025. They're looking at how to define roles for agents within teams, how agents and humans collaborate, how to orchestrate multiple agents, and even how to interview agents for selection and grade, score and rank them, much as you would in an employee's annual review. The banks are already well into the proof-of-concept phase and treating it as much a software engineering problem as a data science problem.

Niamh Lordan17:21

Thanks, James. Gerard, what's your view on agentic AI this year and in the coming years?

Gerard Newman17:33

Do we have enough time to discuss it today? You can't escape it. Focusing on process automation, I think it's going to have a huge impact over the coming years. Process automation is about standardizing your processes and making them more effective and efficient, and agentic AI can sit inside those processes, automate more and potentially replace certain processes.

Gerard Newman18:10

For simple processes that exist in most organizations, an agent may well replace large chunks of the process, removing human involvement while you still monitor and oversee it. You can do some of that today, but what's coming down the tracks will significantly change it.

Gerard Newman18:40

Following on from James's point, I'd see more autonomous agents blended into existing processes: replacing steps, doing preparatory work like gathering information from different sources and bringing it to the person who has to decide, and making recommendations, with a human in the loop making the decision. Overall we'll see intelligent agents working alongside existing process automation in most cases.

Gerard Newman19:16

It's hard to predict exactly when we'll get the more intelligent AI that enables all that decision-making; people are saying toward the end of this year. But overall I think agentic AI will have a very positive impact on process automation.

Niamh Lordan19:36

I love the complementary angle. If agents complement the work we humans are doing and help us all be more productive every day, it's a win-win.

Niamh Lordan19:59

Prediction four. If you've been following FlowForma for a while, you'll know we've been talking about the rise of the citizen developer for a few years now. What we're seeing in 2024 and 2025 is that AI has accelerated this trend even more. Forrester says that in 2025, 30% of generative AI-infused apps will be created by citizen developers. Gerard, are you surprised by those numbers?

Gerard Newman20:41

I totally agree. We've been proponents of citizen development for a long time, and AI is enabling it even further. We can take a cue from software development. In the last year a slew of tools has arrived that let developers create software far quicker, from things like a natural language prompt. That's made existing developers more productive, and it's also bringing new developers on board, who learn how to get the best out of these tools rather than spending all their time on architecture patterns and syntax.

Gerard Newman21:41

That will lead to an explosion of new software applications over the coming years. There are lots of questions about how they'll be maintained, but those are problems for the future. I think the same thing is starting to happen in process automation: generative AI will help citizen developers implement more widespread automation, and make it easier for more business users to become citizen developers.

Gerard Newman22:17

A common barrier to widescale process automation is a lack of understanding of the processes that exist in the organization, and the time the people who know those processes have to work with someone to automate them. People don't have time to invest in their own automation, which is a vicious circle, and I think generative AI will help overcome it.

Gerard Newman22:50

Generative AI gives us a huge knowledge base of processes in different industries. You can try it today with Copilot or ChatGPT: ask about a process and you'll get good answers. We've built that into the tool we use to build processes, so someone can chat with a large language model, sketch out a process or create its skeleton, and then build it.

Gerard Newman23:37

Instead of going to a business user with a blank piece of paper and asking them to explain everything the process does, you can show them an example process and ask how it compares with theirs. The business user is more engaged, because they can see it's a short road to automation, not a long one. It makes existing citizen developers more productive and encourages more people to get involved, lowering the barriers to entry.

Gerard Newman24:39

After that, we'll see a lot more generative AI-based insights into what's happening in your processes. A citizen developer builds and implements a process, gets feedback and improves it, and users become more engaged. Generative AI will help with that feedback loop, so citizen developers don't build something, forget about it and come back years later for a major overhaul.

Niamh Lordan25:25

I think it's really going to help with scalability as well, and the people who own the processes are the ones who know them best. James, any final thoughts on this prediction?

James Corcoran25:46

Gerard's framework is a good way of looking at this. To double-click on software engineering productivity, it's one of the huge use cases for generative AI, with huge return on investment. The thing to bear in mind is that we don't want every person in the organization to become a citizen developer and create their own app, so that everyone has different apps with different views of the world.

James Corcoran26:29

We need to roll out this technology so that technologists and business users collaborate to create standardized apps and processes, with one view of the world. Then you can have the debates you need to make decisions without multiple viewpoints colliding within the frame of reference you're working in. That's one thing to be aware of when it comes to citizen developers.

Niamh Lordan27:11

That was a nice segue to ROI. 2024 felt like the year of AI copilots, and in 2025 there's a clear focus on showing ROI from AI initiatives. Gartner says organizations with tangible ROI from AI initiatives will outperform their peers by 25%. James, can you share some tips on achieving real ROI this year?

James Corcoran28:14

That quote from Gartner is probably about right. I'd start by seeking out high-value use cases, and making sure you're not running around the organization with a solution looking for a problem. Start with the problem and figure out where the business impact is.

James Corcoran28:39

A lot of my clients in large organizations are setting up hybrid structures, with a centralized AI team that prioritizes use cases, manages the program, runs the compute platforms and sets standards and frameworks for AI development. The build-out, and accountability for the success of each implementation, then sits with the business units. That's a decent way to set up AI in medium and large organizations.

James Corcoran29:31

ROI is difficult to measure at the start of a project, because you won't know how much compute or technology you'll need, or how many people it will take to build. Expected efficiency gains are easier to quantify. I'd put AI rollouts in two categories: becoming more productive and efficient, with cost savings, and becoming more innovative, building new products and services to compete.

James Corcoran30:15

On the second category, generative AI tends to give you a standardized answer; it reverts to the mean. Innovation often comes from a contrarian view, and generative AI won't always give you that, at least not without serious prompting. So the initial return on investment is more likely to come from productivity and efficiency gains in areas like customer success and back-office process automation, which are relatively easy to measure.

Niamh Lordan31:25

Thanks, James, that's great advice. Gerard, any tips from your side?

Gerard Newman31:36

It's about identifying the areas where you're actually going to succeed. You're introducing change into the organization, so it's always a change management problem too. It's important to pick areas where the culture is right and willing to take on that change, which helps your overall chance of success and of achieving the return on investment you're hoping for.

Niamh Lordan32:30

That brings us to the end of our five predictions for 2025. Next, we've pulled together James and Gerard's top three tips for success. James, you touched on high-impact use cases. Can you share a little more?

James Corcoran32:58

The last prediction was a good lead-in. Don't fall into the trap of having a solution looking for a problem, so identifying high-impact use cases is critical, and it has to be done in partnership between the business and the tech teams. There's no point either one coming up with a use case on its own. Prioritize repetitive and data-heavy tasks.

James Corcoran33:41

Repetitive tasks are interesting. Many people use generative AI to write a first draft of a document or process, then use a series of prompts to tweak it. That creates a new process that is itself repetitive: you're often asking the LLM to repeat an inference, getting an answer that's 99% the same as a minute ago, and you may be paying for the compute. Spotting those optimizations matters, whether you're building a new tool or rolling AI out across teams. But the key is collaboration between business leaders and technologists to agree the operational area and the first use cases.

Niamh Lordan35:13

Our second tip is about monitoring and optimizing continuously. Gerard, can you share more?

Gerard Newman35:26

When you introduce any new technology that works alongside people, you need to be able to monitor it. Our default position is often to trust technology, and with generative AI in particular we need to be careful, check it's doing what we expect, and keep improving it and making it more accurate.

Gerard Newman35:59

Having data from before you introduced generative AI is really useful, so you can compare, spot deviations from the norm, and manage any aberrations in the process. Continual optimization ties into our own philosophy too: with the data behind it, generative AI is well placed to suggest optimizations to your business operations, like identifying bottlenecks and areas to boost efficiency and effectiveness.

Niamh Lordan36:49

Our last tip: there's a lot of talk about the move toward hyperautomation, and scalability is key to achieving it. James, how do you achieve that?

James Corcoran37:07

If one thing is certain, it's that what's state of the art at the beginning of 2025 won't be by the end of it, which is why I like the word modularity here. Implementing AI in a modular way gives an organization the flexibility to bring in new tools, frameworks and techniques as new technology, software and hardware come to market this year.

James Corcoran37:49

I'd go with cloud-based implementations. The hyperscalers keep adding features to their existing services that make it easier to roll out AI and build tools, products and services on top of them. Modularity is key if you don't want to get caught re-implementing or re-engineering an AI-built product or service in the second half of the year.

Niamh Lordan38:28

Thank you both; these are very concise tips, and we'll share the deck and an infographic with attendees after the webinar. A lot of people on the call may not be familiar with FlowForma, so Gerard, can you talk for a couple of minutes about the product and the AI features you've introduced?

Gerard Newman38:59

FlowForma is a digital process automation platform, specifically targeted at business users who want to automate processes with no code, and that's what enables our citizen developers. It gives you all the tools in one place: forms to capture data, a workflow engine with rules, insights into how your processes are working, and communication features like document generation and email generation. Our customers have been building solutions with it for years across many industries and use cases.

Gerard Newman39:58

Over the last year we've introduced a number of AI features. We started with summarization, which summarizes a process to date so people can catch up when they join a process many steps in. More recently we introduced FlowForma Copilot. You chat with it to identify a process, its key steps and the data collected at each step, refine the responses until you have a reasonable description, and then build it, giving you a working automated process in FlowForma that you can use as is or configure further.

Gerard Newman40:48

Our AI strategy is to keep making the product easier to use and to add effectiveness and efficiency to your process automation through AI, for all the personas who use it: people who fill in forms, people who take part in workflows, and people who build them.

Niamh Lordan41:21

We've shared a lot today: five predictions and three tips for success. If you'd like to discuss any of these trends or see a demonstration of FlowForma Copilot, let us know in the questions box that you'd like a demo, a trial or a consultation, or visit flowforma.com after the webinar. We'd be happy to help you on your journey to success in 2025.

Niamh Lordan42:01

We're close to time, and lots of nice feedback is coming in. First question, maybe for you, Gerard. Emer asks: can AI tools like Copilot help with wider adoption of process automation within an organization?

Gerard Newman42:36

Personally, I think they can. One barrier to process automation is understanding all the processes in your organization, and it can be hard to collect and document all that information. Tools like Copilot help, because you can ask them about processes and get reasonably good answers, which is a much better starting position than a blank page.

Gerard Newman43:12

They also let you run experiments much quicker. You can sketch out a process in a brainstorming session and ask whether you could automate a particular business area. In our own FlowForma Copilot you can even load in drawings and diagrams and it will build a process from them. It's all about removing barriers, speeding things up and making it easy to experiment, and when you decide to proceed, you get to a much more complete point very quickly.

Niamh Lordan44:04

The second part of Emer's question: is it possible to add context to the copilot, so it knows more about her organization's data?

Gerard Newman44:17

That's a slightly tricky one. As we move toward agentic AI, your own organization's data becomes one of the core inputs into intelligent decision-making. With FlowForma Copilot, we've focused the interactions with the large language model on process automation, so we can constrain the answers and get much more focused results.

Gerard Newman44:53

You can provide some of your own data to generative AI, and it will try to tailor answers based on that. There are also techniques to augment the information large language models have available; James mentioned RAG earlier, and that's one option to make it more specific to your business.

Niamh Lordan45:13

James, Paul asks: how do you see AI agents supporting employees on a daily basis?

James Corcoran45:31

Probably somewhere between pilots and copilots. We've seen copilots introduced into office software, making people more productive. Agents are more like pilots: they make decisions based on an instruction and have autonomy to implement them. For 2025 and initial rollouts, I think agents will fall somewhere in the middle.

James Corcoran46:15

The first agent apps were more retail than enterprise: find me and book plane tickets and a hotel in Europe this summer. That's a reasonable use case, and I can see it becoming popular this year. It's very different from asking an agent to recommend a trading strategy for my portfolio and go and make a bunch of investment decisions. So somewhere between pilot and copilot is where enterprise adoption of agents will land this year.

Niamh Lordan47:04

Another question for you, James: what learnings from AI adoption in the fast-moving world of capital markets and big data could be brought into the world of DPA?

James Corcoran47:23

What I like about DPA is that it's proven in production, with demonstrable return on investment. The industry has figured out how to productionize automation and has built the infrastructure and scaffolding around it, like the monitoring and observability Gerard mentioned. Explainability is a key one: it's difficult at the moment to get an agent to explain the decision it made and why, with an audit trail.

James Corcoran48:07

So I really like the play of integrating AI into DPA. You get the best of both worlds and a fast track into production. I'd double down on explainability and traceability, because those are what tend to trip organizations up when they rush AI into production, and the world of DPA has great answers to some of those challenges.

Niamh Lordan48:45

One final question, Gerard, and a quick response: what impact do you think AI tools will have on employee satisfaction and productivity?

Gerard Newman49:06

We'll have to wait and see exactly how it plays out, but introduced correctly, I think AI tools should have a very positive impact on productivity and on satisfaction, because they take away certain tasks or help people do them. I'm not sure how many people like searching for information in SharePoint sites and directories; if AI makes that easier, it helps you do your job and improves satisfaction. But the culture point we mentioned earlier, and how it's introduced, are important.

Niamh Lordan49:56

Lots of questions came in that are very specific to use cases, systems and getting your data ready for AI. We don't have time to answer them all today, but we'll get back to everyone with an unanswered question as soon as possible.

Niamh Lordan50:14

That brings us to the end of today's webinar. Thank you so much to James, our external guest speaker, and to Gerard, and to everyone who joined us for your questions. If you'd like a next step, such as a demo or a trial, let us know, or find more resources at flowforma.com. We'll send you the recording in the coming days, and a short survey will pop up when the webinar ends; we'd really appreciate your feedback. Best of luck to everyone in the new year.