In my experience, successful transformation rarely happens all at once. It typically happens in stages.
This goes for both how we’re implementing AI at Serrala, and how we’re helping our customers to implement it in their process automation as we develop and extend our Platform’s capabilities.
For the past few years, we’ve been working at Serrala to create a new approach to finance automation that brings AI to bear. The main idea behind this is simple: everything needs to come together, because without a great foundation, you can’t really build anything big.
Our focus has been bringing our solutions together within a single platform. The reason is simple: AI can only be as effective as the data and processes behind it. AI is like any other software solution for enterprise before it: without unified, clean data, it simply doesn’t live up to its potential, much less its hype.
Our AI philosophy is simple: AI should help finance teams focus more of their time on judgement, decision-making, and strategic work
The ultimate goal for CFOs in an autonomous finance landscape isn’t simply running better processes. It’s leveraging automation to handle most of what we currently consider “core” finance operations workflows so that more specialized teams can do a whole host of new things: better scenario planning, better reporting, better allocation of value to strategic and tactical priorities. The main objective of the office of CFO is to get insights faster, get to predictions faster, and become more strategic, and your impact in your strategic role becomes higher.
We believe AI should always be used with this goal in mind. Which is why I’ve been working with our customers to create AI agents that bring value to their operations.
Autonomous finance won’t be powered by a single AI system doing everything. It will be powered by specialized agents that each perform a specific task well. A well-built agent needs to execute its task in the same way any financial clerk would have done it.
Once you have several individual agents dialed in on their specific tasks, you can orchestrate them all together into a more cohesive workflow. It looks like one big robot from the outside, but it requires careful balancing of task selection, governance, and humans in the loop.
But the most important piece is that trust layer. The trust layer is where we really apply the 40 years of experience we’ve had building systems with you. It provides the governance, controls, auditability, and human oversight necessary for finance processes. If confidence is low, the workflow should escalate to a human rather than proceed without the appropriate controls
Now, before we get to our three step process, let’s have a quick reality check.
The truth about autonomous finance? It won’t happen overnight
You should prepare for your adoption of AI and your journey to autonomous finance to be a crawl-walk-run process.No responsible finance leader is going to hand an entire process over to AI overnight. You can’t just flip a switch and say “Go ahead AI, do my job.”
Before we even get to step one, we need to pick the right use cases. And how we do that is very simple. We ask one question that cuts right to the heart of the matter: Does this process require reasoning?
AI’s major advantage over traditional automation is its ability to mimic human reasoning when it comes to complex, messy, unstructured data sources. If you want to correlate information in your ERP with a bunch of free text email discussions, you can probably use an agent to do this very effectively. But if a process is governed by clear rules, calculations, and workflows, traditional automation is often the better solution.
If you’re already far along on your finance automation journey, this should be music to your ears, because it means the foundation you’ve built doesn’t need to change. It will only get augmented and improved with AI. For those of you whose organizations are less mature, it doesn’t mean traditional automation is off the table. The right strategy depends on using AI where it’s strong, and other solutions where it isn’t.
A great example of this from our next-generation AP automation solution, which will come to market in the next few months. In this solution, we’ve created a price variance exception handling agent. If there’s a difference in the amount quoted on a supplier invoice versus your purchase order, someone has to investigate the supplier, the history, your tolerance policies, tax variances, and so on.
When you consider back and forth with the vendor, resolving cases like this can often take more than a week. Someone has to investigate supplier history, tolerance policies, tax variances, and related communications before determining the next action.
But if you hand the task to an agent with the right instructions and criteria, it can process all this information in under a minute, give you a likely reason for the variance, and recommend the next step for you. You don’t have to investigate—all you have to do is review the case and verify the fix, or decide to do something else.
This is exactly where AI delivers value because the challenge isn’t execution; it’s investigation and reasoning.
Another great example from the world of AR is our unmatched payments resolution agent. It analyzes unmatched payments, identifies the most relevant matches, and generates a list of recommendations for those payments to be matched. Your teams can accept or reject its recommendations based on audit logs that explain the decision, confidence intervals provided by the platform and their own judgement. This eliminates most of the manual number crunching and data cross-referencing we still see in the cash application process.
All of the agents we currently have in production are designed to work on cases like this, because they represent the expression of the first of our three steps.
Step 1: Build trust in agentic AI through focused use cases
To get the ball rolling, we need to pick use cases carefully and make them as specific and relevant as we can.
You can think of it like hiring a new employee. Just as we wouldn’t ask a new employee to run an entire function on day one, we shouldn’t expect AI to operate autonomously from the start. We’d give them a specific job to do, monitor what they’re doing to see how everything goes, do some performance testing and then make changes and link together responsibilities as you build up trust in their capacity.
This, by the way, is exactly how we’ve been embedding generative and agentic AI into Serrala products. They’re confined to clear features, specific places and specific tasks.
They don’t act autonomously right away. Instead, they present you with a recommendation and a rationale and allow your teams to authorize the action or do it themselves. Ownership stays with the team and with the leaders who control governance rules. The AI is there to recommend.
Paradoxically, limiting the capacity of the AI to begin with is what allows you to accelerate automation more quickly. It provides a safer, slower path to adoption. And again, just like an employee, it allows for regular performance reviews. Our platform allows you to measure AI performance directly with real-time visibility of accuracy and confidence levels. It’s fully transparent, there’s no AI “black box”, and you’re always aware of what use cases you have, how they’re working, and what needs to be done if they aren’t.
Step 2: Orchestrate agents into workflows
Once you’ve built up trust in individual use cases, it’s time to start putting together a more closely orchestrated “crew” of AI task flows.
In this stage, we take our smaller agents and start experimenting with how they work together. A specific “crew” of agents is one designed to work together, step by step, to execute an end-to-end workflow. Like, for instance, taking an incoming payment in the AR workflow, matching it to an invoice, purchase order, and remittance document, and posting it instantly with a final check by a human at the end of the process.
Just like in step one, your teams will still have oversight and control. But now, with multiple agents working together, you benefit from the elimination of handoffs and other bottlenecks that would otherwise delay key processes. AI now has more autonomy to execute the tasks it needs to execute, and we keep the human in the loop just for oversight.
I’ve actually had a couple of customers eager to bypass these two steps altogether. One even said to me “I want to wake up out of bed and start running, Rami.” But it’s important we learn to walk before we try to run, especially if we want to minimize risk and avoid big delays in the future.
When we start stage three, we’re finally ready to run, but it’s much easier. We don’t change what we’ve already been doing, we just let the crew do the job it’s been built to do, and give it a higher level of autonomy.
Step 3: Let agents operate by exception
At this point, you should be confident that your agentic crew can perform its assigned tasks without much human oversight. It’s always on, and always executing, in the areas where it can drive real value. Each task in the flow is fully “owned” by the relevant agent and agent crew, and the human in the loop is responsible for overseeing the whole flow rather than individual agent outputs.
Teams move from reviewing individual transactions to managing exceptions and monitoring outcomes.
Of course, you’ll always need to validate, and you’ll always need human judgement to determine what is and isn’t valuable and to make calls where you need real accountability. You’ll also always need to have audit logs so you can understand exactly what choices each agent has made to fix any errors that do come up in the process later on. And with what confidence scores these calls were made.
But at this point your workflows aren’t just helping improve operational efficiency and speed. Getting the orchestration for a given “bundle” of tasks right will help you focus on the next step, which is, again, controlling and unlocking working capital. As more operational work becomes automated, finance teams can focus more attention on working capital, liquidity, risk, and strategic decision-making.
It also allows us to get to the point where we have a fully integrated dashboard for the entire finance function that we can communicate with and extract data from on the spot with natural language commands.
And the accountability, the oversight and the controls are there and working even when you feel like you no longer need them.
How fast does the transition to autonomous finance occur?
In short: faster than you think.
We’re not worried about how much our customers plan rely on autonomous AI in their finance processes. We’re more focused on getting them to use it for appropriate scenarios in a successful manner. The Finance Platform is set up precisely to allow you to start slow, and expand fast, on your terms.
And this adoption and expansion can happen much faster than you’d think. My favourite example in recent months has been our smart invoice capture tool in AP.
We first launched this capability at the end of 2024. In less than 18 months, 98% of our AP customers had already adopted it. And they’ve gone from 50-60% auto-capture rates to over 90% and sometimes even 100%.
No more templates, no more manual data entry. Hundreds of hours saved for hundreds of businesses each year.
This is why we believe in focused use cases. Introducing specific AI features into the product for specific things. And in terms of results, the pattern is clear. Organizations that start with focused use cases, establish trust, and expand responsibly are seeing meaningful improvements in productivity and efficiency.
We don’t have to flip every single automated workflow to AI because some workflows are fine as they are. The major shift comes from progressively introducing AI and agentic automation at key points where it makes a real difference, progressively, to unlock a snowballing array of autonomous processes and strategic benefits.
How can Serrala help your organization adopt autonomous agentic workflows?
At Serrala, we’ve been helping CFOs and their teams automate finance operations for over 40 years. It’s our job to help you integrate new technologies within your organization in a way that makes sense both for your challenges and their capabilities.
This means our solutions empower your organization to start with the workflows you need today, then expand as your needs grow and your transformation plans mature.
To learn more about how Serrala can help your teams bring agentic AI workflows to bear in your finance function, get in touch today to book a demo.
