Boomi CEO anticipates rise of the AI agent economy

Computer Weekly: Tell us more about Boomi’s footprint and growth in APAC.

Steve Lucas: APAC is rapidly becoming the critical geography for Boomi to win. We spent the first 15 years of our existence growing in North America and Europe, but our growth in APAC, which accounts for over 10% of Boomi’s total revenue, has been massive.

At the macro level, even though inflation has been a challenge for near every country, there are countries and regions that have done a better job taming it than European and North American markets. The combination of continued growth and reasonably well-controlled inflation makes the APAC market incredibly compelling for Boomi.

We’ve invested broadly, not just in Singapore but also in India, where we have about 500 people. Singapore, in particular, is where we base our Asia operations and it’s going to continue to be the case for some time.

Are there specific industries that you’re targeting?

Lucas: It’s banking, followed by manufacturing. The concentration of banks in major financial centres like Singapore and Tokyo makes banking a prime opportunity. Banks, partially due to regulation, have been a little slower to adopt integration technologies over the past decade, so they’re just coming to the point where the right platform with the right level of certification and compliance meets their needs. It’s the perfect storm of growth.

If you look across the broader APAC region, AI has cast a big shadow in a good way. There really isn’t a CIO who can ignore AI anymore. That’s where Boomi comes in to not just do the day-to-day things that organisations need, like integration and automation, but also to give businesses the ability to embrace and build an AI strategy.

Can you dive deeper into how you’re plugging Boomi’s capabilities into your customers’ AI strategy?

Lucas: When the iPhone came out, there wasn’t anybody in the press that didn’t go a day without writing the words app economy. Fast forward to 2024, we’re not really talking about apps anymore. We’re talking about agents, specifically AI agents, and if you check out Hugging Face today, there are over half a million models. The marketplace for AI models and agents has just blown up in a big way.

Two years from now, there will be potentially tens of millions of models, specialised by industry. We’re going to enter this new market called the agent economy. There will be agents for everything, hyper-personalised to you or for an industry.

If you can take the emotion out of a corporate buying decision, you’d no longer have a product that sits on the shelf and never sells. It's purely data-driven. It's fully automated, and solely focused on products that serve you the best and move the quickest. Not how this made me feel as a child and whether we should put it on our shelves

How can I get this LLM and agent for customer service, finance or HR teams? How do I deploy and customise the agents so they’re specific to my business? I don’t want to load my data into a public model that my competitor can take advantage of. That’s how they’re going to build the next generation of applications, reduce cost and transform their business.

How Boomi helps in the agent economy is, first and foremost, we can help organisations get the right information they need and load that into whatever model they want. We can help them get data into the model to finetune it and train it for their business. We have thousands of customers who are using Boomi for things like retrieval-augmented generation and loading data into LLMs and small language models.

Even though models have changed with pre-trained transformers, what hasn’t changed is garbage in, garbage out. If you don’t train the model with the right data and if you don’t finetune it the right way, it doesn’t matter. And so, we’re going to give our customers a pre-selected set of models that they can use with Boomi, load data into the models and finetune them for their business in record time.

That’s interesting because it can be challenging to identify the right model to use, so having a pre-selected set of models is useful. Are you planning to charge more for these capabilities, given the high cost of model-training?

Lucas: In the thousands of scenarios that customers are using Boomi for to move data into models, clean the data, qualify it, improve it, that’s out of the box capability we have. Our customers are using us to connect to data sources, vector databases, LLMs and build workflows. For example, if they get a lead that comes in from Salesforce, they use Boomi to reach out to ChatGPT, which writes a custom email for the salesperson, saving the salesperson an hour of time every day. That’s a trivial but powerful example.

That said, we know most companies don’t have the resources of a Walmart, so when you’re talking about a mid-sized organisation that wants to customise or train a model, that’s where Boomi comes in. We want to make it achievable and affordable.

What’s going to happen is a couple of things. First, Boomi is going to work with any model and we will provide support for those models. Some models will be open source, which are readily available, but as you pointed out, it’s not running but training the model that’s pricey because you need a GPU (graphics processing unit) farm.

Meanwhile, there initiatives that are looking at new ways to train models at a radically lower cost. Cloud providers are also offering GPU-as-a-service that’s going to drive costs down and Boomi’s job is to provide different options. With GPU-as-a-service, our data integration technology, and a set of recommended models, we can bring all that together to drive down costs.

And some of these models might not be LLMs, because you’re dealing with very specific domains, right?

Lucas: The reality is, if you take a retail organisation, it doesn’t need to crawl the whole internet. You’re talking about using a fraction or sliver of data for finetuning, or what we call grounding a model.

It’s generally affordable but here’s the thing: a company with $500m to $1bn in revenue will have a sizable IT budget, but do they have the data lake? Do they have the model selected? Do they have the right data integration tool? Do they have the right people who know how to put all this together? Generally, no. We’re bringing a solution to the table where you can just use Boomi, pick your data source, select the model and click train. That’s something I expect us to be announcing in May.

How are you positioning Boomi vis-a-vis other players in the market? I’ve been hearing similar messages from your competition as well.

Lucas: I would look at what those companies have put in the market. Last year, we released Boomi GPT, which reduces the amount of time to create application integration processes by over 90%. You can literally type in English to connect your Salesforce system to Workday. It automatically connects, designs and loads the processes for you.

Our AI engine, Boomi AI, can also write documentation for business process integrations using a pre-trained transformer. We also have a new self-healing capability that monitors if a process has changed, and if so, we can fix it without you knowing anything about it.

We’re getting ready to announce, in May, a PII (personally identifiable information) capability that monitors data streams for PII data, and if so, tells you where that data is, so you’ll know if you’ve violated the GDPR (General Data Protection Regulation). We’re also going to announce Boomi Process Automation where Boomi can literally monitor your work and automatically build a dashboard for you based on what you’re working on.

Two years from now, there will be potentially tens of millions of models, specialised by industry. We're going to enter this new market called the agent economy. There will be agents for everything, hyper-personalised to you or for an industry

With Boomi AI, we’re going to be providing everything from integration design to data management services, all the way to process automation. Our competitors will say things, but what are they delivering? It’s one thing to make announcements, and it’s another to deliver.

We have something unique that our competitors do not have. We have hundreds of millions of design patterns that we’ve collected over a decade. So, if a company builds an employee onboarding process using Boomi, we anonymise that but we know which systems and how they are integrated. We’ve trained our own LLM on how these processes are designed, so you can literally say ‘create an employee onboarding process’ and it knows what you mean, which systems are commonly used, which ones you have and designs it for you.

I can’t even tell you the inordinate amount of time we spent training this LLM using that dataset to design the right outcomes. The interesting part is some of our competitors rely on things like recipes, which are written by someone who thought they might taste good. Ours is based on what organisations have done with our integration technology, and being able to come up with an infinite number of potential recipes based on reality.

Is there anything else in the pipeline you’re looking at?

Lucas: We’re going to be launching what we call our AI agent architecture. With the agent economy that I talked about earlier, you’re going to have hundreds or thousands of AI agents running different processes inside your organisation. You will want to know what they’re doing and their decisions have to be explainable. If an agent makes a hiring decision, you need to be able to explain that for compliance purposes. With our AI agent architecture, we’ll be aware of these agents and help organisations manage them.

That’s going to take years to build and we will also have our own agents – the design agent, the PII agent and the automation agent, among others. We’re going to work with third-parties and plug their agents into our architecture as well.

Would you say that you’re entering a new market that’s not iPaaS?

Lucas: I don’t know the right word or acronym for it yet, but I know it’s more than iPaaS. I know it’s more than integration. I think it’s in that realm of orchestration. What I’ve settled on, for now, is that all organisations need to have that AI-driven, always-on integration platform. It’s more than what we are today.

I was speaking to the CTO of a large retail chain and one of the things he shared with me was that they are building an agent that can act as a corporate buyer who buys things that go on their shelves.

Think about it – if you can take the emotion out of a corporate buying decision, you’d no longer have a product that sits on the shelf and never sells. It’s purely data-driven. It’s fully automated, and solely focused on products that serve you the best and move the quickest. Not how this made me feel as a child and whether we should put it on our shelves. That level of optimisation is impossible without AI, even in a data-driven world and the right level of analytics.

All we want to do at Boomi is to enable companies that don’t have the resources of Amazon, Alibaba or Walmart to get to that degree of capability without the need to hire 1,000 machine learning engineers. We can do this and we can do it now.