Hiring Enterprise AI Agents: 4 Attributes to Consider

The concept of hiring AI agents for enterprise use cases is quickly shifting from science fiction to operational reality. Companies are already prioritizing AI over traditional human roles in specific areas. Just look at this Hertwill job ad on LinkedIn.

OpenAI’s recent announcement of three enterprise AI agents with a price tag of up to $20,000 a month signals that major players firmly place their bets on this technology—and that enterprise AI agents could bring tremendous value.

Examples of successful enterprise AI agents

AI agents already bring value to enterprises with varied applications.

• Devin AI: An autonomous software engineering agent capable of independently planning and executing entire coding projects.
• Chatbase: An AI customer support agent that goes beyond simple chatbots by taking actions on behalf of users, such as changing passwords and processing refunds. Clients using Chatbase have reported a 65% reduction in support tickets after implementation.
• Salesforce Agentforce: A suite of autonomous AI agents designed to handle tasks across service, sales, marketing, and commerce, bringing benefits like 40-50% faster case resolutions.
• Rossum: An AI agent for paperwork handling the complete processing of transactional documents like invoices and purchase orders, from reading documents to inferring new data and finally performing actions in enterprise ERPs. Implementing Rossum can cut document processing time by up to 90%.

What defines an AI agent?

Before we jump into the hiring specifics, let’s define AI agents.

If you read the above examples again, you will see one common trait: AI agents independently take action and perform the whole job to be done from start to finish.

The simplest definition we can provide is:

“An AI agent is a virtual colleague you can delegate to, rather than a tool that you just use.”

That’s it. The differentiating characteristics of AI agents compared to software tools (even the AI-augmented ones) are autonomy and proactiveness. They suggest a course of action and follow it in the best possible way, given the situation and context.

4 attributes of enterprise AI agents

As hiring AI agents is slowly becoming the norm, you might be asking yourself: what should I look for in an AI agent? Surprisingly—or not so much—it’s the same things you’d look for in a human candidate, with a few nuances.

• Experience: Having a proven track record of performing the job at hand.
• Skills: Having all the specialized skills necessary to do the job and using them when required.
• Self-awareness: Understanding the limitations of their knowledge and having a confidence gauge that makes them question their work and ask for approval or assistance.
• Corporate knowledge: Ability to work with your systems, access your data, and follow operating procedures established in your business.

Summarizing these, the answer lies in specialization. In the enterprise context, there is no room for errors. So, while specialist AI agents can’t do everything, they are very good at specific tasks and jobs, like your most effective employees, who get the job done reliably and on schedule. 

AI agents vs humans

AI agents bring a host of advantages that make them especially valuable in an enterprise setting:

• Work 24/7
• Do not get tired
• No holidays / sick days
• No slow Mondays or hungover Fridays
• Don’t forget things
• Learn instantly
• Follow protocols precisely every time
• Process information instantly

Besides all of that, they are trained on massive amounts of information that a human could never manage. It’s quite an attractive value proposition.

However, at this point of AI agent maturity, humans must guide and oversee the agent’s activity. Even if AI agents can handle 90% of cases offloaded to them, the remaining 10% still need to be taken care of.

We can, therefore, look at AI agents as tremendous productivity boosters capable of replacing humans in repetitive tasks.

Hiring enterprise AI agents for transactional paperwork

“The purchase-to-pay (P2P) process is an ideal starting point for implementing new technologies, which can subsequently be scaled to address additional use cases across various departments.” 

Rene Kuijer, Director, PwC Netherlands

Transactional paperwork such as invoices, purchase orders, bills of lading, proof of delivery, and others are an excellent fit for AI agents.

The reason is that processing these documents is a routine, often monotonous task that begs to be automated. However, some parts of the process are more complex and non-deterministic:

• Understanding payment terms and determining if a big enough payment discount is available.
• Reading handwritten notes on a purchase order and creating an order with correct data in the system.
• Reading a document in a foreign language, for example, in a shared service center setup.

However, even with deterministic processes like tax coding (which is, in fact, a decision tree), the issue is that not all data on the document is available to make those decisions, so the process can be very time-consuming and error-prone, with big compliance risks.

While RPA tools, IDP solutions, and AI-powered OCR exist, they can’t automate complex tasks like invoice tax coding; AI agents can. What separates AI agents from other solutions is their reasoning capabilities and the capacity to take action.

Let’s use our four-part AI-agent hiring framework to see what the best AI agents for paperwork should look like.

Experience

What you find in a typical job ad for an AP specialist is 2-5 years of experience. An AI agent can beat that easily:

• Has seen millions of transactional documents in multiple languages
• Can instantly recognize fraud and anomalies
• Knows what an early payment discount is and why it is important

The key to an experienced AI agent is the quality of its training. As an AI system, it needs to be trained on a large quantity of high-quality data by highly proficient specialists in multiple iterations and learning cycles.

When you are selecting an AI agent for paperwork, you should check the following:

• Does it work with my specific document types?
• Has it been tested in enterprise settings at companies similar to mine?

You want to evaluate both the underlying technology and the implementation success stories.

Skills

AI agents for paperwork need to have all the skills that a human agent would:

• Understand documents in any format or layout
• Understand and work with enterprise master data
• Perform calculations, transformations, 3-way matching, and more
• Communicate with your suppliers
• Execute actions in your ERP
• Involve different stakeholders and request approvals

The biggest advantage of AI agents here is their speed of execution. While humans are arguably better at complex reasoning tasks that require wide business context, AI agents are catching up quickly—after all, reasoning skills are one of the important components that make AI agents different from other solutions, as I discussed above.

Self-awareness

The problem with generic LLMs and generalist agents built on top of them is the desire to please the asker and overconfidence, which results in hallucinations. In a way, their vast knowledge is a curse. Combine that with a lack of proper guardrails, and you have a recipe for disaster.

The solution to these issues is confidence scoring. For each decision that the agent makes, it estimates its confidence, and if it is below the minimum set amount, it escalates for human review.

The great thing about AI agents is that they learn from human feedback, and their confidence grows over time. They become more autonomous and reliable.

Corporate knowledge and ways of working

An AI agent’s experience might mean that it has been pre-trained, but you have your ways of working, your master data, and your systems that the agent needs to work with.

This is an important part of providing guardrails, business context, and your SOPs to the AI agent.

Here is what it looks like in practice for paperwork jobs:

1. Provide the AI agent access to your master data and lists of values.
2. Give the AI agent your standard operating procedures: what actions you want taken on your documents and other rules to ensure compliance and precision.
3. Provide access to corporate tools, like your ERP system and email.

So, just like your best employees, AI agents will understand the task context, access corporate knowledge and tools, and execute appropriate actions.

AI agents for paperwork need to understand the context, follow the process, and execute actions

Rossum: enterprise AI agents for transactional paperwork

At Rossum, we have built a platform for business users to configure and deploy specialist AI agents that automate complex paperwork tasks, saving teams massive amounts of time and reducing errors in data, while keeping the process running smoothly.

Here are some examples:

• Prioritize payment on the invoice with an early payment discount higher than a set amount.
• Involve a manager from a different team to approve payment with a high total amount.
• Calculate the total amount of goods from complex item descriptions and provide it to your ERP.
• Work with your master data, GL data, and reason over document data and select the correct tax code.
• Provide you with custom always up-to-date reports about your transactions and operational metrics.

Learn more and sign up for our waitlist here.