What you need to know while hiring an AI automation agency

Are you facing operational bottlenecks and high error rates? Do you waste money because of slow business procedures? Did you reduce the team size, but they must handle the same workload? Designing business processes and automating at least some of them is the only efficacious solution.

Thanks to advancements in AI technology, we can now build more advanced automation than ever. This is where AI automation agencies enter the stage.

Currently, most AI automation efforts involve either building chatbots or internal tools. Customer-facing chatbots are risky. DPD deployed a chatbot on their website, and the bot called them the worst delivery company. That’s why I focus on building internal tools to speed up the back office operations and make them cheaper. Of course, you may still use a chatbot but don’t make the bot public. I will give you a few examples of such automation in this article.

What is an AI Automation Agency?

The AI Automation agency creates custom AI solutions for businesses. An AI automation agency is a software house, right? Yes and no. A software house typically builds programs based strictly on your specifications. In contrast, an AI automation agency not only develops software but also helps you document and refine your business processes, identifying the best opportunities for automation. And, of course, they will use AI when necessary.

Defining AI Automation

We have used automation for years, such as robotic process automation, automated workflow, etc. Actually, any software is an automation. What’s the difference between that old-school automation and AI automation?

When we use AI, we can perform more advanced tasks. Previously, it was impossible to build an automation that extracted a relevant part of the text from a document unless the extracted part had a predefined structure like an email address, zip code, phone number, etc. Now, we use large language models (LLM), a type of AI, and LLMs no longer have such limitations.

For example, we could classify the customer’s opinion as positive or negative using old-school machine learning technology. Using AI, we can classify the opinion and find the quote explaining why it was classified in such a way. Now, we not only know the opinion was negative but also what the customer complained about.

Not only do we get more actionable information, but we can use the explanation to perfect the automation over time because we know not only what the automation did but also why it did it.

AI isn’t limited to just working with text. We can even automate low-stake, reversible decisions like routing emails to the correct person, preparing offer drafts, finding similar support cases from the past to help solve the current issue, or extracting data from multiple sources to build a customized report.

For now, we must keep a human in the loop, so I wouldn’t use AI to make hiring decisions (the EU AI Act makes such automation illegal anyway), send offers to clients, or make purchases on the company’s behalf.

To ensure the automation addresses your business’s unique needs, you should designate an employee who will work with the AI automation consultant along the way and be available to answer questions, explain the business process, and delegate decisions to the right people in your organization.

The Role of Agencies in AI Automation

Why do you need an agency? Can’t your IT team do it all? I bet they can. The question is how much time they need if they have to build automation and learn about automation at the same time.

Combining your team with an AI automation agency is the most effective setup. The agency builds the automation and teaches your team how to operate it so you can make changes on your own in the future. Meanwhile, your team ensures that the automation integrates well with your existing software. This way, everyone will do what they are best at, and we will get faster results.

“Who works like this?” you may ask. I do. My primary goal is to teach the client’s team how to use AI so they don’t have to call me when they need minor adjustments. I really believe in true consulting, not “consulting” being a code word for glorified outsourcing.

The Benefits of Using an AI Automation Agency

All of the knowledge about AI is on the Internet, and you can find it for free, so why should you consider hiring an AI automation consultant or agency?

Increasing Efficiency in Business Processes

An AI consultant has done it before. They have a set of favorite tools they use to get results fast. While sticking to a predefined toolset may be limiting, it’s all code. We can change it, glue different parts together, and use our code in between.

Having a set of favorite tools and processes means the agency has experience using them and knows what can go wrong and how to deal with them. Your current IT team doesn’t always debate whether they should follow Scrum or Kanban processes or use Java with the Spring Framework or PHP with Laravel. They know what worked before and how to solve problems with their tools.

How can we tell if we achieved the goal? Along the way, we have to set up a monitoring infrastructure to determine how much time the process takes and how much time the automation saves. Monitoring may be bitter because you may learn that something seemingly simple takes a week and involves eight people. Be prepared.

Reducing Errors and Improving Accuracy

Building an AI automation isn’t finished when you run it for the first time. Due to the probabilistic nature of AI and machine learning models, it’s hard to build a solution that never makes a mistake.

AI automation agencies take an iterative approach. First, we build a proof of concept to handle most cases correctly. Then, we analyze what errors AI makes, determine why AI makes those mistakes, and adjust the code.

Let’s talk about the elephant in the room. 100% accuracy is achievable only in the simplest automations. AI hasn’t reached a human level yet; even if it did, humans make mistakes, too.

An AI automation agency deals with this uncertainty by constraining what AI can do and asking a competent human operator to approve any critical decision. The AI system may also gather feedback from your employees, so you can use the feedback as training data for building the subsequent versions of the automation.

Streamlining Repetitive Tasks

The most significant money saving comes from saving time spent on repetitive tasks. Do your employees need to find information in a document directory when preparing an offer? What happens when they do it? They have looked at those documents for 3 hours a day for the last five years. First, they take a break and make tea because the task will be tedious, so they want to postpone the work. Then, they mindlessly search the documents while thinking of something else. Anything. Lunch. Renovating their home. Asking out William/Emily from the department next door.

Should your best people spend time pressing CMD+F to open a search window and repeatedly typing the exact text? AI can do this.

Do they need to copy-paste parts of the product description from the internal documents into the offer sent to the client? AI can prepare a draft for them. Let your team focus on what makes the offer unique and tailored for the client.

The Role of AI Automation in Back-office Operations

What can AI do for you? Everyone is building chatbots. While chatbots are a way to deliver automation to the people using them, the automation may do much more than chatting.

Reducing Cost of Business Operations through Retrieval Augmented Generation

Retrieval Augmented Generation is a fancy way of saying that we use AI to query a database, search documents, or request data from other software integrated with the AI systems. Then, we ask AI to consolidate the retrieved data and write a response to a person using the system.

What can RAG do? If you sell manufacturing equipment, automation may find relevant information in the technical specifications and present the data to the salespeople preparing the offer. In the financial industry, automation may find relevant parts of the terms and conditions. In customer support, AI may look for similar support cases from the past and draft a solution suggestion based on the steps that worked in the past. Or even better, automatically classify the support case and draw the solution from a collection of your standard operating procedures.

Ensuring Regulatory Compliance with Automated Monitoring

We can also use AI after contacting the customers. If all phone calls are recorded, AI may analyze the transcripts to find what was said and determine whether the salespeople stick to the sales script, don’t provide financial advice if they are not allowed to, or handle customer objections in the way you trained them.

Leveraging AI Technology for KPI Tracking and Data Analysis

Do you need to bother a data analyst whenever you want to know something about your business? Did you ask the data engineers to build tons of dashboards, but there is no way to remember which one shows which data? AI can help with both problems.

Imagine an AI system where you type your question in a text box, and the AI queries all relevant databases and delivers a one-sentence-long answer a minute later. It is most likely at least 30 times faster than a data analyst. The analyst can focus on more complex problems and use human creativity instead of dealing with dull, one-off questions.

Key Services Offered by AI Automation Agencies

While the scope of possible solutions is infinite, as with any software, AI Automation agencies typically focus on those five things:

Business Process Automation with AI Decision Making

The simplest but often most beneficial thing an AI automation consultant can do is automate a scripted process. If a task must always happen in a certain way but requires making some decisions, let someone automate the process for you, use AI to make those decisions, and ask a human to review and approve the final result.

It’s not shiny. It doesn’t look cool in demos. Often, you can’t even tell if they used AI or not. But such automation makes money for your business by saving time and letting people do more work in a workday.

The problem is that you need a process. To automate a process, you need a process followed by the people doing the task manually. If everybody does the work slightly differently, we will have to come up with a unified process first. If that’s impossible, you have 2-3 processes covering 80% of all cases and can automate those.

Customer Support Automation

As mentioned earlier, a large part of customer support work consists of looking for a solution in a similar case from the past or your standard operating procedures. Nobody invents a new method for every customer’s email. At least, I hope they don’t do it.

I wouldn’t recommend sending the AI’s solution directly to the customer. (After all, social media is full of examples showing how AI makes mistakes.) However, customer support can use AI’s output as a starting point, saving them a little time.

Even if the automation cuts the average time required for handling a support case by only 10%, how many more can the same team handle without hiring and training new employees?

Data Analysis and Reporting

The other thing I mentioned earlier is asking a data analyst one-off questions about your business or department. Those short, simple questions throw them off focus and prevent them from doing more complex work. Imagine your analysts working without interruptions as an AI system swiftly retrieves answers from your databases.

You can even make the AI system available as a bot in Slack, MS Team, or whatever chat software you use. Now, you send your question to a bot instead of the data analyst, and in a minute or two, you get an actual answer instead of a response saying they will handle your question as soon as they finish working on the big thing they are doing right now.

Predictive Analysis

Machine learning prediction didn’t disappear when we started using Large Language Models. Now, they can be even better because we can automate data preparation.

What if your churn prediction system could use the information you have quantified in the database and data extracted from phone calls with the client or survey responses they sent?

What if the recommendation system could learn what is important to customers based on their reviews of products and services they purchased in the past?

Data Retrieval Automation as a Decision Making Assistant

Companies have terabytes of data with which they have no clue what to do. AI will not magically solve the problem. (Especially if you forgot what is in the data and what it represents.) However, you can use AI to query multiple data sources and prepare an actionable report to answer your questions.

It’s not just a report with facts, and it’s not just a question-answering tool. Such an automation incorporates your values in the answer it generates.

Imagine this: You have an automation that generates meeting notes from transcripts. Thousands of summarization tools exist, but yours doesn’t generate just another summary. Instead, AI finds the business objective, explains how the objective refers to your OKRs, and finds the action items and the people responsible for those actions. When you approve the meeting report, the automation adds the actions to to-do lists of appropriate people and generates an agenda for a follow-up meeting. It may even periodically ask people for updates on their assigned tasks.

Steps to Engage with an AI Automation Agency

AI Automation agencies are not Amazon. You won’t go to their website, add an AI system to a shopping cart, pay, and get a working application the next business day. They are more like a mix of a standard software house and a business consulting company. Of course, they will not tell you how to run your business, but they need to understand how you do it to build the right automation.

That’s why a typical engagement with an AI automation consultant looks like this:

Initial Consultation and Needs Assessment

During the initial consultation, the consultant will assess what you are looking for and whether they can help you. A good consultant should be able to explain a high-level overview and provide some examples. The consultant may send you links to show what is possible.

If this turns into a sales pitch where they tell you they can do everything you need, you are in the wrong hands. They have no way of knowing what they must build yet.

Only two outcomes of the initial consultation are possible: “Maybe we can work together” or “No, we can’t do that.” I talked a few people out of using AI for their workflow because they would be disappointed and had to spend too much time manually fixing AI’s decisions. Mostly because they had no real business process, and everyone working on the task did whatever “felt right” at the time. You can’t automate chaos.

Understanding Workflows and Tasks

During the second conversation, the consultant should ask questions about your workflow and the business process you automate. You could go through an example case together during the call. You will also discuss the data sources you access while working on the process.

This time, the consultant should devise a way to automate some parts of the process. The actual implementation will likely require iterative development in collaboration with someone from your organization. This meeting is the time to discuss the rules of the cooperation.

It would be best to involve the people who will use the automation. They understand the workflow and can explain what happens, but they also need to learn the benefits of automation. Being involved from the beginning is an excellent way to see the benefits.

Proposal and Solutions Design

The consultant should come up with a demo of the proof of concept. The initial version may work on a small subset of the data but should more or less explain how the entire solution works. They should also elaborate on how the complete solution will look like and what work is required.

You should discuss the consultant’s work and the work required from your team responsible for integrating the AI system with your applications.

Implementation and Integration with Existing Systems

The integration stage will likely involve several iterations of changes because people always discover edge cases they forgot to mention earlier. Similarly, the implementation may be split into milestones where you begin with an AI system handling the most common case and add additional capabilities later.

Adopting an agile approach and starting with a Minimum Viable Product (MVP) is crucial. The MVP allows users to begin using the tool quickly and provide valuable feedback, essential for refining the final product. Otherwise, you will build something that slows your business down and annoys your employees.

Training and Change Management

An unused AI system is a waste of money. Along the way, the consultant should train your team to operate the solution you build. If the system is integrated into your existing application, it’s often better when the actual training is performed by your team but with the consultant’s help. The consultant may handle the entire training if you build a new application.

Change management, however, is entirely your responsibility. An external consultant cannot force people to change their habits. That’s why I suggested involving the final users early in the process.

Ongoing Support and Optimization

AI isn’t perfect. You will want to handle new cases, change the current behavior, or include additional steps in the workflow. Involving the consultant in every such change makes no sense. I like to train the developers along the way, so you can ask your team to handle the changes.

Of course, the consultant may be always available to answer their questions if you decide that’s the engagement model you want to pay for. You may also start a subsequent project where the consultant builds new features on top of the existing AI system.

Future Trends in AI Business Automation

According to Jason Liu, AI automation will move from Retrieval Augmented Generation and question answering into custom report generation tools that fill out templates using AI and data requested from other systems. Jason thinks we will move from using AI as a more advanced search engine to a decision-making tool that can incorporate your values while making the decision.

At the same time, Andrew Ng thinks agentic workflows will be the future of AI. Agentic workflows allow AI to independently perform tasks and modify data. We will get used to delegating tasks to AI and patiently waiting for a response.

Both trends are exciting because we can automate time-consuming tasks and make business processes more predictable, efficient, and cheaper.

Do you need help building AI-powered business process automations for your business?
You can hire me!