Which Back-Office Work Should You Automate With AI First?

What should you automate with AI first?

Start with one workflow that your team does every day, that follows a process people can describe, and that ends in something a person can check. The first project has one job: deliver a result the team notices, so the next project gets support.

That matters because most AI spending hasn't shown up in the numbers yet. In McKinsey's 2026 State of AI survey, 37% of respondents attribute at least some EBIT impact to AI use, about the same share as the year before.¹ A vague first project, such as "roll out an AI assistant to everyone," makes it hard to say what changed.

At Soren AI, a good first workflow has three things:

1. Return you can attribute directly to the work. Hours saved on a named task, a backlog cleared, faster turnaround on a specific request.
2. A clear definition of success. Everyone can tell whether it's working. Nothing ambiguous.
3. A win the team notices. Something that removes a task people dislike, which earns trust for what comes next.

Which back-office processes make good first AI projects?

The strongest candidates are repetitive tasks that still need some reading: too varied for a rigid script, too routine for your most experienced people. Here are the ones we see most often.

Intake is a common first build for law firms because every new matter starts there and the delay is easy to see. At one law firm we worked with, intake that used to take three days now takes about 20 minutes.

Which work should you not automate first?

Skip anything where a mistake is costly and hard to catch, where the process is different every time, or where you can't get to the data. Those can come later, once you have a working system and a team that trusts it.

• Final decisions with real consequences, such as approving a loan, accepting a client or sending advice. AI can prepare the work; a person should make the call. This is what people mean by human in the loop.
• Rare tasks. If it happens twice a quarter, the savings won't justify the build.
• Processes nobody owns or can describe. Fix the process first, or you'll automate the confusion.
• Work locked in systems you can't access yet. Getting access often takes longer than building the workflow.
• Tasks a simple rule already handles. If the steps never change, ordinary automation is cheaper and easier to audit than AI.

That last point applies to agents too. We explain when you need one and when you don't in our guide to what an AI agent is.

How do you score and rank candidate workflows?

List five to ten candidates, then rank them by expected return against the effort to build them. There's no fixed formula; the weights depend on your business. These six questions cover what matters.

1. How often does it happen? Daily or weekly work pays back faster than monthly work.
2. How long does each one take today? Time it on a few real examples, with the waiting included.
3. What does a mistake cost? Lower is better for a first project.
4. Can we get to the data? Note which systems it touches and whether you have access.
5. How will we know it worked? Write the success measure down before you start.
6. Where does a person review it? Name the step where someone approves the output.

The workflow that scores well on all six is your first project. Keep the runner-up list. It becomes the roadmap for the next six months.

Why do so many teams need to centralize information first?

Because AI can only work with what it can find, and most teams' knowledge is spread across inboxes, shared drives and people's heads. Atlassian's 2025 State of Teams research, a survey of 12,000 knowledge workers and 200 executives, found that leaders and teams waste 25% of their time just searching for answers.²

This is the most common surprise in our assessments. A team expects to start with automation and finds that the bigger win is one searchable place for its documents and data. That's why a company knowledge base is one of the two most common first workflows we build, alongside intake. It also makes every later workflow easier, because the information is already in one place.

How do you measure whether the first workflow worked?

Measure the workflow before you change it, then measure the same thing after. Few organizations do this. In the Thomson Reuters Institute's 2026 report, which surveyed more than 1,500 professionals in 27 countries, only 18% of respondents said they knew their organization was tracking the ROI of AI tools.³

• Turnaround time: how long from request to finished work.
• Hours spent: staff time per item or per week.
• Backlog: how many items are waiting at the end of each week.
• Error or rework rate: how often a person has to fix the output.

The companies getting the most from AI also change the work itself rather than adding a tool on top. In McKinsey's 2026 survey, nearly three-quarters of AI high performers (the 6% of respondents who attribute 5% or more of EBIT to AI and report significant value) said they had fundamentally redesigned workflows because of AI, compared with one-quarter of other respondents.¹

How does Soren pick the first workflow with a client?

We start by learning how the work actually runs. In the first two weeks our engineers meet the team, interview the people doing the work and map where time goes. By the end of week two there's a clear scope and goal, and we're already building.

The first workflow is low-hanging fruit with return you can attribute directly, and it's typically live within two to three weeks of getting access to the systems and information it needs. Our engineers stay after launch to support and improve it. More on how we work, or see what happens in an AI transformation assessment.