Skip to content
All guides

Write down how the work really gets done. Then let an agent try it.

Bain's Technology Report 2026 says the biggest AI agent opportunity at large companies sits in the work between apps, and that unwritten know-how is what holds it back. How a small business can start with one workflow this week.

What Bain published

On September 29, 2026, Bain & Company published its Technology Report 2026. One chapter, "The $100-Billion SaaS Opportunity Hiding in Cross-System Labor", looks at the work people do between business apps: pulling a number from one system, checking it in another, reading an email and deciding what happens next.

Bain wrote it for software companies and investors, and its numbers describe large US companies. We read it for a simpler reason. Almost every small business runs on this kind of in-between work.

What "work between apps" means

Bain calls it cross-system coordination work. Its examples include pulling budget data from one system, "checking inventory in a spreadsheet, interpreting free-text responses," and deciding when to escalate (p.24).

In a small business it looks like this:

  • Copying a new order from an email into the invoicing tool
  • Checking the calendar before confirming a booking
  • Chasing an unpaid invoice, then updating the customer record
  • Matching a supplier bill to what was actually delivered

Bain says this work has been "unable to be automated by rules-based software or traditional robotic process automation (RPA)" (p.24). Fixed rules break when information is scattered across apps or when a message is ambiguous. AI agents can read across sources and use judgment within limits you set, which is why Bain sees a large new market here. It estimates about $100 billion in the US, with more than 90% of that opportunity still untapped (p.25).

The catch: the know-how lives in people's heads

Bain lists six factors that decide how much of a workflow an agent can realistically take on. The one it singles out is written-down knowledge. Its examples of what tends to be missing include "a veteran buyer's vendor preferences" and "the unwritten exception logic that governs how a policy is actually applied" (p.27).

Bain's conclusion: "the availability of this digitized contextual knowledge is almost always the binding constraint on automation" (p.27).

In plain words, an agent can't follow a rule nobody wrote down.

How much is automatable, at large companies

Bain's estimates of the share of workflow tasks that agents could automate, by function (p.28):

  • Customer support and engineering: roughly 40% to 60%
  • Finance and HR: 35% to 45%
  • Sales and IT: 30% to 40%
  • Legal: 20% to 30%, held down by the cost of mistakes

These are Bain's estimates for large companies, not measurements of small businesses. Bain ties the range to factors such as whether the result can be checked, what a mistake costs, how many systems are involved and how often the work breaks pattern (pp.26 to 28).

Why it matters to a small business

In a small team, this knowledge often sits with one or two people who have always done the work. They know which customer always pays late, which supplier sends the wrong invoice format, and when a refund needs a second opinion. None of that is in the software.

That makes the first step for an AI agent cheap and unglamorous: write the workflow down the way your best person actually does it, exceptions included. Bain's chapter on AI-native companies makes the same point about effort, noting that companies Bain works with "have found that for every dollar spent on technology, four are spent on people and process" (p.51).

The practical part: four questions before you pick a workflow

These are adapted from Bain's six factors (pp.26 to 28) for a small business:

  • **Can you check the result?** An invoice that matches the order, or a booking that shows up on the calendar, is easy to verify. Start there.
  • **What does a mistake cost?** Pick work where an error is cheap and easy to undo. Keep anything involving payments or legal commitments with a person.
  • **How many apps are involved?** Two or three is a good first pilot. Bain notes that five or six systems are harder to coordinate reliably (p.28).
  • **How often does it break pattern?** Choose work with a few common exceptions, not constant surprises.

This week

  1. 01

    Pick one in-between task

    that eats time every week, such as matching supplier bills to orders.

  2. 02

    Sit with the person who does it

    and write down every step, including each "if this, then that" exception and who they ask when they are unsure.

  3. 03

    Mark the costly steps.

    Anywhere a mistake would lose money or upset a customer stays with a person.

  4. 04

    Let an AI tool or agent try the rest

    , with a person checking every result for the first two weeks.

  5. 05

    Keep a short log

    of what the agent got right, what you fixed and what you had to explain. Add each explanation to the written steps.

Our take

The best first AI agent project is not a new app. It is one workflow you can finally describe on paper.

No partnership with Bain & Company is implied here.

Sources

The $100-Billion SaaS Opportunity Hiding in Cross-System Labor (Bain & Company, David Crawford, Chris McLaughlin and Greg Fiore, Sep 29, 2026), Technology Report 2026, pp.23 to 28.

The AI-Native Enterprise: Absorption Is the New Advantage (Bain & Company, David Crawford, Anne Hoecker, Jue Wang and Chris McLaughlin, Sep 29, 2026), Technology Report 2026, p.51.

Know someone who should read this?

Send it to a colleague, or hand it to your AI assistant and let it tailor the advice to your business.

Send to a personHand it to your AI

Want this working in your business?

Tell us what you are trying to fix and we will point you at the smallest useful next step.

Talk to us