How to pick the first workflows to automate with AI
The ops team called us after six months with their AI rollout. Meeting summaries, board deck drafts, expense reports - all automated. Nothing in the P&L had moved. We had seen it before.
Ease of automation and business value pull in opposite directions, so the workflow that is simplest to automate is usually the one worth automating least.
Short answer: Most companies automate what is easy: report generation, meeting summaries, data entry. The research is clear that redesigning a workflow moves EBIT, and task automation rarely does. The right first workflows have high transaction volume, a direct link to revenue or retention, and a human bottleneck that slows a decision. Leverage is the filter.
Why do companies automate the wrong workflows first?
Companies pick the easiest workflow because two forces push them there, and neither has anything to do with value.
The first is who asks. The people closest to a workflow's friction get to nominate it for automation, and they nominate what annoys them. Expense reports. Status updates. The Tuesday metrics email. These are real irritations. They are also low-stakes, internal, and already serviceable without AI, which is exactly the profile of a workflow that returns little when you automate it.
The second force is the sales cycle. Vendors need a demo that lands in week three, so they steer you toward something clean and rule-based that will visibly work. A program that needs an early win picks the use case that produces one. That use case is almost never the one that matters.
The numbers say the default is failing. BCG found that 74% of companies have struggled to scale value from AI, and Gartner reports that only 28% of AI use cases in infrastructure and operations fully meet their ROI expectations. Most programs are not short on AI. They are pointed at the wrong work.
What actually drives the value
The value comes from redesigning how the work is structured. Bolting AI onto a task that already runs fine does almost nothing. This is the finding that separates the companies seeing EBIT impact from the ones running pilots that never scale.
According to McKinsey's State of AI, only about a fifth of organizations using gen AI have redesigned any workflow at all. The high performers behave differently. McKinsey finds they are 2.8 times more likely to have fundamentally redesigned their workflows, 55% of them against 20% of everyone else.
MIT Sloan's Ravin Jesuthasan makes the mechanism plain: the biggest productivity gains come from rethinking how tasks are sequenced, grouped, and handed off. Speeding up one task in place leaves most of the value on the table. Automate a task inside a broken process and you get a faster broken process. Redesign the process and the gain compounds.
This is why the easy workflows disappoint. A clean, rule-based process is already optimized. There is little left to redesign, so AI shaves minutes off something that was never the constraint. The workflows with room to move are messier, slower, and tied to a decision someone is waiting on.
Where is the value concentrated?
The value sits in core business functions. BCG puts a number on it: 62% of AI's value lies in functions like sales, marketing, manufacturing, supply chain, and pricing, and a year later BCG raised that estimate to roughly 70% concentrated in core functions. The same report found AI leaders growing revenue at double the rate of laggards and capturing 40% more cost savings.
The back office is where most programs start. The core is where the money is. Three workflow types account for most of the high-leverage opportunity.
Customer response and qualification at volume
The first type is anything that handles customer interactions at high volume with a clock running. Inbound inquiry routing, service triage, lead qualification. Every hour a qualified lead waits is an hour a competitor can answer first, so delay here has a direct dollar value.
The selling-time problem makes the case. Salesforce found that reps spend the majority of their time on non-selling tasks, and that sales teams using AI are 1.3 times more likely to report revenue growth, 83% with AI against 66% without. Triage and qualification are where that time leaks out.
Revenue-generating knowledge work
The second type is the expert work that sits directly on the revenue line. Prospecting research, proposal and RFP drafting, credit and risk memos, pricing analysis. The output decides whether a deal moves, and a skilled person is the bottleneck.
McKinsey reports a bank where AI-drafted credit memos lifted revenue per relationship manager by 20%. The same work shows agentic AI accelerating marketing campaign creation by 10 to 15 times. The pattern repeats wherever a senior person spends hours drafting before they can sell.
Operational decision workflows with high data volume
The third type is the decision workflow buried in more data than a person can hold. Demand forecasting, supply chain exception detection, capacity planning inputs. A human sifts the data, spots the exceptions, and feeds a decision that other things wait on.
BCG names this as one of its primary value domains, and the scarcity is the opportunity. McKinsey found only 3% of companies had scaled a gen AI use case in operations. The work is hard, which is why few have done it and why the advantage holds.
Which workflow comes first? The decision filter
A workflow is a first candidate when it has high volume, a human bottleneck, a revenue or margin link, and output that varies in quality. The wrong candidates have the opposite profile. Run yours through this before you green-light anything.
| Trait | Automate first (high leverage) | Avoid first (low leverage) |
|---|---|---|
| Volume | High transaction or interaction count | Low frequency, one-off |
| Bottleneck | A person slows a downstream decision or customer action | No one is waiting on it |
| Business link | Direct line to revenue, retention, or margin | Internal audience only |
| Current output | Inconsistent in quality | Already serviceable without AI |
| Data shape | Messy, needs judgment, room to redesign | Already structured, simple transformation |
If a workflow lands in the right column on volume, bottleneck, and business link, it belongs on the shortlist. If it sits in the left column on most rows, it is the comfortable choice, and comfortable is what the data says fails.
Want the next article when it's out? Sign up for the newsletter to stay current on making operations AI-native.
How do you decide between two strong candidates?
When two workflows both clear the filter, pick the one where the bottleneck is most expensive and the output most repeatable. A workflow with thousands of monthly interactions and a person gating each one will return more than a high-stakes process that runs twice a quarter.
Volume and repeatability matter because AI earns its keep on the hundredth instance. The first run rarely pays for itself. A twice-a-quarter board decision is high stakes and low volume, so a human handles it well enough and there is no leverage to capture. A thousand inbound leads a week is high volume with a person stuck in the middle, and that is where redesign pays. Start where the math is biggest and most predictable, then move down the list.
Frequently asked questions
What types of workflows are easiest to automate with AI first? The easiest are clean, rule-based tasks on structured data: report generation, data entry, meeting summaries. They are also the lowest-leverage, because a process that is already orderly has little left to redesign. Easy and valuable rarely sit in the same workflow, so ease is the wrong thing to optimize for.
How do you know if a workflow is ready for AI automation? Check four traits. High transaction volume, a human bottleneck that delays a decision or a customer, a direct link to revenue or margin, and output that varies in quality today. A workflow with all four is ready and worth it. One missing trait is a caution. Several missing is a pass.
How long does it take to see ROI from AI workflow automation? It varies, and many programs never get there. Gartner found only 28% of operations use cases fully meet ROI expectations. The ones that succeed redesign a core-function workflow rather than automate a back-office task. Picking the right workflow does more for your timeline than any tool choice.
Should you automate customer-facing or internal workflows first? Lead with customer-facing work when it carries a revenue or retention link. BCG places roughly 70% of AI's value in core functions like sales, marketing, and supply chain. Internal back-office workflows are easier to start with and that is the trap, because they tend to be low stakes and already serviceable.
What is the difference between automating a workflow and redesigning it with AI? Automating speeds up a task inside the existing process. Redesigning changes how the work is sequenced, grouped, and handed off. MIT Sloan finds the redesign is where the productivity gain lives. Automate a broken process and you get a faster broken process. Redesign it and the gain compounds.
Where to start this week
Pull your three highest-volume workflows that touch a customer or a dollar, and run each through the filter above. Your first build is the one with the most expensive bottleneck and the most repeatable output. Skip the comfortable choice. Pick the workflow a person is stuck inside while revenue waits.
We write about making operations, sales, and marketing AI-native. No tutorials, no hype. Join the newsletter to read the next one when it's out. Or just the newsletter signup form.
Sources
- McKinsey, "The State of AI" (March 2025). mckinsey.com
- BCG, "Where's the Value in AI?" (October 2024). bcg.com
- BCG, "AI Leaders Outpace Laggards in Revenue Growth and Cost Savings" (September 2025). bcg.com
- Gartner, "AI Projects in Infrastructure and Operations Stall Ahead of Meaningful ROI Returns" (April 2026). gartner.com
- Salesforce, "State of Sales" (2024). salesforce.com
- McKinsey, "Reinventing marketing workflows with agentic AI" (2024-2025). mckinsey.com
- MIT Sloan Management Review, Ravin Jesuthasan, "Want AI-Driven Productivity? Redesign Work" (2024-2025). sloanreview.mit.edu
- McKinsey, "From promising to productive: real results from gen AI in services" (August 2024). mckinsey.com
By Christopher Kliebenstein. We build and run AI-native workflows for operators who want results, not demos.