

Most AI business cases file faster decisions under soft benefits because nobody has measured them. In back-office workflows, that speed is mostly waiting time between hand-offs, and you can count it from the records you already have.
A LinkedIn post by Dr. Drijesh P. lays out a simple equation for enterprise AI returns: hard benefits plus soft benefits, minus total cost of ownership, divided by total investment. It puts labor reduction, process automation, and error and fraud reduction in the hard column. Faster decisions, improved customer experience and brand trust go in the soft column, which the post calls harder to quantify but strategically critical. The post also argues that the biggest returns often sit in back-office automation, ahead of the more visible front-end generative AI tools. I agree with that last point, and I think it changes how the soft benefits column should be filled in.
In a back-office workflow, a faster decision mostly means less waiting. A supplier nonconformance, a credit exception or a new product request moves through a sequence of people, and most of its elapsed time is spent sitting in someone's queue between steps. That time looks hard to quantify only because nobody records it. Once you count it, it moves from the soft column to the hard one, and the business case you bring to the CFO becomes much harder to dismiss.
I think the pain in most mid-market operations is visibility across hand-offs more than any missing feature. When a quality engineer finishes a review and emails it to an approver who is traveling, the request can sit for days while nobody downstream knows where it stands. The work itself may take an hour. The calendar time is set by the gaps between people, and by the follow-up emails and phone calls needed to find out what happened.
When Laminar built LeaseIQ for Healthpeak Properties, the lease approval, legal document review and tenant improvement workflows it replaced had been running on email, spreadsheets and phone calls. Work in that form leaves no dependable record of when a step started or who was holding it, and that is exactly the information an ROI model needs. Many processes that companies now want to hand to AI agents are in the same condition.
This matters for the equation because an AI agent mainly speeds up the work inside a step. It can draft a summary, check a package for missing fields or route a request in minutes. If the request then waits three days for a signature, the faster draft barely changes the total time. A business case that claims faster decisions without separating work time from wait time is guessing at the one number the CFO will question first.
Regulated companies have an extra step here, because some of the waiting is mandatory. A medical device design change has to be reviewed and approved under the company's change control procedure. A credit decision above a set threshold may need a second signer. A supplier issue may need quality sign-off before material is released. Those reviews stay in place after any automation project, and an honest model should not count their removal as a benefit.
What can shrink is the time a request spends waiting for someone to notice it, waiting while someone chases missing information, or going back around because the first package was incomplete. Sorting each wait into required review or avoidable delay is work your quality and compliance people should do with you, since they will ask the same question when the project comes up for review. Having the answer ready before the proposal reaches them saves a round of back-and-forth and gives them a reason to support the numbers.
The obvious objection is that the data is a mess. Timestamps live in email threads, the ERP, the quality system and someone's tracking spreadsheet, and none of them agree. Messy data is normal in a growing mid-market company, and I think it is better to plan around it than to treat it as a reason to skip the baseline. You do not need a complete history. Pull a sample of 30 to 50 recent items of one request type, reconstruct the dates between steps from whatever records exist, and mark where the trail goes cold.
Then sit with the people who handle those requests and ask where items wait longest and why. Their answers will usually line up with the gaps in the records. What you end up with is a range of days in each queue, with a clear note of how it was calculated. A range the CFO can trace back to real records is more credible than a precise soft-benefit estimate. It also lets finance attach a dollar value they agree with, whether that is inventory on hold, delayed revenue, late fees or staff hours spent chasing status.
This kind of work is usually capital spend the CFO approves, and neither operations nor IT tends to own that budget on its own. The post notes that returns often take two to four years to show up. A measured baseline gives you something to report in the first year, when the full payoff is still forming, because you can rerun the same measurement after launch and show whether the queues actually got shorter.
Dealing with a similar challenge?
We work with mid-market companies in regulated industries to build AI workflows that actually hold up.
Let's TalkSean Cummings
Founder of Laminar Consulting Services. Specializes in AI workflow automation for regulated industries — medical device, financial services, and complex logistics operations.


