Where the project and the business run on the same peopleConfidential by design · Anonymous by default
Resources›Automation and AI
Pillar Article
Updated October 2026
7 min read

Automation Projects Not Delivering Results: Where the Time Goes

Automation projects not delivering results often aren't failing inside any one step. The time is lost between them.

§ Automation

Angela has two things open on her screen. One is the spreadsheet Sam has kept since last winter's planning meeting, and every row on it now says Done. The other is this week's shipping report, which shows orders for stocked parts going out the same day about as often as they did a year ago.

That's the number the list was supposed to move. The company's distributors compare suppliers on how fast stock ships, and two of the larger ones have told Paul, who owns the company, that it isn't fast enough.

Angela runs operations. At the planning meeting, each department named the step that slowed it down most. Sam, who keeps the company's ERP running, built the automations with the company's RPA software and some help from the ERP partner. Customer service got an AI tool that reads purchase orders out of email and enters them in the ERP. Finance got a workflow that emails Hal, the controller, whenever the ERP puts an order on credit hold, so he can approve the release from his inbox. Shipping got a bot that books the carrier and prints the label as soon as an order is packed. Accounts payable got one that reads supplier invoices and matches them to purchase orders. Every project did what it was supposed to. Customer service hardly keys an order anymore, and neither does payables.

The number has barely moved.

Ten orders that missed the truck

Angela pulls ten orders from this week that didn't ship the day they came in. For each one, the email shows when it arrived and the carrier software shows when the label printed. The ERP doesn't record much of what happened in between, so she fills in the rest from the exceptions folder, the approval emails in Hal's inbox and a few questions to Gina and the warehouse.

The steps the automations took over are fast now. An order the tool can read is in the ERP within minutes, and once it's packed, the label prints in seconds. Three of the ten came in late in the afternoon, so they were never going to make the truck. The other seven arrived before noon and still missed it. Almost none of their lost time happened inside a step. It happened between steps, in three places.

The exceptions folder

The tool reads about four out of five purchase orders on its own. The fifth goes to an exceptions folder, usually because it has a handwritten note, a part number the tool doesn't recognize or a price that doesn't match.

Before the tool, customer service keyed every order first thing in the morning, odd ones included. Now nobody's morning starts with the folder. Gina, who leads customer service, checks it when the phones quiet down, which is usually after lunch. Four of Angela's seven orders sat there for more than three hours.

Gina's team got their mornings back. Nobody was asked to take on the folder in their place.

Waiting on Hal

The other three were on credit hold. The ERP has always held orders from customers who are over their limit or past due, and Hal is the only one who can release one.

Before the workflow, someone from customer service walked a held order over to his desk. Most holds cleared in a few minutes, often because the customer had already paid and the payment hadn't posted yet. Now each hold reaches Hal as an email asking him to approve it. It's one more message in an inbox that gets more than a hundred a day, and this is the week he's closing the month. Two of the three customers had already paid.

After two o'clock

The truck that takes same-day orders leaves at three. The warehouse's cutoff for same-day is two o'clock, which leaves an hour to pick, pack and stage the afternoon's orders, so anything that clears after two goes out the next day. A label that prints in seconds doesn't help an order that clears at two-thirty.

All seven of Angela's orders cleared after two, some by only twenty minutes. A few hours in the folder or in Hal's inbox doesn't look like much on its own. With a three o'clock truck, it's the difference between today and tomorrow.

Angela looks back at Sam's spreadsheet. Every project on it was proposed by one department for its own step. None of them covered the hours an order spends waiting between one department and the next.

At twenty to five on Friday, Hal gets to his inbox and approves eleven holds. The truck left at three, so all eleven will ship Monday. Two of them are for one of the distributors that called Paul.

When automation finishes and the number doesn't move

It's common for automation to make each step faster and leave the result about where it was. In a McKinsey survey published in March 2025, redesigning workflows had the biggest effect of 25 attributes on whether generative AI improved a company's earnings, yet only 21 percent of respondents whose organizations use generative AI said they had fundamentally redesigned at least some workflows. That survey measured generative AI only. In our experience, rule-based bots run into the same thing, since a step done faster only helps if that step was what held the work up.

There are a few common reasons an automation program stops short of the number it was meant to move, and a struggling one often has more than one. Where the time goes on a handful of late orders, invoices or tickets usually shows which. Most systems record when work arrives and when it's finished. The hours in between are often only partly recorded, which is why a delay between steps can go unseen for so long, and why piecing those hours together usually takes the people who handled the work.

Time lost between steps shows up as quick steps with long gaps between them. Work an automation can't handle lands in a folder or queue that isn't part of anyone's first task of the day. Approvals that used to happen in person can end up waiting in an inbox. A cutoff, like a truck, a bank deadline or a shift change, turns a few hours of waiting into a full day, so a gap costs the most when it runs up against one.

An automation that breaks as its inputs change looks different, because the trouble builds over time rather than showing up all at once. A customer changes its order format or a supplier changes its invoice layout, and the share of work going to exceptions creeps up. Sometimes people quietly go back to doing the step by hand. Since the work still gets done, a broken automation can go unnoticed for weeks while the time it saves shrinks.

Then there's the time an automation saves. A business case often counts on those hours going to something specific, like taking on more volume or moving someone to other work. If nobody decided what fills them, other work tends to expand into them, and the number the project was meant to move stays where it was.

Small automation projects tend to go wrong between departments, while large ones tend to go wrong inside them. A project as large as an ERP rollout usually gets mapped from the customer's order to the shipment before anything is built, because the system has to handle all of it. A list of small projects, each drawn around one team's step, rarely gets that map. That's what makes each one easy to approve and quick to finish, and it's also why the handoffs between teams end up on nobody's list.

Questions About Automation Projects Not Delivering Results

Why are automation projects not delivering results when every bot works?

Often because each automation was scoped to one department's step, while much of the delay sits between steps. Work an automation can't handle lands in a folder or queue nobody checks first thing, approvals that used to happen in person wait in an inbox, and cutoffs turn a few hours into a full day. Other common reasons are automations that break as their inputs change, and saved time that nobody decided how to use. Tracing where the time went on a handful of slow transactions usually shows which is at work.

Why is RPA not delivering ROI?

Sometimes the bots work and the time they save never reaches the number the business case counted on, because the delay was elsewhere or the freed hours weren't put to a specific use. Sometimes the bots stop working as well as they did, because a customer or supplier changed a format and the share of work going to exceptions crept up. Because the exceptions still get handled by hand, a bot that has stopped keeping up can go unnoticed for weeks.

How is a list of small automation projects different from one large one?

Small projects are easier to approve and finish because each one is drawn around a single department's work. A large project, like an ERP implementation, is usually mapped from start to finish first, so the handoffs between departments come up early. Neither guarantees results. They tend to go wrong in different places, small projects between departments and large ones inside them.

How many companies redesign their workflows when they automate?

Fewer than you might expect. In a McKinsey survey of 1,491 people published in March 2025, only 21 percent of those whose organizations use generative AI said they had fundamentally redesigned at least some workflows. Of 25 attributes McKinsey measured, workflow redesign had the biggest effect on whether generative AI improved earnings.

§ Next Step

Up to thirty minutes, confidential, and an honest read on why your automation projects aren't moving the number.

Get Your Free Pulse Check →
Automation and AI

An automation or AI tool can do exactly what it was built to do while the result it was bought for barely moves.

More in this cluster →
§ Begin

Start with your free Pulse Check.

Thirty minutes. Confidential. No card required.

Get Your Free Pulse Check →Up to thirty minutes · Confidential AI · No pressure, ever
Free
Up to 30-minute Pulse Check
READ DELIVERED SHORTLY AFTER
Comprehensive
Comprehensive Pulse Check
DETAILED ROADMAP · 24 HRS AFTER FINAL INTERVIEW