Deloitte Says 84% See Payback. In a Rental Yard, Start With the Credit Memos You Wrote Off Last Quarter

Eighty-four percent. That is the share of organisations investing in AI that reported positive return in Deloitte's State of AI in the Enterprise 2026. It is a real number and it is useless to you, because it is not your number. Your number is sitting in the credit memo column of last quarter's rental invoicing — the damage, fuel and cleaning charges you billed and then gave back because a customer argued and nobody could produce the photo.

Pull that column before you buy anything. On a 400-unit mixed fleet — compaction, mini excavators, scissor lifts, telehandlers, light towers — that column is usually somewhere between $1,500 and $4,000 a month, and it is the cleanest place in this business to prove whether AI paid for itself, because it has a before, an after, and a dollar sign.

The one process worth mapping first: the machine coming back at 4:40pm

Not the phone. Not marketing. The return gate. A skid steer comes off a roll-back at 4:40 on a Friday, the yard coordinator is already staging Monday's deliveries, the driver wants to clock out, and the check-in inspection gets a glance instead of a walk-around. The bucket edge has a fresh gouge. The fuel is at three-eighths. The hour meter reads 46 hours against a four-week contract written with a 160-hour allowance, so nobody thinks twice about it.

Three weeks later the customer disputes a $340 damage line. Your service tech knows the gouge was not there when it left, but the check-out photos are in Record360 and the check-in photos never got taken, and the branch manager writes a credit memo to keep a customer who rents $6,000 a month. Nobody logs that as a loss. It shows up as a credit.

The way to prove AI paid for itself is to pick one process with a countable error, measure it for 90 days before you switch anything on, add human review to the new version, and then compare the same count. Not a vibe. A count.

The baseline to capture before you turn anything on

Ninety days, five numbers, pulled from Point of Rental, Wynne RentalMan, Texada or whatever runs your counter, plus your damage documentation tool. Do not skip this. Once the new process is running you cannot go back and reconstruct it.

• Total returns in the period, by branch.
• Percentage of returns with a complete check-in photo set attached to the contract.
• Dollars of damage, fuel and cleaning billed per 100 returns.
• Credit memos issued against those charges, in dollars and as a percentage of what you billed.
• Median days from return to the damage line hitting the invoice.

That fifth one matters more than owners expect. A damage charge that appears on an invoice 19 days after the machine came back gets argued. The same charge sent the same evening, with two photos and an hour-meter reading, mostly gets paid.

flowchart TD
  A["Baseline: 90 days of returns and credit memos"] --> B["Switch on photo check-in at one branch"]
  B --> C["Agent drafts damage, fuel and hour-meter lines from photos"]
  C --> D{"Yard coordinator agrees with the draft?"}
  D -->|Edit| C
  D -->|Approve| E["Charge posts to the contract the same evening"]
  E --> F["Monthly: billed damage vs credit memos issued"]
  F --> G{"Recapture up and disputes flat?"}
  G -->|No| B
  G -->|Yes| H["Roll to branches two and three"]

What the reviewed version looks like at the return gate

The driver or yard coordinator walks the machine with a phone, the same eight or nine shots your check-out standard already calls for: four corners, tires or tracks, attachment, hour meter, fuel gauge, cab interior. That part is not new — Record360 and the check-in screens in most rental systems have done it for years, when people actually do it.

What is new is that the shots are compared against the check-out set on that same contract, and a draft billing line comes back before the driver has parked the truck: bucket cutting edge — new damage, compare frame 3 out and frame 3 in; fuel three-eighths against full out, 11 gallons at the contracted refuel rate; hour meter 46 against 12 at delivery, inside allowance, no overage. It reads like something your best coordinator would write on a good day when he was not in a hurry.

Then a human looks at it. The coordinator approves the fuel line, kills the cleaning line because the machine went to a dirt job and that is expected, and edits the damage line down to the cutting edge only. It posts that evening with photos attached. The customer's project manager sees it while the job is still fresh in his memory, which is the entire trick.

The arithmetic that settles the argument

One branch, illustrative figures — replace them with your baseline.

A difference of roughly $4,870 a month at one branch, against a running cost that is a small fraction of that. Two things to be careful about. First, some of the "before" gap is not AI at all — it is simply that people take the photos when the process asks for them, so run one month of photo discipline without any drafting to separate the two effects. Second, if your dispute rate goes up rather than down, stop. You are billing things you should not be billing, and the credit memo column will tell you within six weeks.

The self-storage version: the move-out nobody dated

Same logic, different form. On the storage side the countable error is the move-out that was never processed: tenant clears the unit on the 28th, does not tell anyone, the lock check does not catch it for nine days, billing continues, the tenant calls angry on the 12th and you refund a month plus the tenant protection plan fee.

Baseline the same way: number of refunds and rate adjustments issued per month, the median gap between the last gate entry and the recorded move-out date, and the number of units showing vacant in storEDGE that still had a lock on the walk. The nightly job is a comparison of gate activity, autopay status and the lock-check sheet, producing a short list of probable move-outs for the manager to verify on the morning walk. The measure of success is refund dollars and days-to-vacant, not a feeling that things are tidier.

Where the human still signs

Damage disputes with a customer who rents six figures a year are a relationship decision, not a photo decision. Keep the branch manager's authority to waive intact and do not measure him on recapture alone, or he will bill a $180 tire on a $70,000 account and you will lose the account.

Anything involving insurance, subrogation or an incident — a tip-over on a scissor lift, an injury, a machine that came back with a damaged fall-arrest anchor — goes straight to a person, and the machine goes on hold until a qualified tech signs it off. No draft billing line should ever be allowed to close out a safety item.

Nor should you let the drafting set prices. The rate for a cutting edge, a refuel or a cleaning comes from your rate table, reviewed by your service manager. And for storage, refunds and prorations touching a lien or auction account never get automated — that is state statute territory.

Frequently asked questions

How long before I know whether it worked?

One quarter, if you baselined properly. You need enough returns for the credit memo percentage to mean something — roughly 300 returns or more. Below that, run two quarters and expect noise. Do not let anyone declare victory at week three off a good month.

Our coordinators already hate the check-in tablet. Why would this be different?

Because it removes typing rather than adding it. The old ask was: take nine photos and then write up the damage. The new ask is: take nine photos, and approve or edit a draft. If your team still resists, that is your answer about the current process — the photos were not getting taken before either, which is exactly why the credit memos exist.

What does it actually cost to run?

Less than most owners expect. Frontier AI is down roughly tenfold from 2025 prices, and this is a small job on a few photos per return. The real cost is the ten to fifteen hours of setup — connecting the photo tool to your rental system, loading your rate table, and writing down what your branch counts as billable versus fair wear.

Can I use the same approach on something other than damage billing?

Yes, once. Pick the next process with a countable error — unsigned delivery tickets, missing purchase order numbers, expired certificates of insurance. The discipline is the same: baseline, one branch, human review, count again. Widening to four processes at once is how these projects end up unmeasurable.

A note from CallSphere

Recapture generates phone calls — customers ringing the counter to ask about a damage line, or to call a machine off rent before it accrues another week. CallSphere builds AI voice and chat agents that answer the counter line and the office line, including evenings and weekends, take call-offs and callback requests, answer routine rate and availability questions, and log each one with the contract details attached so the follow-up lands with the right person the next morning.