Can revenue leakage be fully automated?

No, not fully, and you would not want it to be. Finding, checking and pricing discrepancies between CRM, contracts, billing and payments can largely be automated. Resolving them can be partly automated: preparing a missing invoice or creating a task can happen automatically. But deciding whether to invoice a customer retrospectively, reverse a discount for a strategic account or reopen a contract are commercial choices that a person should make.

The chain from leak to resolution

To see what you can automate, it helps to break the work down. Tackling revenue leakage consists of six steps.

1. Detect. There is a discrepancy: a deal without an invoice, a price that does not match the contract, usage above what is being invoiced.
2. Validate. Is it really an error, or is there an explanation? Perhaps the discount was agreed, but it is only recorded in an email.
3. Price. What is the discrepancy worth, per year and in total?
4. Decide. What do we do about it? Correct it going forward, recover it, leave it, discuss it with the customer first?
5. Execute. Raise the invoice, adjust the price, correct the subscription, inform the customer.
6. Verify. Has the correction been applied properly, and is the money coming in?

Each step can be automated to a different degree.

What you can automate well

Detecting

This is the step where automation delivers most. A system that continuously puts CRM, billing, contracts and usage data side by side finds discrepancies that a person would only see after months, or never. It does so every day, for every customer, without getting tired or skipping a month.

Detection works with two kinds of logic. Rules: a won deal must have an invoice within thirty days, a contract with an indexation clause must get a new price on the indexation date. And patterns: this customer structurally pays less than comparable customers, this account shows a decline similar to earlier customers who cancelled. The first is classic automation, the second is where AI plays a role. How that works is covered in how to detect revenue leakage automatically and whether AI can detect revenue leakage.

Pricing

Once the discrepancy is known, the amount is usually a calculation: the difference between the contract price and the invoiced price, multiplied by the period. That can be fully automatic, provided the contract data is available.

Verifying

After a correction, a system can check whether the new invoice was sent, whether the amount is right and whether it has been paid. That too is easy to automate.

What you can partly automate

Validating

Part of the validation can be automated. A system can check, for example, whether there is a credit note that explains the difference, whether the deal was later cancelled, or whether there is a second invoice under a different customer name. That filters out a large share of the false alerts.

But part of the validation requires knowledge that is not in any system. The salesperson gave an extra discount verbally. The customer agreed a different amount with the managing director. The contract was amended at renewal but the PDF is not yet in the system. Only a person knows that.

Executing

Part of the execution can be automated, especially for unambiguous corrections:

• Preparing a draft invoice for a won deal without an invoice.
• Creating a task for the account manager with the evidence and the amount.
• Sending a reminder for an overdue invoice.
• Preparing a proposed new price after an indexation date.

What you should not do automatically: send an invoice directly, change a price directly in billing or email a customer directly about an error. Not because it is technically impossible, but because an error there lands straight with the customer.

What a person must keep doing

Deciding

This is the core. Suppose the system finds that a customer has had a 15 percent discount for two years that should have ended after one. What do you do?

• You can stop the discount immediately and inform the customer.
• You can recover the difference for the past year.
• You can let the discount run until the next renewal and discuss it then.
• You can leave the discount in place because the customer is strategically important and is negotiating a renewal.

All four are defensible. Which is right depends on the relationship, the size of the account, the negotiating position and the strategy. That is not a calculation. The difference between a system that recommends and a system that decides is set out in AI recommendations vs AI decisions.

The overview

This model is also known as human-in-the-loop: the system does the work that is scalable and repeatable, and a person approves at the moments where an error is expensive or visible. More on this in human-in-the-loop AI.

Why full automation is a risk

Suppose you automate it fully anyway. What can go wrong?

False alerts become actions. A system sees a discrepancy that is not there, because an agreement was made outside the system. Without human validation, the customer receives a corrective invoice for something that was legitimate. That costs more trust than it brings in.

Errors scale. A wrong rule in an automated process does not make one mistake, but a hundred. If the rule "index all contracts by 3 percent" accidentally also hits contracts without an indexation clause, a hundred unjustified price increases go out.

Commercial context is missing. A system does not know that the customer is about to sign a large contract next month, or that the managing director made a personal agreement. An action that is arithmetically right can be commercially damaging.

Nobody understands what is happening any more. If everything is automatic, after a year nobody knows why a particular correction was made. If something then goes wrong, it is hard to put right.

Worked example: what one wrong automated action costs

Suppose a system automatically sends corrective invoices for missed indexations. Because of an error in the link with contract management, 40 customers without an indexation clause are hit, at an average of EUR 1,800 per invoice. This is an example.

• EUR 72,000 in unjustified invoices.
• 40 customers who call, complain or quietly start having doubts.
• Credit notes, apologies and explanations: about 2 hours per customer for account management and finance, 80 hours in total.
• If 2 of those customers leave at their next renewal, each with a contract of EUR 30,000 a year, that costs EUR 60,000 a year.

One approval step would have prevented this. The time it takes is a fraction of the damage.

How to draw the line sensibly

1. Automate detection fully. This is where the biggest gain and the smallest risk lie, because nothing goes out of the door.
2. Automate pricing and verification. Also low risk, with a large gain in time.
3. Automate validation where you can. Filter out the false alerts that can be explained by data. Put the rest in front of a person.
4. Have execution prepared, not sent. Drafts, tasks, proposals. A person presses the button.
5. Leave decisions with people, backed by good information. The system provides the evidence, the amount and a proposal. The owner decides.
6. Widen automation step by step. Once a particular type of correction has been approved a hundred times without error, you can consider letting that type run automatically. Not before.

That last step matters. Automation does not have to be static. Start cautiously and extend it where the evidence is there. What that delivers more broadly is covered in what Revenue Intelligence delivers.

Frequently asked questions

Can AI resolve revenue leakage on its own?

AI can find discrepancies, price them and propose a resolution. It can also prepare a task or a draft. Sending an invoice or changing a price on its own is technically possible, but for most companies it is not wise without human approval.

How much of the work can realistically be automated?

Most of the searching and most of the calculating. That is also the part that currently takes the most time, or does not happen at all. The decision and the communication with the customer remain human work, but with far better information than without automation.

Will it become fully automatic in future?

For unambiguous, frequent corrections with little risk, probably more and more. For decisions that affect the customer relationship, human judgement will remain sensible. The question is not whether it is technically possible, but whether you want a system negotiating with your customers on your behalf.

What do you need to automate detection?

Access to the systems your revenue flows through, a way to link customers across those systems, and rules or models that define what counts as a discrepancy. Plus someone who picks up the findings, otherwise automatic detection is just an automatic list that nobody reads.