Newsroom

Picture the automation lead at a mid-sized insurer at 7:10 on a Monday. Overnight, a broker portal moved its policy-number field two inches to the right. Forty-one claims sit in an exception queue. The bot that should have filed them did exactly what it was built to do: it looked for the field where it used to be, didn't find it, and stopped.

By 9:30 a developer has patched the selector. By noon the backlog is cleared by hand. Nobody records the morning as a cost. It shows up later, spread across a team that never seems to get to the new automations on its roadmap.

That morning is the line item most automation budgets leave out.

Licenses are the visible part

When a scripted-automation program is reviewed, the conversation usually starts with what is easy to count: licenses, infrastructure and the number of bots in production. Those are real costs, and they arrive on an invoice.

The costs that don't arrive on an invoice are harder to see:

• Break-fix time. Developer and analyst hours spent diagnosing and repairing bots after a screen, portal or document format changes.
• Re-runs and backlog. Work that piles up while a bot is down, then gets cleared by people, often at overtime rates.
• Exception handling. Items a bot hands back because they don't fit its rules: an unusual format, a missing field, a judgment call.
• Opportunity cost. The automations your team didn't build because it was busy keeping the existing ones running.

None of these are failures of the people running the program. They follow from how scripted automation works. A bot follows coordinates and selectors. When the screen changes, the instructions no longer match, and the bot stops.

Put a number on it

You don't need a consultant to estimate what your estate costs to keep alive. You need six numbers most teams already have:

1. How many bots run in production
2. How many of them broke last month
3. Average hours to diagnose and fix a break
4. Loaded hourly cost of the people who fix them
5. Items a month that bots route to people as exceptions
6. Average minutes a person spends on each exception

Multiply breaks by hours by cost and you have monthly maintenance. Multiply exceptions by minutes by the loaded cost of the operations team and you have monthly exception handling. Add both, annualize, and set the result next to your license and infrastructure spend.

A worked example (illustrative numbers, not a benchmark): 120 production bots; 18 breaks last month at 6 hours each and $85 an hour is about $9,200 a month in maintenance. 4,000 exceptions at 7 minutes each and $45 an hour is about $21,000 a month in exception handling. That is roughly $360,000 a year that never appears on a license invoice.

Your numbers will differ. The point is to have them, because they change the question from "how much do our licenses cost?" to "how much does each bot cost to keep alive, and is it worth it?"

Which bots are worth keeping

Once you have a cost per bot, the estate sorts itself into three groups:

• Stable and valuable. Bots on clean, structured, rarely changing screens. Keep them running; they are doing their job.
• Fragile and valuable. Important processes on screens that change often: portals, Citrix sessions, legacy desktop apps. These are the ones eating your team's week.
• Fragile and marginal. Bots that cost more to maintain than the work they save. Retire them.

The second group is where a different approach pays off. Computer-use agents don't follow coordinates. They look at the screen the way a person does, find the field they need, and carry on when the layout changes. They can also read the document that didn't fit the rules and route only the real judgment call to a person.

That doesn't mean replacing everything. Bots that run reliably should keep running. It means being deliberate about where your team's hours go.

Start with your own numbers

We built a run-cost calculator that runs the math above with your inputs, shows every formula, and gives you a cost per bot. It takes about two minutes and doesn't require an email address to see the result.

If you'd rather have the full picture, our free Bot Estate Audit does it bot by bot: a fragility score for each one, a run-cost model, and the ten bots worth retiring first, delivered in two weeks from an inventory export and two short workshops.