08AI implementation
Back-office AI implementation: tickets, approvals, reporting, and admin
The back office is not the most interesting function in a company. It is usually the most expensive one nobody argues about. That is the whole case for this page, and it rests on a single structural fact: back-office work is cheap per unit and enormous in aggregate. Nobody defends a four-minute approval. Nobody schedules a meeting about a password reset. So nothing gets counted, and a cost nobody counts is a cost nobody cuts. We went looking for a published figure to put on that sentence and could not find one we would stand behind; every version we traced led back to a vendor selling the fix, not to anyone who had actually measured it. So the argument here runs on structure, not on a number.
01
The case for the back office
The back office is not the most interesting function in a company. It is usually the most expensive one nobody argues about.
That is the whole case for this page, and it rests on a single structural fact: back-office work is cheap per unit and enormous in aggregate. Nobody defends a four-minute approval. Nobody schedules a meeting about a password reset. So nothing gets counted, and a cost nobody counts is a cost nobody cuts. We went looking for a published figure to put on that sentence and could not find one we would stand behind; every version we traced led back to a vendor selling the fix, not to anyone who had actually measured it. So the argument here runs on structure, not on a number.
Sales work gets attention because it is measured against a target. Support work gets attention because customers complain. Back-office work gets attention only when it fails. The rest of the time it is spread so thinly across so many people that it is invisible on every report that matters.
Spread thin is exactly the shape automation is good at.
02
One week of back-office work, counted
Take a company of 100 people and count the small things rather than the big ones.
20 password and access requests. 60 approvals moving through email, each waiting on someone who is in a meeting. A weekly report someone rebuilds from 4 exports. 40 meetings whose notes exist only in one attendee's memory. A stock count reconciled by hand. 300 emails that needed a category and a destination rather than a reply. 12 documents retyped from a PDF into a form.
No line on that list is worth a project. The list is worth a headcount, which is precisely why it never gets addressed. There is no single owner and no single villain.
03
What gets automated, item by item
- 01
IT tickets and access requests. Password resets, software licences, permission requests, new starter and leaver access. Requests get read, checked against the person's role, provisioned where the rules allow, and escalated where they do not. Leaver access removal in particular gets safer, because the current version depends on somebody remembering.
- 02
Approval chains. Purchases, expenses, time off, contracts, discounts. The system routes by policy rather than by whoever answers first, chases the person actually holding it up, escalates when a threshold passes, and keeps the record. Most of the delay in an approval chain is not deliberation. It is a message sitting unread.
- 03
Inventory and scheduling. Reorder points recalculated against real consumption instead of a number set 2 years ago. Shift and resource schedules drafted around constraints, for a person to adjust rather than build from an empty grid.
- 04
Dashboards that write themselves. Not the chart, which you already have. The paragraph under it. Numbers pulled, checked against last period, movements identified, and the reasons drafted with the underlying rows attached, before anyone opens the file. The analyst edits the reading instead of assembling the data.
- 05
Meeting notes and email triage. Notes captured, decisions and owners extracted, actions pushed into whatever system tracks them. Incoming mail categorised, routed, and drafted where it is routine, with a person sending.
- 06
Document handling. Contracts, forms, statements, delivery notes and scanned paper read into structured data, filed, and checked against what they should contain. Retyping is the single most automatable task in most companies and the one people are least willing to admit to.
04
Why this function usually pays back first
Three reasons, and none of them are about the technology.
It is high volume and low variety, which is the shape automation handles best. It carries almost no external risk, because a customer never sees an internal approval chain, so a failure is embarrassing rather than damaging. And nobody is territorial about it. Try to change a sales process and you will meet a sales director with opinions. Offer to remove password resets from an IT team's week and nobody will defend it.
That makes the back office a good place to start even when it is not where the loudest problem is. It is where the first result can be shown quickly and where the argument for the next piece of work gets made.
05
Common failure points
Automating a process nobody has looked at in 5 years means paying to make a bad process faster. A form with 14 fields where 5 are needed does not want automation, it wants 9 fields deleted. That is a free change and it should happen first.
The second failure is the approval chain that exists to protect nobody. Plenty of chains have 4 steps because 4 people were once nervous. Automating that chain preserves it forever. We would rather ask which steps are genuinely controls and which are habit, and automate only the first kind.
06
What this typically runs on
Usually Jira Service Management, ServiceNow or Freshservice for tickets, Okta or the identity system already in place for access, Asana, Monday or Airtable for process and tracking, Power BI or Looker Studio for reporting, and Google Workspace or Microsoft 365 underneath all of it. Almost always tools you already pay for and use at a fraction of what they can do.
There is no charge for the build. What Praxis takes afterwards is a share of the gain that the work is measured to have produced, and the measurement is agreed before anything is switched on. How that is set up is a page of its own.
Questions
What should you know before the first call?
- What counts as back-office work?
- Back-office work is everything the customer never sees: IT tickets and access requests, approval chains, internal reporting, inventory and scheduling, meeting notes, email triage and document handling. It is defined by being internal and administrative rather than by which department does it, which is why it is often spread across every team and owned by none.
- Why is the back office where the biggest savings usually sit?
- Because the cost is spread rather than concentrated. Each task is small enough that nobody schedules a project around it, so nothing gets counted, and a cost nobody counts is a cost nobody cuts. Automation works well on exactly that shape: high volume, low variety, and no single owner defending it.
- Do you sell back-office automation software, or build it?
- We build it inside the systems you already run. Back-office automation tools are sold as products you then have to configure, staff and connect to everything else, and the connecting is the part that was hard in the first place. What we hand over is that connection, working on your ticket queue, your approval chain and your reporting, with no new product for anyone to log into. Where a tool you already pay for can do a piece of it, we use that one rather than adding another.
- Should a process be automated or fixed first?
- Fix it first, then automate what remains. Automating an unexamined process pays to make a bad process faster and freezes it in place. Deleting unnecessary form fields and approval steps that protect nobody costs nothing, and it usually reduces how much automation you need to buy.
- Will back-office automation put administrative staff out of work?
- It removes tasks rather than roles, and the tasks it removes are the ones nobody wanted: retyping, chasing, resetting passwords, rebuilding the same report. Where the work is genuinely reduced, that is a decision for the business to make openly, and we would rather say so plainly than pretend the question does not exist.
- How long does back-office automation take to build?
- It varies with how many systems have to be joined and how clean the underlying data is. Access requests and document handling are usually among the fastest to stand up because the rules are already written down somewhere. Approval chains take longer, because the first step is agreeing which steps in them are real controls.
Where this leads
This page is one part of the whole offer: automate your business with AI.
Start with a free consultation
Tell us what a normal week of tickets, approvals and reports looks like and who rebuilds what on a Monday. We will tell you what can be routed and drafted automatically, what should be deleted instead, and what we would build first. The call costs nothing and ends with a plain list either way.
No obligation · a scoping conversation first