Executive AI OSNo. 07
The new back office: AI handles the tasks, the team decides
Where to start handing administrative work over to AI, who can lead the change and how to measure the business benefits
Invoices, purchase orders, supporting documents and supplier enquiries can keep an administration team busy all day, getting information to the right place. That leaves less time to work out which customers generate the best margins, anticipate cash requirements or produce more accurate forecasts.

The opportunity for AI is clear: it can take on some of the work of gathering, cross-checking and recording information, freeing up more of the team's time for analysis and decisions. For business leaders, that means two opportunities: supporting growth without a matching increase in manual work, and improving financial analysis at the business's current scale.
Over the past year, I have been helping SMEs train their teams. In that work, I have seen reductions of between 65% and 75% in the operational workload of a traditional administration department. The challenge for leadership is to turn that extra capacity into better results.
01
Where the opportunity lies
In my experience working with SMEs, the best starting point is where the team spends the most time gathering, checking and moving information. Look for three things when choosing your first process:
- Repetition: the same task comes up time and again.
- Speed matters: delays hold up other tasks or decisions.
- A familiar process: the team knows the steps and checks involved, even if they have not yet been written down.
Choose a process that meets those conditions, test it with the team and measure how much manual work it saves and which checks the team still needs to carry out.
AI is particularly useful for making sense of information in different formats, comparing it with other sources and drafting replies or creating records. To do this well, it needs business rules that explain how we work. The team decides where to check its work, handles exceptions and makes the decisions that require human judgement.
This applies across sectors; the value depends on the work involved. Look at the number of transactions, the documents each one requires and the time spent gathering them, cross-checking them and dealing with exceptions. Alongside invoices, teams may need to handle delivery notes, purchase orders, certificates and forms required by customers or under relevant regulations. Entering the same information in several documents, checking they are up to date and chasing missing paperwork all take time.
Logistics and distribution companies, manufacturers, construction firms and hotel chains often handle large volumes of transactions and paperwork. Even a small improvement per transaction can make a substantial difference. But volume alone is not the deciding factor: a business with fewer transactions may need extensive documentation for each one. Assess the whole administrative process, allowing for the time spent checking the work and the cost of keeping it running.
02
A practical example: invoice processing
If we connect our AI tool to the inbox used for supplier invoices, it can read incoming invoices and check them against a spreadsheet of supplier details and purchase orders. Invoices that pass the agreed checks can be logged in a connected spreadsheet, while potential duplicates and discrepancies are flagged for review. Our business rules tell it when we make payments, which terms apply to each supplier and how to draft an email acknowledging an invoice or querying an amount that does not match the purchase order. The team still checks the work at agreed stages and approves all payments.
Fig. 1From inbox to team review
- 01 Runs the skill
Every hour, the agent checks the supplier invoice inbox.
- 02 Checks invoices
Checks purchase orders, supplier details and business rules.
- 03 Posts to Slack
Logs results and flags discrepancies, duplicates or missing details.
- 04 The team reviews
Reviews each invoice’s status and resolves exceptions.
The workflow can extend to entering invoices in the ERP and arranging payment, with the necessary integrations and team approval.
The procedure can be saved as a skill: a reusable set of instructions and resources that tells AI what to do and what to check. With the required connections and permissions in place for email, the spreadsheet and Slack, a scheduled task can run the skill at regular intervals. For example: "Every hour, process new invoices in the accounts payable inbox and post an update in the #admin Slack channel showing which invoices have been logged and which issues the team needs to resolve." Share this example with your admin team to identify the business rules, information sources and review stages they would need to test it in their own work.
03
A new role alongside the team
In smaller companies, an AI Operations Manager can lead AI adoption across the business. In larger organisations, the AI Operations team may include an AI Back-office Specialist, focusing on administration, accounting and finance. They work with the people who run those functions to redesign processes, translate their expertise into instructions for AI and oversee the resulting workflows.
The role depends on close collaboration with the people doing the work. Together, they identify which steps still serve a purpose, where rules are missing and which exceptions need human judgement. Technical expertise turns those decisions into a process that works in practice and that the team can adopt.
Responsibility for accounting and financial decisions, including approvals, stays with the people in charge of those areas. Leadership sets the priorities and decides how to put the extra capacity to use.
04
Choose the metrics that matter to the business
Leadership's most important decision is how to use the time the team gets back. My test for these initiatives is straightforward: they must help us handle more transactions, understand our margins better or anticipate cash requirements. There are two ways to put that extra capacity to work:
Grow without a matching rise in admin work. Process more transactions with the same team, keeping paperwork from piling up and avoiding delays in invoicing and collecting payments.
Improve results without increasing sales. Spend more time analysing margins by customer or product, negotiating better supplier terms and anticipating cash needs. Factor in discounts, transport costs, seasonal changes and customer payment terms.
Before you start, decide what you want to improve and how you will measure it. If the aim is to handle more demand, track transactions per person alongside outstanding work and errors. To improve cash flow, track how quickly invoices are issued and which payments are overdue. To protect margins, measure the full cost of the process and the improvements secured in supplier terms.
Establish a baseline, name the person responsible and agree when to review the results. Include the time spent supervising and correcting the work, as well as the cost of the tools. This shows whether the improvement lasts and what it is worth to the business.
The question that matters is "How did we use that extra capacity, and what difference did it make to the business?" Freeing up time is only the first step. The real gain comes from turning it into better business results.
