When owners think about AI, they usually think about getting more customers: better ads, faster follow-up, more content. That is where most of the money goes. The MIT NANDA report summarized by Fortune found that more than half of generative AI budgets go to sales and marketing tools, yet the biggest returns showed up in back-office automation, including cutting outsourced processing and external agency costs.[1]
That finding has caveats. It is a press summary of a report, and "return" there reflects the study's own definitions. But it fits what you see on the ground, and the reasons are practical.
Why back-office work is easier to get right
- It follows rules. An invoice either matches a purchase order or it does not. A timesheet either totals correctly or it does not.
- Volume is high and steady. The same task happens every day, so savings add up and you get data fast.
- Errors are contained. A mistake lands in a queue for review, not in front of a customer.
- Results are measurable. Minutes per item, days to close the books, invoices paid on time.
- Nobody has to like it. Staff do not feel their craft is being replaced when the tool does data entry.
Why sales and marketing is harder
Quality is subjective: is this email good? Brand risk is higher because mistakes are public. And attribution is murky. If leads go up the same month you changed three things, you cannot tell which change did it. None of that means sales and marketing AI is a bad idea. It means the return is harder to prove, so it makes a weaker first project.
Where owners say they use it
In Homebase's survey, the most common uses among small business adopters were data analysis and business insights (47%), payroll and timekeeping (45%), inventory management (44%) and administrative paperwork (43%).[2] In Bluevine's survey, 29% were automating operations and 23% used AI for financial management.[3] Different samples and question wording make the figures hard to compare directly, but both point to operations and admin work.
Eight back-office workflows worth examining
| Workflow | What the AI does | Human gate | Number to track |
|---|---|---|---|
| Accounts payable | Reads invoices, codes them, matches to orders | Approval before payment | Cost and days per invoice |
| Accounts receivable | Drafts reminders, flags late payers | Review of firm-tone messages | Days sales outstanding |
| Document intake | Sorts and extracts data from forms and attachments | Spot checks and exceptions | Minutes per document |
| Scheduling and dispatch | Proposes assignments from rules and availability | Dispatcher approves changes | Jobs per tech per day |
| Payroll prep | Flags missing punches and overtime anomalies | Manager approves before submit | Corrections per cycle |
| Reporting | Pulls data from systems into the weekly summary | Owner reviews numbers | Hours spent building reports |
| Customer follow-up admin | Logs calls and updates the CRM | Weekly hygiene check | Records updated on time |
| Contract and policy lookup | Answers "what does the agreement say" from your files | User checks the source | Time to find an answer |
A word about outsourcing and agencies
The MIT summary specifically mentions eliminating business process outsourcing and agency spend as a source of return.[1] If you pay a vendor per task, per hour or per page for routine work, you already have a price tag on the process. That makes the business case simpler: compare the tool and review cost to what you pay today. Just confirm you can bring the work in-house with enough oversight, and keep the outside vendor until the new process proves itself.
Start here
Pick the back-office process that takes the most staff hours and has the clearest rules. Time it for two weeks. Then decide whether to automate part of it or all of it. Our articles on invoice processing and on calculating return walk through how.
Sources
- Fortune, summary of MIT NANDA "The GenAI Divide: State of AI in Business 2025"
- Homebase, "2026 Main Street AI Gap Report"
- Bluevine, "2026 Small Business AI Trends Report"
Figures are quoted from the sources above and were reviewed in October 2026. Where a source is a secondary summary of other research, we say so in the text. Survey results reflect each survey's own sample and definitions. Examples labeled as illustrations use made-up numbers to show the method. Nothing here is a forecast or a guarantee of results.
Want this worked out for your business?
Tell us what is wasting your time.