From experimenting to operating
AI is becoming familiar, but using a tool and embedding it into a business process are different steps. The Federal Reserve’s 2026 small-employer report found that 46% of surveyed firms used AI. Among users, just 7% reported full integration. Accuracy and adapting tools to business needs were leading challenges. The research covered U.S. firms with 1–499 employees; its optional AI questions came from a 2025 survey. [1]
Our interpretation: a useful automation service must understand the process, connect its systems, and establish how the result will be checked. Generating an answer is one step. Updating the right record and handling an exception may be the work that matters.
The work between applications adds up
In Intuit’s 2024 survey of 630 U.S. owners and executives at businesses with 10–99 employees, respondents reported 25 hours weekly on manual data entry or reconciliation across applications. Seventy-two percent wanted more automation; 64% wanted better integration. This sponsor-funded online survey describes its respondents, not a guaranteed saving for every business. [2]
That points to concrete starting places: emailed orders copied into an operating system, documents chased before billing, customer requests passed between teams, and reports assembled from several exports. A workflow audit should establish which of these actually costs your business time.
Four practical places to look
These are our priorities for exploration, informed by the research—not a statistical ranking of buying intent. Each example needs validation against the business’s volume, systems, and approval requirements.
| Business context | Example starting workflow | Measure |
|---|---|---|
| Accounting & finance | Document intake to reviewed accounting entry | Handling time and corrections |
| Wholesale distribution | Emailed order to verified system record | Order turnaround and rework |
| Field services | Incoming enquiry to scheduled next step | Response time and follow-up coverage |
| Manufacturing administration | Supplier confirmation to purchasing update | Processing time and unresolved exceptions |
Accounting offers especially direct evidence: Intuit’s 2025 survey of 700 U.S. accounting professionals found that integration and time-consuming data entry were each a challenge for 41%. The sample included both small and larger firms. [3]
NAW’s distribution research identifies customer service, inventory, and sales among AI priorities. Its public summary is an industry signal rather than a representative SMB estimate. [4] In Jobber’s survey of 1,050 U.S. home-service owners, customer communication, quoting, and paperwork were prominent time drains. [5] NIST MEP’s manufacturing-client research highlights continuous improvement and cost reduction; its clients are not a sample of every manufacturer. [6]
Legacy systems need a practical assessment
An incomplete or unreliable API can leave a manual step in an otherwise connected process. Browser interaction can provide another route where the application and access permissions allow it. That capability already exists in established automation products, including UiPath and Microsoft Power Automate. [7]
The useful question is whether the complete workflow can run reliably in your environment. We assess available integrations, authentication, interface changes, and the human decisions that should remain. Research supports integration pain; it does not establish how many SMBs specifically have unreliable APIs, or prove universal compatibility.
Our approach: inspect, explain, improve. Acquint’s observability suite gives you visibility into agent actions. Describe a correction in plain-language chat, and that feedback can be automatically incorporated into the workflow to support self-healing behavior. Inspect the next run to verify the change, with review and escalation boundaries agreed for your process. This describes our approach, rather than an independently measured recovery rate or a promise that every failure repairs itself.
Measure capacity and cash savings separately. Time recovered can help a team handle more work. It becomes a cash saving only when a real expense falls. The Federal Reserve found productivity improvements were more common than changes to labor costs among AI users. [1]
Start with a business case you can test
Count the eligible transactions, record current handling time, and identify errors and delays. Then estimate the remaining review effort alongside implementation, software, runtime, and support costs. Agree what successful completion looks like before launching a pilot.
This is the purpose of our free initial audit: identify the friction, prioritize feasible opportunities, and propose a measurable next step. Implementation is scoped separately. Request your free AI workflow audit.
Sources & reading notes
Reviewed September 21, 2026. Surveys differ in timing, sample, and definitions; their adoption rates should not be combined. Findings inform our approach and are not Acquint customer results.
- Federal Reserve: 2026 Report on Employer Firms2025 fieldwork; weighted convenience sample; AI module optional.
- Intuit: 2024 Business Solutions SurveyPaid online panel; U.S. businesses with 10–99 employees.
- Intuit: 2025 Accountant Technology SurveyResearch commissioned by a software provider.
- NAW: AI in Distribution, 2026Public analysis of distribution-leader research.
- Jobber: 2026 Home Service Trends ReportDecember 2025 owner survey; includes businesses of different sizes.
- NIST MEP: Manufacturer ChallengesFindings from FY2024 MEP client surveys.
- UiPath RPA and Microsoft Power Platform guidanceVendor descriptions establish advertised capabilities, not comparative performance.