Finding Where AI Fits in Your Business

Many discussions about AI begin with a tool, but a more productive approach is to start with the work itself. The sections that follow outline a method for identifying where AI can add value, discerning when it is not suitable, and selecting a first project worth testing.

First, examine where people connect systems. In most businesses, the most time-consuming and error-prone work occurs at the interfaces between systems. Information often arrives via email, PDFs, spreadsheets, and must be manually read, interpreted, and entered into another system. Signs of these slow and error-prone tasks include staff copying information from various sources into another system, individuals reading lengthy documents to extract relevant details, incoming requests being sorted and forwarded manually before action can be taken, and weekly reports compiled from multiple disparate sources.

Questions to consider include where work accumulates before someone can act upon it, what tasks would be halted or slowed if a critical person were absent for a week, where identical information is entered into multiple systems, which tasks scale directly with the volume of customers, orders, or requests, and where mistakes most frequently originate and how they are detected.

Engaging the individuals performing the work rather than solely the managers can provide valuable insights into workarounds and their actual time consumption.

Next, recognize tasks that are well-suited for AI. Good candidates typically include sorting emails, extracting details from documents, matching records across systems, and drafting documents based on existing material. These tasks share several characteristics: a person can verify the outcome, they involve reading or writing text rather than calculations, they occur frequently enough that even minor time savings accumulate, the input varies and requires interpretation, and someone must understand the content's meaning without applying deep expertise or making high-stakes judgments consistently.

For example, consider a property management company that receives maintenance requests from tenants via email and a web form. Tenants describe problems in their own words, such as "water under the sink" or "something is leaking in the cabinet." A coordinator reviews each request, determines the category (plumbing, electrical, or general repair), assesses urgency, and assigns the issue to the appropriate vendor.

The requests are frequent, the wording varies, and each request necessitates a small amount of interpretation. AI can assist by suggesting the category and urgency, with the coordinator merely confirming or adjusting the suggestion. This allows the coordinator to begin from a predefined framework rather than starting from scratch for every request.

Conversely, AI is likely not the best starting point when the problem is undefined, the task is infrequent, the required information is not available, the desired accuracy is exact every time, or the benefit is minimal. Adopting AI without a clear problem in mind often results in a feature that goes unused. Ruling out a task is valuable as it narrows down to the opportunities that are worth pursuing and often reveals simpler solutions that can be implemented immediately.

Before committing to AI, consider if a simpler alternative could solve the issue. Solutions such as fixed rules within the existing system to route requests based on customer-specific logic, direct integrations to move data between systems automatically, straightforward automations for repetitive tasks like sending reminders or generating standard reports, or even redesigning forms to collect information in a structured manner, can often be more affordable, predictable, and maintainable than AI solutions.