Almost every discovery call we run in 2026 includes some version of the same question: “where should we add AI?” It's the right question to ask — but usually at the wrong time. AI features amplify a product; they rarely fix one. Before we recommend any AI integration, we look for a handful of signals that tell us the foundation can support it.
Signal 1: You have a repetitive decision, not just repetitive work
Automation handles repetitive work. AI shines when there's a repetitive decision — classifying a support ticket, prioritizing a lead, extracting fields from a messy document. If a rules engine could do the job with ten if-statements, build the rules engine. It's cheaper, faster, and it never hallucinates.
Signal 2: You have the data — and it's yours
The best AI features are grounded in data your product already collects: your users' history, your catalog, your domain documents. If the feature only works with generic model knowledge, users can get the same answer from any chatbot — you're adding latency and cost without adding moat.
Signal 3: A wrong answer is survivable
Every model is wrong some percentage of the time. Great AI products are designed so that a wrong answer costs seconds, not trust: a draft the user edits, a suggestion they can dismiss, a summary with the source one click away. If a wrong answer is catastrophic — payments, compliance, medical — you need a human in the loop, and the feature should be designed around that from day one.
When we say no
We regularly talk clients out of AI features — when the core flow still has friction, when the data isn't there yet, or when a plain algorithm solves it. Shipping a mediocre AI feature is worse than shipping none: it trains users to distrust the product.
If those three signals are present, though, the upside is real. Our fastest-growing client products all pair a strong core workflow with one well-placed AI capability — not ten scattered ones.
Have a product in mind?
Tell us what you're building — we'll tell you how we'd ship it.