A two-week AI audit that recommended against most of it
Mapped an immigration consultancy's whole intake-to-filing process, scored each step both on how much time it could save and how risky it would be to automate, and handed back a sequenced roadmap, including an explicit list of what not to touch.
The practice knew a large share of its work was repetitive and suspected AI could help, but had no way to tell which part. In immigration work, the payoff and the risk are spread very unevenly across what looks like one single process: some steps are pure clerical busywork, others are regulated advice that an unlicensed system must never touch.
In Canada, giving immigration advice for a fee is restricted to licensed representatives. Any automation touching enquiries has to be built so it's structurally incapable of giving advice, not just told not to, which rules out the obvious build and changes what the right one looks like.
Two weeks. We mapped the workflow as it actually runs, not as it's documented, then scored every step on two things at once: hours saved, and regulatory or reputational risk. Scoring on time saved alone would have ranked the highest-volume step first, which was exactly the step with the most legal risk.
A map of the workflow with time and volume figures, a two-part risk-and-value score for every step, a sequenced and priced roadmap, and an explicit list of what we recommended they not automate, with reasons.
Our first pass scored automation potential on a single number, which put first-contact eligibility enquiries at the top by volume alone, directly against the licensing boundary. We rebuilt the scoring to weigh risk as its own factor, which substantially reordered the roadmap and changed what we recommended building first.