Sample
What a finished audit looks like.
Redacted structure from a real delivery. Client identity, systems, and figures are replaced with placeholders — we do not publish a client's stack or their costs.
Direct answers to the goals you named
Each workflow you asked about gets its own recommendation. Example shape:
| Your stated goal | Recommendation | Why |
|---|---|---|
| Handle more inbound calls | Do it differently, not with an AI receptionist | Your booking system already collects requests and requires human confirmation. Adding a second booking path creates conflicts. Automate the follow-up message instead. |
| Reduce time on patient replies | Do it — narrow and fenced | Template drafting for non-sensitive messages is worth it. Anything identifying a customer stays out of a general AI workspace. |
| Get better at scheduling | Do not do it yet | Scheduling is owned by an existing system with a live calendar. Until we know which one and who confirms, a new tool adds a second source of truth. |
Note the shape: two of three are "yes, like this" rather than refusals. That was the main failure of an earlier version of this audit.
Keep / cut / defer, with your numbers
| System | Monthly | Owner | Renewal | Decision |
|---|---|---|---|---|
| General assistant (seat 1) | $20 | Owner | Monthly | KEEP — narrow non-sensitive drafting lane |
| General assistant (seat 2, unowned) | $20 | unknown | unknown | CUT — no owner, no defined weekly job |
| Meeting recorder | $18 | unknown | unknown | CUT — not used on sensitive calls, unused for the stated workflow |
| Specialist documentation tool | unknown | — | — | DEFER — only if the ranked bottleneck is documentation |
Where a field was left blank it stays "unknown". We do not fill in a plausible figure to make the table look complete.
The ranked bottleneck, worked in depth
Options compared on fit, cost, effort, and risk, ending in a single recommendation and the tradeoff you are accepting. Not a list of possibilities.
Monday actions and a 30-day sequence
Every finding carries a specific next action with a named owner. The sequence covers weeks 1-4 and ends with a cancel rule so the stack does not drift back.
Ratings are editorial product assessments, not lab benchmarks. How we score.