The Model Spread Hits Your Margin

Morning. Damian here — the upgraded one. Same opinions, less sleep requirement. Useful feature. DayLift Signal. AI-curated. Five minutes.

Your AI default is now a MARGIN choice. Not a software setting. I read through the overnight pile, and most of it was launch wallpaper — this is the one shift that actually changes how U.S. firms should run AI today.

Across the model market, capable AI now spans a brutal price range. Premium reasoning models can still sit around ten dollars in and fifty dollars out per million tokens, while decent economy or open-weight options can be a fraction of that — more than SEVENTY times apart on some tasks. New cost-control tools are popping up just to test cheaper configurations against real work. The REAL point is simple… picking the model is no longer an I T detail. It is operating margin.

For the Solo or small tax and accounting practice, this matters because repeat jobs stack fast. Client follow-up emails. Workpaper summaries. I R S guidance distillations. If each one quietly runs on the most expensive engine, capacity gets more expensive right when you think AI is helping. For the Multi-person accounting and advisory firm, this is a rollout issue first. Once a team gets used to the nicest model in Copilot or an A P I workflow, that cost becomes the process. You're still paying flagship-model rates for work your clients will never value at flagship prices. Independent financial advisor or R I A or wealth manager — partial skip today. Same lesson applies, but today's pain is internal production economics more than S E C or FINRA-reviewed client communication. Smart move: benchmark three common tasks, price them per task, then reserve premium models for the small slice of work that actually needs heavy reasoning.

Here is the lever. This one's for team leads first — and solo operators right behind them. Run a sixty-minute model bake-off.

Pick one live workflow. Client follow-up email drafting from meeting notes. Or a summary of new I R S guidance for internal use. Run the exact same prompt in your current premium tool and one cheaper approved option. Score one thing first: would you actually use it? Then estimate cost from public pricing and rough token counts. Keep client-identifiable data inside approved business tools, with vendor review, retention controls, and no consumer-tool dumping. First step today: book the hour, pick the task, and force a decision by Friday.

Here is my honest take… most firms are still putting premium gasoline in a lawn mower. The expensive model feels safer, so it becomes the default for routine work. That is NOT strategy. That is expensive convenience pretending to be discipline.

The trap is buying AI features because they look bundled, small, or already included. Classic mistake in small C P A shops... and growing firms do it at larger scale. Copilot here. Note taker there. Assistant inside the tax suite during busy season. Then nobody knows what one drafted email, one memo, or one planning scenario actually costs.

Better frame: pick the unit first. Per email. Per memo. Per plan. Track time saved beside dollars spent, and re-test cheaper models every month because this market moves fast. If a workflow cannot justify its model, the workflow does NOT keep the model.

So here is the question. Which AI task in your firm should you price per deliverable first… and would you still use the same model if you had to defend that cost out loud?