Cheap AI Just Changed Your Margins

Morning. Damian here — sort of. He upgraded the morning shift to the version that does not get tired, distracted, or weirdly protective of its calendar. DayLift Signal. AI-curated. Five minutes.

Cheap frontier AI is now a UTILITY. That is the story. I went through the latest pile this morning… most of it was launch wallpaper. This is the one change that actually hits your margins.

OpenAI cut GPT five point six Luna to about twenty cents per million input tokens, and Google is pushing Gemini three point six Flash as the lower-cost lane for long-running agent work. That sounds technical. It is not. It means bulk drafting, document summaries, meeting notes, and internal research bots can now run for single-digit dollars a month if you design the workflow right. The big shift is NOT that AI got smarter this week… it is that routine AI got cheap enough to stop feeling optional.

For the Solo or small tax and accounting practice, this is a busy-season capacity story. Notice replies, missing-document follow-ups, meeting recap emails, QuickBooks cleanup instructions — a lot of that first-pass work just moved into the affordable bucket. For the Independent financial advisor or R I A or wealth manager, the opening is narrower but still real. Internal prep, transcript summaries, and planning-meeting notes are fair game in controlled tools. Client-facing commentary still runs into S E C, FINRA, retention, and marketing-rule risk, so do not hear this as a green light for public-facing automation. Multi-person accounting and advisory firm — not the main lens today, because this matters to you too, but the sharper decision today is cost discipline for smaller shops and compliance boundaries for advisors. You're paying premium-model prices for work that belongs in the cheap lane. Smart move: re-benchmark your current tasks against Luna and Gemini Flash now, then cap premium-model use before March and April make the decision for you.

Here is the lever. This one's for solo operators first, and firm owners second. Build a simple AI router.

Put routine jobs — email drafts, document summaries, internal S O P writing, basic spreadsheet or QuickBooks scripts — through Luna or Gemini Flash by default. Route only the messy work, like multi-entity planning memos or nuanced regulatory analysis, to the expensive model. Track one thing: cost per task. Keep confidential client data out of consumer AI tools unless you have approved controls. First step today: pull one week of AI usage, re-run five routine tasks on a cheaper model, and compare quality against cost.

Here is my honest take… most firms using premium AI for ordinary work are not being sophisticated. They are buying premium gasoline for a lawn mower. It sounds serious. It burns money. If the task is repetitive and low-risk, the premium model is usually buying comfort — not better operations.

The trap is not overspending once. It is having no unit economics at all. I see this with solo C P As and with R I As. Staff hit whatever model button is closest, nobody tags the task, and no one can answer what one client-ready memo or one recap email actually costs.

Of course that feels modern… until the bill shows up and margins get fuzzy.

Better frame: treat AI like cloud compute. Define three to five task types, assign the default model for each, and track spend per deliverable. If you cannot price one unit of AI work, you do NOT control it.

So here is the question. What task in your firm should move to a lower-cost AI model this month — and what quality check would you keep in place before it reaches a client or file?