Section 1
Where the margin actually moves
Look at your delivery cost line and split it into three buckets: the work only a senior person can do, the work anyone trained can do, and the work that is purely moving information between systems. The third bucket compresses hard, often to near zero at the margin. The second partly compresses, into review time rather than production time. The first does not move at all, and if you cut it you damage the product. The gain is real, but it is smaller than vendor arithmetic suggests, because review time is real time. A team that produces a draft in four minutes and spends twenty-five checking it has improved by less than the demo implied. The underlying definitions are in [What Is AI Automation? A Plain-English Guide for Founders](/blog/what-is-ai-automation-a-plain-english-guide-for-founders).
Section 2
Speed is the part customers notice
Margin is your problem. Speed is theirs. Response time, turnaround and time to a first answer are visible to buyers in a way that internal efficiency never is. This is where competitive pressure actually bites. If a competitor answers enquiries in ten minutes and quotes the same day, your two-day turnaround becomes a reason to lose, regardless of who does better work. Speed also compounds internally: faster cycles mean more iterations, more decisions per quarter, and a shorter path from customer signal to a change in what you sell. Choosing the systems that make that possible is covered in [Choosing the Right AI Automation Tools for Your Business](/blog/choosing-the-right-ai-automation-tools-for-your-business).
Section 3
The pricing problem you inherit
Here is the trap. If you bill by the hour and then halve the hours, you have cut your own revenue and called it an efficiency gain. Anyone whose costs fall has to decide where the saving goes: into price, into margin, or into more scope at the same price. All three are defensible, but the decision has to be deliberate and it usually forces a change in how you package the work. Fixed-price deliverables, retainers and outcome-based agreements survive this transition. Hourly billing does not, at least not without an uncomfortable conversation about why the same output now costs the client less.
Section 4
What breaks when the cost base changes
Cheaper output invites more of it. Teams that could produce ten proposals a month start producing sixty, and the bottleneck moves downstream to whoever has to review, approve or deliver on them. Watch for three failures. Volume outpacing review capacity, so quality control becomes rubber-stamping. Cost creep, where per-run pricing looks trivial per unit and arrives as a serious invoice at volume. And skill decay, where juniors never learn the work because the machine does the first draft and nobody explains what a good one looks like. Each is a management problem, not a technical one, and each shows up about two quarters after launch.
Section 5
The trust cost nobody prices
Trustworthiness, design, evaluation and use are the four headings NIST puts around AI risk, and trust is the line item that never appears in the return model. It is also the one that can wipe out a year of gains in a week. Set the boundaries in writing: what the system can see, what it can change, which judgments stay with a person, who answers for the outcome. Disclose machine involvement where a customer would care. Keep a human owner on anything touching money, employment, legal exposure or safety. On margin arithmetic, one avoided incident is usually worth more than the entire efficiency programme. Read [Storytelling in the Age of AI and Automation](/blog/storytelling-in-the-age-of-ai-and-automation) for how this shift lands with your team.
Section 6
The numbers that prove the model changed
Track four things and you will know whether the business model moved or just the demo did. Gross margin per delivered unit, before and after. Time from enquiry to first substantive response. Revenue per full-time employee. And the ratio of production hours to review hours, which is the honest measure of how much work really disappeared. Review them quarterly, not weekly. Business-model change shows up on a slower clock than tool adoption, and the early numbers usually flatter you because the hard cases have not arrived yet.