Writing / Practice
Is AI saving your practice time?
Only 38% of the 896 A&E firms in Deltek’s 47th Clarity study report a measurable positive business impact from AI, even though 70% say they have adopted it (Deltek, 12 May 2026). A further 39% say it is too early to tell or they are still piloting, which is the polite form of nobody measured.
18 October 2026 · 6 minute read
Why most firms cannot answer the question
Adoption has run ahead of measurement. In Deltek’s 2026 study, AI adoption rose to 70% of firms from 53% a year earlier and generative AI use to 78% from 64%, yet only 38% could point to a measurable positive business impact and 39% said it was too early or still a pilot (Deltek, 12 May 2026). The sample is mostly larger A&E firms. At the other end of the scale, the AIA’s own study found that 6% of the profession regularly used AI for their jobs, with chatbots, image generators and text tools the most common, and that 8% of firms had implemented AI with 20% more in the process (AIA, 11 March 2025).
The same AIA study explains why the saving is hard to count. 84% of respondents were optimistic about automating manual tasks, and 90% were concerned about inaccuracy, data misuse, security, authenticity and transparency (AIA, 11 March 2025). An output you have to check is not a saved hour until the checking is counted. No survey we found shows AI changing what architects charge, so the claim this guide helps you test is the modest one: did it save time, and where did the time go?
Set a baseline before you change anything
A saving is a difference, so you need the number before the tool. Four measures cover the business side of a small practice, and each takes a week or two of ordinary work to capture.
- Hours per task. Pick the four tasks you expect AI to touch: the first reply to a new enquiry, a fee proposal from brief to sent, meeting notes and actions, and the invoice chase. Log the time for each instance for two weeks, in quarter hours, with no AI.
- Proposal turnaround. Days from the client brief to the proposal in their inbox. Take it from your sent folder for the last ten proposals.
- Enquiry response time. Minutes or hours from the enquiry arriving to your first substantive reply. Take it from the timestamps on the last twenty enquiries.
- Conversion. Enquiries that became consultations, and proposals that became signed fees, over the last twelve months. This is the number that tells you whether faster was also better.
The baseline is the whole method. A firm that starts logging on the day it switches the tool on has nothing to compare against and will believe whatever it hoped.
The four-week log
Copy the table into a spreadsheet, one row per task per week. Week one is the baseline with no AI. Weeks two to four use the tool on the same tasks, and the column that decides the result is checking and rework, because an hour saved drafting and spent correcting is not saved.
| Week | Task | Hours without AI | Hours with AI | Checking and rework | Net hours saved | Quality note |
|---|---|---|---|---|---|---|
| 1, baseline | First reply to a new enquiry | Record each instance | Not used | Not used | 0 | Describe your usual standard |
| 1, baseline | Fee proposal, brief to sent | Record each instance | Not used | Not used | 0 | Describe your usual standard |
| 1, baseline | Meeting notes and actions | Record each instance | Not used | Not used | 0 | Describe your usual standard |
| 1, baseline | Invoice chase and reminders | Record each instance | Not used | Not used | 0 | Describe your usual standard |
| 2, 3 and 4 | First reply to a new enquiry | Week 1 figure | Record each instance | Record each instance | Week 1 minus (with AI plus checking) | Errors caught, tone, client reaction |
| 2, 3 and 4 | Fee proposal, brief to sent | Week 1 figure | Record each instance | Record each instance | Week 1 minus (with AI plus checking) | Scope errors, pricing errors, revisions asked for |
| 2, 3 and 4 | Meeting notes and actions | Week 1 figure | Record each instance | Record each instance | Week 1 minus (with AI plus checking) | Missed actions, wrong attributions |
| 2, 3 and 4 | Invoice chase and reminders | Week 1 figure | Record each instance | Record each instance | Week 1 minus (with AI plus checking) | Days to payment compared with week 1 |
Add the three pipeline measures as a second sheet: proposal turnaround, enquiry response time and conversion, recorded weekly against the baseline. Four weeks is short for conversion, so read that line over a quarter.
Rules that keep the log honest
- Count every minute of checking. The 90% of AIA respondents worried about inaccuracy were describing real work.
- Log the instances you abandoned. A draft you threw away and rewrote by hand counts as time with AI plus the full rewrite.
- Do not log hours that were never going to be billed anyway. Saving an hour of admin you used to do at nine in the evening is worth something, but it is not an hour of fee.
- Have one person own the sheet. Four people estimating from memory on Friday produces a story rather than a log.
Reading the result
After four weeks you will have one of three answers. If net hours saved is near zero or negative once checking is counted, stop using the tool on that task and say so; 39% of Deltek’s firms are still piloting, and a clean no is a result. If the saving is real but small, keep it and move on, because the hours matter only when they add up to a day a week. If the saving is large, the question changes from whether to where, and that is the question most firms never ask.
Look at the pipeline sheet before you decide. A faster first reply that did not change conversion saved you time. A faster first reply that did change conversion earned you a fee, and our article on why studios lose projects they have already won explains why the second case is common.
Turn saved hours into capacity, not discounts
A day a week saved is a fifth of a working week. There are two things to do with it. The first is to sell it back to the client as a lower fee because the proposal took an hour instead of a day. The second is to keep it and spend it on the work that earns the next fee: answering enquiries within the hour, qualifying harder, running a paid feasibility visit, going to site. The benchmarks give no support to the first. No survey shows AI changing what architects charge, and the fee is for the judgement, the liability and the result, none of which the tool carried.
Treat the saving as capacity and price it that way. The utilisation and capacity planner turns a weekly hour saving into annual fee capacity on your own rates, which is the figure to hold in mind when a client suggests the fee should fall because the work got faster.