Technology and AI
What to Automate First at an Engineering Firm
The best first automation at an engineering firm is the task that repeats every week, follows the same steps and costs billable people’s time. There are usually a few of those in the office, well before any in the design.
On this page
Terms Used
- Automation: software doing a task that follows fixed steps, without a person doing each step.
- Integration: a link that moves data between two systems, such as timesheets into billing, without retyping.
- Trigger: what starts an automation: a day of the month, a timesheet submitted, a file arriving.
- Exception: a case that doesn’t fit the steps and goes to a person.
What Makes a Good First Automation
Three tests pick out the best candidates:
- It repeats. Every week or every month.
- It follows the same steps. The same inputs, the same output, little judgment in between.
- It takes billable people’s time. An hour of a project engineer’s time on a task is worth more to automate than an hour of the software’s time saved elsewhere.
The tasks that pass all three are usually in the office, around the engineering: moving data, assembling documents, producing reports.
The Usual Candidates
| Task | What automation looks like | Who still decides |
|---|---|---|
| Hours keyed into two systems | Timesheets flow into billing and payroll without retyping | Whoever approves timesheets |
| Monthly invoices | Draft invoices built from the month’s timesheets and percent complete | The project manager approves each one |
| Weekly project reports | Hours against fee and utilization, produced every Monday from the systems | The partners read and act |
| Proposal assembly | Project sheets, résumés and standard text pulled into a first draft | The proposal lead |
| Report assembly | Standard sections, tables and data placed into the firm’s template | The engineer completes and checks it |
| Document filing | Incoming drawings and correspondence named and filed to the right project | Exceptions go to administration |
| Reminders | Timesheets due, invoices overdue, insurance and licence renewals | The person reminded |
How to Choose from Your Own Records
The measure is the hours each task takes in a month: hours a month = times a month × hours each time. A weekly project report that takes 5 hours comes to 20 hours a month (4 × 5).
- List the repeated tasks. Ask each person what they do every week that follows the same steps.
- Estimate each task’s times a month, its hours each time, and who does it.
- Value the hours at what they could earn on projects, for the people who bill.
- Start with the task at the top, unless it is very hard to automate; then take the next.
| Example, round numbers | Times a month | Hours each | Who does it | Hours a month | Billable people’s hours |
|---|---|---|---|---|---|
| Proposal assembly | 4 | 8 | Engineers | 32 | 32 |
| Monthly invoices | 1 | 30 | Project managers | 30 | 30 |
| Hours keyed into billing | 4 | 6 | Office manager | 24 | 0 |
| Weekly project reports | 4 | 5 | Project managers | 20 | 20 |
In this example, proposal assembly goes first: 32 hours of engineers’ time a month, which at $150 an hour could earn $4,800 on projects (32 × 150). Keying hours into billing takes almost as long but no billable time, so it ranks lower, unless the office manager’s hours are what the firm is short of. Timesheets help here: if non-billable time is coded into a few categories, the largest ones are visible in a month. See how to improve utilization.
Automation, AI, or Both
Automation and AI are often used as one word, and they are different tools. Automation follows fixed steps: move these hours, build this invoice, file this document by its number. It is predictable, and many of the most useful automations at an engineering firm use no AI at all. AI is for the steps where the input varies: reading an RFP to list its requirements, drafting a project description, sorting correspondence that arrives without a project number. The two combine well: an automation gathers the data and places it, an AI step drafts the words, and a person approves. See AI at an engineering firm for where AI pays back.
Often No New Software
Many automations link systems the firm already pays for: the timesheet tool, the accounting package, the document store, email. Many of those systems offer a built-in link, a connector or an export, and a connection between two of them can remove a task completely. New software is worth buying when no existing system can hold the data. Before buying, check what the current systems can already do; before building, check what they can already send and receive. Our article on choosing project and accounting software covers the larger decision.
Design the Exceptions
Every automation meets cases its steps don’t cover: a timesheet coded to a closed project, an invoice for a client with special terms, a drawing with no project number. Deciding in advance what happens to them is what keeps an automation trusted.
- A named owner. One person who knows what the automation does and is told when it stops.
- An exceptions list. Cases that don’t fit go to a person, with the reason, instead of being forced through.
- A log. What ran, when, and what it changed, so a wrong invoice can be traced.
- A first month of checking. Every output compared with what a person would have done, before anyone stops checking.
- A plan for changes. When a connected system updates, the owner checks the automation still works.
What to Leave Alone
- Engineering judgment. Design decisions, checking and anything under a seal stay with the engineer. Automation can prepare the inputs and the paperwork around them.
- Rare tasks. A task done twice a year is cheaper to do by hand.
- Broken processes. Automating a process that does not work makes it fail faster. Fix the steps first.
- Tasks still changing. Wait until the steps have settled; automating a moving target means building it twice.
Measure the Hours Back
Record the task’s hours for a month before, and a month after, whether the automation uses AI or not. That record is the only way to know what an automation gave back, and it settles whether to keep it, change it or switch it off. Then check where the freed hours went: hours only earn once they land on billable work, which the resource plan shows.
For the steps that do use AI, the industry’s own results are a reason to measure. Deltek’s 2026 study found only 38% of architecture and engineering firms reporting a measurable positive business impact from AI (Deltek, 12 May 2026). The study measured AI, not automation in general.
A Checklist for the First Automation
- List the tasks that repeat weekly or monthly and follow the same steps.
- Estimate times a month, hours each and who does each one.
- Rank them by billable people’s hours.
- Check the top task’s steps work by hand before automating them.
- Name an owner and decide where exceptions go.
- Record the hours for a month, build it, and record them again.
- Check where the freed hours went.
How We Approach This
We build the drafting, the links between your systems and the reports, in-house, as part of our software and automation work, starting from a map of where the unbilled hours go in your own records.
Questions
What Can an Engineering Firm Automate?
Business tasks such as moving hours between time and accounting systems, preparing draft invoices, weekly project and utilization reports, proposal and résumé assembly, filing project documents and reminders.
How Do You Choose What to Automate?
List the repeated tasks, estimate the hours each takes a month and who does them, and start with the one that takes the most billable people’s time, unless it is very hard to automate.
Is Automation the Same as AI?
No. Automation follows fixed steps, such as moving hours into billing. AI handles steps where the input varies, such as drafting a project description. Many useful automations use no AI at all.
Does Automation Need New Software?
Often not. Many tasks can be automated by linking the systems the firm already uses. New software is worth buying when no existing system can hold the data.
How Quickly Does a First Automation Pay Back?
Compare the hours it frees a month with the hours to build and look after it. For example, 40 hours to build and 20 hours freed a month pays back in about two months (40 ÷ 20), if the freed hours go to billable work.
Sources
Updates
- : Rewritten at full length: who still decides each task, the hours-a-month sum and a worked ranking of four tasks, automation against AI, building on the systems already paid for, designing the exceptions, measuring with Deltek’s AI figure kept to AI, and a checklist; Deltek re-read.