- The best first uses sit around the expert work, not inside it: proposals, intake, project paperwork, status reporting.
- A professional still signs the work. AI drafts and gathers; a named person reviews and decides.
- Start with one job, measure it for a few weeks, then widen.
- Your team decides what the system can see. Client files stay inside the access rules you already have.
- The real prize is your firm's own history: past proposals, past projects, past answers, made findable.
The work around the work
Ask a principal at an engineering or advisory firm where the week went. The answer is rarely "design" or "analysis."
It's the proposal due Friday. The kickoff notes nobody wrote up. The monthly status report rebuilt from three systems. The resume pages updated for the fourth time this quarter.
That's where AI earns its place in a professional services firm. Not by doing the expert work, but by clearing the ground around it.
Where AI actually helps
Proposals and RFP responses
An RFP lands on Tuesday. It asks for 40 items, half of which your firm has answered a dozen times before. Someone digs through a shared drive for the best old answer, copies it, and fixes the client name.
A proposal assistant reads the RFP, lists every requirement, and pulls the closest past answers, project descriptions and team bios from your own library. The proposal manager starts from a sorted first draft instead of a blank page. A compliance check at the end flags any requirement still unanswered.
The writing still needs a person. The searching doesn't.
Client intake and scoping
A prospective client sends a long email, two attachments and a site plan. Someone has to turn that into a scoping note: what they want, what's missing, which practice should take it.
AI can produce that note in a minute, with a list of questions to ask on the first call. The partner reads it, corrects it, and walks into the call prepared. Our inbound lead response page covers how that hand-off usually works.
Project paperwork
Meeting notes, action items, transmittals, change logs. Each one is short. Together they eat a project manager's afternoon.
A system can draft minutes from a recorded call, pull the decisions and owners into a list, and file them against the right project. The project manager approves before anything goes to the client.
Status reports and utilization
Every month someone exports hours from the time system, budget from the project system and milestones from a spreadsheet. Then they write the same three paragraphs for each client.
Once the data is connected, AI can draft the narrative from the numbers: what moved, what slipped, what needs a decision. The project lead edits it. That's a report written in ten minutes instead of an hour. See reporting by hand for the pattern.
Finding what the firm already knows
A firm with 20 years of projects has answered most questions before. The trouble is finding the answer. Search across past reports, calculations notes and lessons learned, with links back to the source, turns that history into something people actually use.
Knowledge for new staff
A new project engineer asks the same twenty questions every new hire asks. Where's the template for a transmittal? How do we name drawing files? Who approves expenses on this client?
An assistant that answers from the firm's own handbook and standards, with a link to the source, saves senior people dozens of interruptions a month. When it doesn't know, it says so and points to a person.
Where it doesn't help (yet)
Some work should stay with the professional, full stop.
Engineering judgment. Licensed engineering is performed under a professional engineer in responsible charge. The National Society of Professional Engineers describes the model law definition as "direct control and personal supervision of engineering work," and says reviewing documents after someone else prepared them doesn't meet it. An AI draft of a calculation or a design basis doesn't change who carries that responsibility. Use AI on the paperwork around the design, not as the designer.
Advice the client relies on. A recommendation to a client's board is your firm's opinion. AI can gather the material. It shouldn't write the conclusion.
Anything that needs current facts you haven't given it. A model doesn't know this week's code amendment or the client's latest numbers unless your system feeds them in, with sources.
Pricing and fee decisions. A model can summarize past fees on similar work. Setting the fee is a partner's call.
Questions to settle before you build
- Which job costs the most senior hours each month, and who reviews it today?
- Where do past proposals, project descriptions and resumes live, and which versions are trusted?
- Do any client contracts limit where their documents can go or who can process them?
- Which AI provider will process the data, and under which business terms?
- Who owns the system once it's running, and who decides when it changes?
None of these takes long to answer. Skipping them is what turns a six-week project into a six-month one.
If you're weighing several candidates for the first project, our guide on what to automate first has a simple way to score them.
How to start safely
Pick one job. Proposals are often the best first choice: the volume is steady, the old answers already exist, and the person who reviews the draft is the person who wrote it before.
Keep a person on every output. AI drafts. A named person approves. For work that leaves the building, that person is the same one who would have signed it anyway. Our guide to designing human review for AI covers how to make that review fast instead of a rubber stamp.
Measure before and after. Hours per proposal. Days from inquiry to scoping note. Time spent on monthly reports. Write down today's number before you change anything.
Let your team set the access. The system should see only what the person using it could already see. Client confidentiality terms, project-specific restrictions and ethical walls should carry over, not be rebuilt from scratch.
Check your client contracts. Some engagements limit where client data can go. Read them before client documents go into any AI tool.
The NIST AI Risk Management Framework is voluntary, but it's a useful shared vocabulary when your leadership and your clients ask how the firm manages AI risk.
What a realistic first release looks like
Take proposals as the example. In the first weeks, the team gathers the best past answers and cleans them up. The assistant is built to read an RFP and draft from that library.
Then it runs alongside the real work. The proposal manager uses it on live bids and marks what was useful and what was wrong. Those notes shape the next round.
After a month or two, you compare hours per proposal against your before number. If it helped, you widen it: more proposal types, then resumes, then project sheets. If it didn't, you've learned cheaply and can pick a different first job.
That rhythm matters more than the tool. Small release, honest measurement, then the next step.
What changes for your team
The proposal manager stops being a searcher and becomes an editor. That's a better job, and most people take to it quickly.
Project managers spend less time writing minutes and more time on the decisions in them. The minutes still go out under their name, so they still read them.
Partners see drafts sooner. That only helps if their review fits their week. A two-page summary with the three changes that matter gets read. A 40-page draft waits.
New hires get up to speed faster. When the firm's past work is searchable, a second-year engineer can find how the firm handled a similar site in 2017 without asking around the office.
Nobody's role disappears in a first project. The work shifts from assembling to checking. That's worth saying out loud early, because people worry about it whether or not you mention it.
What most guides miss
Your proposal library is messier than you think. Five versions of the same project description, two with the wrong square footage. A proposal assistant will happily reuse the wrong one. Spend the first week cleaning and tagging the best answers. It pays off on every proposal after.
Resumes and project sheets drift. The best quick win is often keeping staff resumes and project descriptions current and consistent, so every proposal starts from the right facts.
The partner is the bottleneck, not the writer. If drafts arrive faster but still wait three days for a partner's review, nothing improved. Design the review step for the partner's schedule: a short summary, the changes that matter, and one click to approve.
Tools you already pay for may cover part of it. Office suites and project systems keep adding AI features. Use them where they fit. Custom work earns its place where your firm's own process and history make the difference.
How we do it
We work alongside your team, starting with a free intro call to find the one job worth fixing first. We connect to the systems you already run, such as Deltek Vantagepoint or Kantata, with access your team grants. Then we stay on as the work grows. More on our professional services page and our AI systems work.
