Your Firm Automated the Yes

The go/no-go meeting takes 10 minutes. A county wants a team for a bridge rehabilitation, the proposal is due in two weeks, and the first draft is already done.

A year ago, that deadline would have forced a real conversation. This year someone ran the RFP through an AI tool on Thursday afternoon, and by Friday the project sheets were matched and the resumes pulled.

Saying no now feels like throwing work away. So nobody asks why the firm lost this county's last job, what that pursuit cost, or whether the engineer named on page four is free in March.

The firm says yes. Nobody actually decided.

That is what AI has done to the pursuit so far. It made saying "yes" easy. Deciding when to say "no" is as hard as ever, and it still happens in a meeting, on what a few people happen to remember.

The writing was the half AI could reach. The decision is the other half, and it runs on the firm's own record, which no drafting tool improves. It is the gap between faster people and a smarter firm, and here you can put a number on it.

What did the extra proposals actually buy?

More work than the year before, to be fair.

Deltek's 47th Clarity study of architecture and engineering firms found proposals submitted up 32 percent in 2025 and proposals awarded up 24 percent. Much of that is a rebound: the prior year's edition, on a different sample, reported proposals down 38 percent and awards down 59 percent.

The cost shows up in conversion. Capture rate is the share of proposed dollars a firm actually wins. In Deltek's study the median fell 3.8 points to 44.4 percent, the sharpest drop in its nine-year trend and the lowest since 2019, while the median win rate slipped about a point to 49 percent. Deltek says the fall likely reflects higher volume, more competition and a shift toward smaller projects.

Neither survey I'm citing shows that AI caused any of this, and I won't claim it. Unanet says many firms in its survey chased more work because of economic uncertainty, and Deltek says the softer win rate may reflect more speculative pursuits in the mix.

What Deltek does show is where the tools went. Proposals and marketing content are now the second most common use of generative AI, at 50 percent of firms, up from 39 percent. Only administrative work like meeting notes ranks higher.

Why doesn't a go/no-go process fix this?

The standard advice is to get one. Most firms already have. Deltek reports that 80 percent use a go/no-go process, though only 38 percent run it on every opportunity. Unanet's survey of about 300 US architecture, engineering and construction leaders puts it at 92 percent, with fewer than half following a formal one.

Know who is giving the advice: Deltek and Unanet both sell software to these firms, and Unanet sells go/no-go tools. To their credit, both say the decision should run on data, and I agree. But a tool can only score what the firm has recorded. Unanet's own proposal product says its output is more useful the more current and organized the source material is.

A go/no-go scorecard is a list of questions about the firm's own history: whether it has won this kind of work, whether this client buys on price or on relationships, who would run the job, why it lost the last one. A firm that cannot look up those answers is scoring on memory, which we know is often fallible.

That may be most firms. Unanet's Akshay Mahajan says 53 percent of firms in its survey lack a rigorous framework and lean on instinct and experience, or on no defined process at all. Instinct is not a flaw. It is what a firm runs on when its experience can't be looked up.

Zweig Group reports that 89 percent of firms name principals and owners as primarily responsible for business development, while project managers carry 59 percent of the effort and principals 12 percent. The people in the room for the call are often not the people the last pursuit taught.

Here is the number I would show every firm president. Deltek asked what most influences the decision to submit, and having the staff to deliver fell four points to 41 percent of firms, while probability of winning rose two points to 40 percent. In the same study, proposals grew 32 percent and staff grew 1.2 percent. Firms are bidding more, with barely more people, and fewer of them say staffing drives the decision.

What does a firm keep from a pursuit?

Every pursuit teaches something: which client reads past the cover letter, which competitor prices under you, why the county picked someone else.

Today that lesson usually goes home with whoever sat in the debrief. It is the same recall problem that slows a scope, and it arrives before the work is even won.

This is the line between a tool and organizational intelligence. An AI-augmented RFP response drafting tool makes one proposal faster, and a competitor can buy the same one. A firm that can query its own pursuit history makes every later decision better, and that history stays when the principal who built it retires.

The other side of the table is moving too. "We used to get maybe 10 responses to an RFI or RFP," Anil Chaudhry, a senior advisor at the US Department of Transportation, said onstage in October, according to FedScoop. Now, he said, it's 500. That is one federal official describing procurement in general, not design selection, where we haven't seen it documented.

But when every firm's proposal reads well, a well-written proposal proves less.

A pursuit you should never have chased now costs less to lose. It still costs principal time, a named engineer's attention, and the shot at work you would have won. And nobody gets credit for the pursuit the firm skipped.

What would you check first?

Pull your last 20 pursuits, wins and losses. For each one, answer four questions without calling anyone: who made the call, what it cost, why you won or lost, and whether the team you named is the team that showed up.

Each call you have to make marks where the firm's record stops and someone's memory starts. Count them. That count is where an AI Workstream begins: the pursuit decision as one workflow, defined end to end and measured, because a workflow nobody has written down can't be improved by anything.

Then change what you report. Track win rate, capture rate and cost per pursuit before and after AI, not proposal count. If proposals doubled and those three went nowhere, you bought speed and nothing else.

Every firm in your market can now say yes faster, so saying yes faster is no longer worth much. The advantage has moved to knowing which yes to skip, and being able to show why.

Go back to that 10-minute meeting about the county bridge. The draft is done either way. What changes is what sits beside it: why the firm lost this county last time, what that pursuit cost, and whether the engineer on page four is free in March, each one looked up rather than remembered. The meeting can still take 10 minutes, and this time someone decides.

---

Where to go next: an AI Workstream takes three to five weeks, defines one workflow end to end, and ends with a scoped recommendation for what to build. The pursuit decision is a strong first candidate, and the fee is credited toward the engagement it recommends.