AI Can Draft Your Proposal in Two Hours. It Still Can’t Estimate Your Project

Somewhere in your agency right now, a brief just landed on a Tuesday. By Friday, a full offer needs to be out the door. In between lies the usual gauntlet: an introductory call with the lead, aligning calendars across developers, designers, and PMs, all while trying to write something that is both realistic and clear.

That is how pre-sales have worked for a long time. 

Enter AI. It has finally sped up the part that always hurt the most: the first draft. Feed it a brief, an email thread, and some scattered call notes, and in a few hours you have a structured proposal with project phases, open questions, and alternative scenarios.

But generating a document faster isn't the same as writing the right one.

What AI actually changed 

To understand what changed, it helps to look at the traditional pre-sales machine. A brief comes in. The Business Developer (BD) organizes a call, gathers the info, and pulls the whole team into an internal meeting. Everyone sizes up their part of the work, bringing their own perspective, doubts, and a bit of misplaced optimism. The BD orchestrates it all together, and in a best-case scenario, the process eats up three to five business days.

That pace came at a heavy hidden cost. Your senior people spent days bickering over estimates instead of working hours on active projects. When those inquiries quietly died and never converted, those were unbillable hours the agency simply had to swallow.

At first glance, it looks like AI completely reshuffled how roles are distributed in pre-sales. On paper, developers are no longer the starting point; they’re just validators. Designers don’t need to sit in on early meetings, and internal reviews happen asynchronously. The BD becomes far less dependent on the rest of the team and can respond to a lead faster.

But is that actually how it works in practice? Or is it just an illusion of efficiency?

Why an estimate is still just an estimate

Here is where we hit the core problem. AI works with the information it's given, and in most cases, that information isn't fully defined yet.

Without highly specific input, AI will estimate a standard project, not the actual project sitting in front of your team. It easily recognizes a typical WooCommerce or Laravel build, but it can't judge the sheer complexity of a legacy ERP integration unless someone explicitly tells it. If your senior developers can't say with confidence how complex a piece of work will be before a proper technical discovery, AI certainly has no way of knowing it either.

And throwing more text at it doesn't fix the problem.

Research on large language models shows that a bigger context window doesn't make a model less likely to invent an answer. Finding the right paragraph in a brief and pulling the correct commercial number out of it turn out to be two entirely separate skills. Feed AI the entire project history, and it can still hand you back a confident but completely wrong estimate.

There are dozens of things that need to be decided before an estimated number means anything at all: which exact technology to use, what hidden integrations lurk in the background, the client's actual budget limits, and what the internal approval politics look like on both sides.

We saw this play out recently with a lead who came in for a complex eCommerce project: multiple services, multiple products, a heavy booking system… On paper, the brief looked rock-solid. But on the introductory call, we found out the lead was in the final stages of signing with a third-party provider offering a combined ERP, CRM, and payments solution in one deal. That single piece of information changed our entire approach. Integration effort, outline, tooling - all of it depended on a decision the client hadn't even made yet.

The technical solution isn't something you can just "decide later." It directly dictates the scope, the quality, the price, and the timeline. A lead comparing a Shopify quote, a WooCommerce quote, and a Laravel quote for the exact same brief isn't comparing three completely different projects that just happen to share a cover page.

The new failure mode: confident, wrong, and perfectly formatted

Traditionally, the risk in estimating was out in the open. Everyone involved knew an estimate built on limited information had a margin of error baked in.

The new risk is much more dangerous because it's invisible. A neatly formatted AI document with a clean structure doesn't mean the numbers behind it are grounded in reality. If you never told the AI your hourly rate, margin targets, how long your internal QA takes, anything specific about that client's backend setup or how many revision rounds are actually included, it doesn't know. It simply fills the gap with an assumption that sounds exactly as confident as a fact.

The biggest risk today is a BD sending an AI-generated offer to a lead without passing it through a technical review. There's a tangible cost when someone on your team can't answer a follow-up question because they don't actually understand the document they just sent. It happens far more often than agencies care to admit.

Cutting the core team out of the early phase gives you the appearance of speed, while at the same tome isolating the person writing the proposal from the people who actually have to ship it.

There's a more subtle problem too: people can tell when they're looking at a template. It shows in generic corporate phrasing like "this solution will streamline your operations," in sentences that share the exact same robotic rhythm, or in a tone that stays a little too polished for a gritty document about budgets and timelines. In a recent survey of over 660 tech readers, more than three-quarters said they stop reading the moment they suspect AI wrote something.

The introductory call remains irreplaceable

Lately, our client calls keep hitting the same wall: people wondering why an integration or design sprint takes weeks when AI is supposed to handle it instantly.

It's a fair question built on a deeply unfair premise. AI shortens the distance between a brief and a first draft. It doesn't shorten the distance between a first draft and a hard decision the client still has to make, or a custom integration that still has to be built, tested, and fixed when it inevitably breaks.

No prompt can replace what happens on a live kickoff call! The call is where you hear what the lead didn't write down: legacy platforms, internal bottlenecks, organizational politics, B2B pricing logic... That's why everyone who has to sign off on the estimate should be on that call, not reading someone's notes afterwards. Notes keep the facts and lose the part that imapacts the number.

It’s also where you form a gut judgment on whether this lead will be a reliable partner or a steady source of scope creep. That instinct, built by an experienced BD or PM over years of getting burned, isn't something you can teach an AI through a prompt.

Take a recent project we did for a large retail brand. The scope itself wasn't wildly complex, but early on, the client was upfront about their heavy internal decision-making process and the fact that they expected ongoing updates along the way. We used a Time & Materials model to keep the structure flexible, but T&M wasn't the real key to success. The crucial factor was keeping our design and engineering leads deeply embedded in the planning process from day one. That continuous team presence allowed us to absorb client shifts smoothly, resulting in far better process than any rigid handoff would have allowed.

Under a few conditions...

Yes, we use AI in our pre-sales workflow every day, and we aren't looking back. It handles the heavy lifting of the initial outline. But because of everything above, we've established three non-negotiables:

• The intro call happens every time. Even if the brief looks thorough, we talk to the client. That’s where the real constraints surface.
• Variable scope means Time & Materials. If the technical blueprint isn't fully mapped out, we default to T&M. Slapping a fixed number on an under-defined project just pushes the risk further down the timeline, where it costs more to fix.
• Our engineering and design leads sit in on every client call, challenge assumptions in real time, and own the technical scope alongside the BD.

Getting an offer out faster is a real advantage. But moving faster only counts if you can stand behind every line item when the client asks the one follow-up question you didn't see coming. That's the part AI doesn't do for us, and the part we refuse to skip.