Verikal.
← Back to library

PLAYBOOK

How to scope an AI project in 48 hours

The exact questions we ask in a discovery call, and the diagnostic we send back.

Published May 8, 2026 · 8 min read

Most “AI strategy” engagements take six weeks and end with a forty-page deck nobody reads. We do something different. A 30-minute call. A written plan within 48 hours. The plan is yours to keep, whether you hire us or not.

This is what’s in the call, what’s in the plan, and what we’re deliberately not doing. Steal it for your own engagements if you’re consulting. Use it on yourself if you’re a founder trying to figure out where to start.

What the call is for

We have one job in 30 minutes: find the biggest leak. Not all the leaks. The biggest. The one that’s costing the most time, the most money, or the most attention this quarter.

We are aggressively not trying to:

  • Map your full operation. (Six-week engagement. Not this.)
  • Pitch you on a service line. (We pitch at the end of the plan, in writing, with the math.)
  • Get you to commit to anything on the call. (The 48-hour written plan does the persuading.)
  • Show you our credentials. (You can read those before the call. We don’t want to spend the time on it.)

That sounds like restraint. It is. The single most common mistake in consulting scoping is trying to do everything in one pass. You end up with a generic plan that impresses nobody and ships in nothing.

The eight questions

We ask these in roughly this order. Sometimes we skip one. Often we go deep on one for fifteen minutes and the rest pour out as side commentary.

1. Walk me through last Tuesday.

Specific day, specific person. Whoever’s on the call walks us through their actual workday. Not a typical day. The most recent specific Tuesday. Specifics force truth out. “A typical day” gets you marketing copy. “Last Tuesday” gets you the bit where you spent two hours fixing a quote that shouldn’t have needed fixing.

2. Where in there did you wish a person could’ve stepped in?

Not an AI. A person. We’re looking for the moment of resentment — the moment where the founder says “I shouldn’t be doing this.” That moment is almost always the leak.

3. Where do things drop?

Tickets that don’t get answered. Quotes that don’t get sent. Invoices that don’t get chased. Customer onboarding steps that get skipped. The question is which kind of thing falls through the cracks the most often. There’s usually a clear winner.

4. What lives in WhatsApp?

We ask this even of companies that swear they don’t use WhatsApp for business. They do. Everyone does. The question is which work artifacts — jobs, approvals, decisions, hand-offs — live there permanently and not in a system. That’s where memory leaks out of the company.

5. What’s manually copied between two tools?

Almost every business has at least one. Often three. Order from Shopify to accounting. Lead from form to CRM. Quote from Google Doc to PDF to email to filing cabinet. Each manual copy is a place an agent could live.

6. What does your team ask you that they could ask anybody?

The questions that come back to the founder over and over — pricing clarifications, policy lookups, “is this customer special?”, “what did we promise them last time?” — are knowledge that should live in a tool, not in a person’s head. The follow-up agent for this is small and high leverage.

7. What does it cost you when each of those things go wrong?

We need an order of magnitude, not a precise number. Is the leak costing you a thousand dollars a month, or fifty thousand? The plan ranks leaks by cost, not by how irritating they are. (The most irritating leak is rarely the most expensive one.)

8. If we shipped one agent in the next four weeks, what would have to be true on the other side?

This is the closing question and it’s the most important one. It surfaces the constraints we couldn’t guess at. (“It can’t touch our SAP instance.” “Legal needs to review any customer-facing copy.” “Our COO is on holiday for three weeks.”) Plans that ignore those constraints get politely shelved.

The 48-hour plan

We send a one-pager. Not a deck. Not a Notion page. A markdown document, in writing, usually under a thousand words. It has five sections.

1. The leak

One paragraph. What we heard, in our words. We name the specific work that’s leaking, the volume per week or month, and the rough cost. If we’ve mis-heard, you fix this section before we go further. Everything else hangs off it.

2. What the agent would do

Two to four paragraphs. The behavior of the agent, in plain language. The triggers, the read of the world, the rules layer, the model’s job, the guardrails. We try to make this feel like a colleague describing the work, not a pitch.

3. Three options

Always three. Not one. Not five.

  • DIY. What it looks like for your team to build it themselves. Tools we’d use. Rough effort. Risks. We’re honest here even when it costs us the engagement.
  • We build it once. Fixed scope, fixed price. What’s in. What’s out. When it ships.
  • We build it and stay on. Build plus monthly retainer. When this makes sense (and when it doesn’t).

Three options force you to evaluate trade-offs. One option feels like a sales pitch. Five options is paralysis. Three is the right number.

4. What we’re explicitly not doing

The other leaks we noticed but aren’t addressing. Why each is real but not first. This is also where we name anything we’re skeptical of in the request — if you came in asking for an “AI strategy presentation” and we don’t think that’s the right thing, we say so here.

5. Numbers

Costs, timelines, who does what. If we’re building, this section has the line item. If you’re building, this section has “your team, four weeks.”

Two days. Not three. Not five.

We commit to 48 hours because we’ve found that anything longer rots. The call is fresh in everyone’s head for about two days. If the plan lands on day six, it’s entering a different conversation than the one we just had. Day two, you read it and remember every example we used. Day six, it’s a document from a stranger.

It also forces us to be brief. A 48-hour deadline is incompatible with a forty-page deck. It’s compatible with a thousand-word plan. The shorter document is usually the better one.

What you can steal

If you’re scoping projects yourself — AI or otherwise — here’s the part to take with you:

  1. Specific days beat typical days. Always ask about last Tuesday.
  2. Find the biggest leak before you find all the leaks.
  3. Three options. Not one. Not five.
  4. Write down what you’re explicitly not doing.
  5. Ship the plan in 48 hours, even if it’s short.

And if you want us to do it on your business, the call is free and the plan is in writing within 48 hours. That’s the offer.


Found a leak in here? Let’s find the bigger one.

Book a free 30-min call
How to scope an AI project in 48 hours — Verikal — Verikal