Plenty of founders now open a chat window before they open a spreadsheet. So I ran a test. I told Claude I was starting a new B2B business and wanted to build a reliable growth plan for 2027.
Its first answer was solid. Size the market, define the ICP, and build a real model with pipeline math from leads through MQL, SQL and opportunity to Closed Won. It asked three sharp questions. I answered: pre-launch, mid-market deals of $5K to $50K, and I wanted a Premonio model.
It searched the web, decided Premonio wasn’t right for me yet, and built its own spreadsheet instead. Three tabs, four lead sources, a scenario toggle and two charts. The base case for 2027: about $904K in new ARR, 54 new customers, $420K in go-to-market spend, a $7,729 blended CAC and a 7.4-month payback.
It took minutes, and it looks like a plan. Here’s what’s actually inside it.
What it got right
Credit where it’s due. The formulas are live, not hardcoded, and it checked them. Deals are lagged by each channel’s sales cycle, so revenue ramps instead of appearing on day one. Every input sits in a blue cell you can edit.
Most importantly, it was honest. It labeled every assumption an illustrative placeholder, “not researched benchmarks,” and said the output is only as good as those inputs. That candor sums up the problem: a model built on placeholders isn’t a plan yet.
Six places it stopped short
1. The funnel it promised collapsed into one number
Its own advice said a B2B model needs conversion rates at every stage: lead, MQL, SQL, opportunity, Closed Won. The model it built has one rate per channel, lead to customer: 3% for outbound, 2% for inbound, 8% for partnerships and 1.5% for paid. The two charts show cumulative ARR and ARR by channel, with no view of the funnel itself. When a plan misses, the first question is where the funnel broke, and this model can’t tell you.
2. Budget is a label, not a lever
Each channel gets a flat monthly budget: $12K outbound, $8K inbound, $5K partnerships and $10K paid. But lead volume doesn’t depend on it. Leads start at a fixed number and grow at a fixed monthly rate, whatever you spend. Double the paid budget and you get the same leads at twice the CAC. The question every CEO actually asks, “what happens if I move $10K from paid to partnerships?”, has no answer here.
3. The aggressive scenario is free
The scenario toggle multiplies lead volume by 0.7, 1.0 or 1.3, and spend never moves. So Conservative books about $633K of new ARR and Aggressive about $1.17M, both on the same $420K. Real scenarios have a price. More leads mean more budget, more reps or better conversion, and each of those costs something.
4. It shows the problem but doesn’t act on it
The dashboard shows paid taking 29% of spend and returning 10% of new ARR, at a $13,148 CAC against a $10K average deal. Partnerships take 14% of spend and return 40% of new ARR, at a $4,165 CAC. Nothing in the model, or in the chat, suggests moving a dollar. There’s no optimization step at all; reallocation is left to whoever reads the table closely enough.
5. The assumptions are guesses, with nothing to check them against
Starting leads, growth rates, conversion, sales cycle and deal size are all placeholders. There’s no benchmark range beside any of them, and no flag when an input is out of line. Its sales cycles are one or two months for $10K to $25K deals, although its own first answer said B2B cycles often run one to six months or more. You’d only catch that if you already knew the answer.
6. It’s a snapshot, not a system
The model stops at new bookings: no renewal, no upsell, and churn is a flat 1.5% a month on total ARR. Every deal closes exactly on its cycle month, and sales capacity never appears. Most of all, nothing connects the model to what happens next. When real leads and deals arrive, you’re back in a chat window, prompting from scratch.
The twist: it told me to wait for data
When I asked for a Premonio model, Claude said Premonio is designed for companies that already have pipeline data. It suggested coming back once I had three to six months of it.
That’s the opposite of what we built. GOALS generates plans without prior CRM data, using proprietary benchmark data and conversion-rate chains that run from the first activity unit through Closed Won, Renewal and Upsell. Our homepage says so, and we’ve written about why you don’t need months of CRM data.
Then it built a plan with no data and placeholder benchmarks: the exact job it had just said needed data.
I’m not sharing this to complain. It’s a useful lesson in how LLMs work. They answer from whatever they find and infer, confidently, and they fill gaps with plausible defaults. That’s fine for a first draft of an email. It’s risky for a plan that sets hiring, budget and board targets.
Where an LLM does fit
LLMs are very good at the language side of planning: framing the strategy, asking the right discovery questions, explaining a plan to a board. What they lack is a calibrated engine underneath. That means benchmark data to anchor each assumption, a full-funnel model where budget drives volume, and a loop back to actuals.
So the answer isn’t “don’t use AI.” It’s to put AI in front of an engine, not in place of one. That’s how GOALS works. Revie, its AI assistant, runs a conversational questionnaire and turns your answers into a configured scenario with confidence scores. A benchmark report card shows every assumed conversion rate and velocity, the benchmark range behind it, and flags anything more than 1.5 standard deviations from the median. When real data arrives, Sync Actuals keeps the plan a Living Scenario instead of a one-off file.
Where this is heading: the LLM you already use becomes the front door. ⚠ We implemented an MCP server so a founder working in Claude or Gemini could ask for a revenue plan and get one back from a calibrated engine, not a blank spreadsheet.
The engine makes the plan based on deterministic, proven algorithms that causally connect lead generation activities for every lead source with all lead stages and budget requirements. And it can also handle more sophisticated modeling aspects such as “cohorting”, i.e. modeling that leads generated in, say, January, might become opportunities in May, and finally closed won business in, say, September – modeling such time delays requires more sophisticated math than LLM can provide. Or how to allocate budgets from less efficient lead sources to more efficient lead sources without violating capacity constraints, otherwise all spend will go to the one more spend efficient lead source, which is clearly not feasible. However, despite the LLM’s calculational ineffectiveness, AI can credibly remove the consultant from the front of it and the analyst from the back of it.
Seven questions to ask before you trust an AI-built growth plan
- Can I see every funnel stage and its conversion rate, from first activity to Closed Won?
- If I move the budget between lead sources, do leads, pipeline and revenue move with it?
- Does the aggressive scenario cost more than the conservative one?
- Where did each assumption come from, and how far is it from the benchmark?
- Does it tell me which lead source to fund more, or only report the results?
- Does it include renewals and upsell, not just new bookings?
- When actuals come in, does the plan update, or do I start a new chat?
If the answer to most of these is no, you have a useful sketch. Build the plan somewhere that can answer yes.
FAQ
Can ChatGPT or Claude build a B2B growth plan?
They can build a starter model quickly, with live formulas and editable inputs. In our test, the model used one conversion rate per channel, didn’t link budget to lead volume and ran on placeholder assumptions. Treat the output as a sketch, not a plan.
Do you need historical CRM data to build a growth plan?
No. A plan can start from benchmark conversion rates and velocities for your market and deal size, then be recalibrated as actuals arrive. GOALS is built to do this for B2B companies with zero CRM history.
What should a B2B growth model include?
Stage-by-stage conversion from first activity to Closed Won, budget that drives lead volume by source, realistic sales-cycle timing, renewal and upsell, and a way to compare plan against actuals.
