Meta description: We built an AI advisory council that analyzes our business every night and delivers actionable recommendations by morning. Here's exactly how we did it.
This morning I woke up to six detailed business analyses sitting in my Slack. Revenue strategy, growth hacking, content gaps, sales coaching, operational bottlenecks, market positioning. All written overnight while I was asleep.
No consultant fees. No meeting prep. No waiting until next quarter for a review.
We built this yesterday. Here's what happened.
The Problem With Traditional Business Advice
Most founders get advice in one of two ways. Either they pay a consultant a few thousand dollars for a report that takes three weeks to arrive, or they Google around until they find a Reddit thread that sort of applies to their situation.
Neither is good.
The real issue is that good business analysis requires holding a lot of context at once. Your current pipeline. Your content performance. Your operational health. Your market position relative to competitors. A human consultant gets a snapshot. They miss the ongoing pattern.
I'd been thinking about this for months when I watched a YouTube video by Matthew Berman. He demonstrated something he called a "business advisory council": a team of parallel AI agents that each analyze different aspects of your business from a specialist's perspective. The concept was immediately obvious: instead of one AI giving generic advice, you run eight specialists simultaneously. This is essentially AI workflow automation applied to strategic thinking. Each one has a different lens. Then a synthesizer distills it all into ranked priorities.
I want to give Matthew full credit here. His framework is what inspired the version we built for em8.io. What we did was adapt it to our specific data sources and build it into our existing automation stack.
What Is an AI Business Advisory Council?
The concept is simpler than it sounds.
You gather your live business data. CRM pipeline, content schedule, blog performance, calendar entries, historical decisions. Then you run multiple AI "experts" in parallel, each prompt-engineered to think like a specific type of specialist. They all get the same data. They each return a focused analysis from their angle. Finally, a synthesizer reads all six outputs and surfaces the three highest-impact actions for right now.
It's not magic. It's structured prompting with real data.
The key insight is parallelism. Running six analyses simultaneously costs the same as running one sequentially, but gives you six different perspectives on the same problem. A revenue strategist will notice something a content strategist misses. An operations expert will flag something neither of them sees.
The output lands in Slack every morning. It takes about 90 seconds to read. Then I know exactly where to focus.
Our 6 Experts and What They Analyze
Here's how we structured our council. Each expert gets the same data context but has a completely different system prompt.
💰 Revenue Strategist Looks at deal pipeline health, pricing strategy, retainer conversion rates, and revenue predictability. If there's a deal stalling, this expert will flag it. If we're leaving money on the table with a client, this one spots it first.
📈 Growth Hacker Focused on client acquisition velocity, lead generation efficiency, and growth experiments worth running. Thinks in channels, conversion rates, and compounding bets.
📝 Content Strategist Scans our content pipeline and published articles for gaps. Looks for SEO topics we're missing, LinkedIn and X content that could drive inbound, and whether our content calendar is serving our growth goals.
🤝 Sales Coach Reviews open leads and suggests how to move them forward. Gives specific language for follow-ups, identifies which prospects have gone cold, and flags any objection patterns worth addressing.
⚡ Operations Expert Watches cron job reliability, delivery timelines, and any system health signals. This one is surprisingly useful. It catches operational drift before it becomes a problem. A deadline slipping on a client project, a pipeline that hasn't run in 48 hours.
🎯 Market Analyst Looks at the automation market landscape, competitor positioning, and emerging opportunities in Make.com, n8n, and AI agent tooling. Helps us know when to double down on a positioning angle and when the market is moving in a different direction.
Each expert returns three specific, actionable recommendations. Not vague suggestions. Concrete next steps tied to this week's data.
Real Recommendations It Gave Us (From Today's Run)
I'm anonymizing client names, but these are actual outputs from this morning's council run on February 18, 2026.
The Revenue Strategist flagged that we have multiple active leads who haven't moved in over a week. Its recommendation: prioritize converting project-based work to retainers before closing, since monthly recurring revenue provides the predictability that project work doesn't. Specifically, it noted that our pipeline has several clients who would likely accept a "maintenance + support" retainer as an add-on.
The Growth Hacker identified cold outreach as the highest-ROI acquisition channel we're not fully running yet. It had visibility into our pipeline data showing that inbound SEO leads have longer sales cycles than direct outreach leads, and recommended starting an Instantly sequence this week rather than next.
The Content Strategist noticed we haven't published in several days and flagged that our workflow automation cost article is ranking but could be updated with 2026 pricing data to recapture position. It also suggested that a case study on a real automation build would perform better than topical content right now.
The Operations Expert caught that a scheduled task had a dependency not yet resolved: an API key we'd been waiting on. It flagged the delay proactively, which is the kind of thing that used to fall through the cracks.
None of this is groundbreaking in isolation. But having all six of these surface on the same morning, ranked by priority, means I actually act on them. The bottleneck in most businesses isn't knowing what to do. It's the cognitive overhead of keeping track of everything at once.
How to Build Your Own
This works whether you're using Make.com, n8n, or writing scripts directly. The architecture is the same.
Step 1: Define your data sources
The council is only as good as the data it sees. For us, that means: - ClickUp tasks and lead pipeline - Published and scheduled content - Google Calendar for meeting context - Daily memory files (our AI's running log)
Start with what you already have. Don't wait for perfect data coverage. Two sources is enough to start getting useful output.
Step 2: Build a data collector
Write a function that pulls all your sources and compresses them into a single context block. This becomes the shared "briefing" every expert receives. Keep it under 4,000 tokens if you can. Most of the signal is in structured summaries, not raw dumps.
Step 3: Write your expert prompts
Each expert needs a system prompt that establishes their role, and a user prompt that presents the data and asks for exactly three specific recommendations for this week. The constraint to "this week" is important. Without it, AI advisors give you evergreen strategy advice that's true but not actionable.
Here's the structure we use for each expert prompt:
You are a [Role]. Focus on: [2-3 specific domains].
DATA: [compressed data block]
Give 3 specific [domain] actions Emil should take THIS WEEK. Format each as: → [action] because [evidence from data]
Step 4: Run experts in parallel
Use Promise.allSettled() in Node.js, or parallel branches in Make.com/n8n. All six experts should run simultaneously. On Claude Sonnet 4.5 via OpenRouter, this takes about 30-45 seconds total.
Step 5: Add a synthesizer
The synthesizer reads all six expert outputs and returns only the top three priorities, with duplicates removed. This is the most important step. Without synthesis, you get 18 recommendations and decision paralysis. With it, you get three clear actions.
Step 6: Deliver it somewhere you'll see it
Slack, Telegram, email. It doesn't matter. What matters is that it arrives at a consistent time when you're in a decision-making mode. We send ours at 07:00 UTC so it's there when we start work.
If you're using Make.com, this is a scenario with six parallel HTTP modules calling your AI provider, then one final HTTP module for synthesis, then a Slack action. If you're on n8n, it's nearly identical. We covered the Make.com vs n8n decision in detail if you're still choosing.
Results and ROI
We've been running a version of this for a few weeks and the honest answer is: the value isn't in any single recommendation. It's in the consistency.
Before this system, business strategy happened reactively. Most founders understand business process automation for operational tasks — but few apply it to strategic review. A slow week prompted a retrospective. A new opportunity triggered a sprint. Everything was driven by what was most visible, not what was most important.
Now the council runs whether or not I'm thinking about strategy that day. The Revenue Strategist catches pipeline drift before I notice it. The Operations Expert flags delays I wouldn't have spotted for another three days. The Content Strategist reminds me of the article I said I'd publish this week.
In terms of actual numbers: one pipeline flag from the Revenue Strategist prompted a retainer conversation with a client we were about to close on a fixed project. That conversation added roughly €800/month in recurring revenue we would have left on the table. That's more than enough ROI for the entire system.
The build took one day. The cost per run is under $1 in API fees.
The automation ROI math is straightforward: a single useful recommendation per week justifies the infrastructure entirely.
FAQ
Do I need to be technical to build this?
If you can use Make.com, you can build a basic version of this. The architecture is just HTTP requests to an AI API inside a loop. If you want to run it as a scheduled Node.js script like we do, you'll need basic JavaScript knowledge, but nothing advanced.
Which AI model should I use?
We use Claude Sonnet via OpenRouter for the expert analyses, and Claude Opus for the synthesis step where quality matters most. GPT-4o works equally well. The prompt engineering matters more than the model choice.
What if my business is different?
The six experts we use were designed for an automation agency. Your business needs different specialists. A SaaS founder might replace the Sales Coach with a Churn Analyst. A creator might add a Platform Strategist. The framework is modular. Swap the roles to match your reality.
How do I prevent the AI from giving the same advice every day?
Make sure the data context changes daily. If the AI sees different pipeline data, different content metrics, and different calendar events each day, the recommendations will stay fresh. Static prompts with static data produce static output.
Can I add a feedback loop?
Yes, and this is where it gets interesting. We built a learning layer that records which recommendations we acted on and which we rejected. After a few rejections with similar patterns, it auto-generates filter rules. The council learns not to recommend TikTok to a B2B-focused agency. Over time, the signal-to-noise ratio improves without any manual tuning.
What's Next
We're actively improving this system. The next version will pull in Google Search Console data so the Content Strategist can see actual ranking positions. We're also experimenting with having the experts "argue" with each other before synthesizing, a debate step that surfaces edge cases the first pass misses.
If you're a founder spending time on operational review that could be automated, this is one of the most valuable builds you can run. A few hours of setup. Daily output. No recurring consultant fees.
If you want help building your own AI advisory council, or integrating it into your existing Make.com or n8n setup, we build exactly these kinds of systems at em8.io. The examples of business automation we've shipped give you a sense of how we approach these builds.
Start with two experts and one data source. Add more once you've seen it work. The architecture scales naturally.
Built at em8.io. AI automation for founders who want to work smarter, not just faster.