Context and Growth Strategy

Context is what turns a general language model into your marketing team. It is reused in every report, every ticket and every chat answer. The customers who get the most from Cogny are the ones who spend the first hour here, and ten minutes a week afterwards.

The layers, from strongest to weakest

The growth strategy overrides everything. A rule written there applies to every report and ticket. Custom instructions on a single report apply only to that report. Chat applies to nothing beyond the conversation, which is why rules typed into chat never stick.

Step 1: review the generated documents

When the workspace is created, Cogny researches your website and the open web and writes three documents: Industry Context (500 to 800 words), Company Information (400 to 600 words) and Competitor Analysis (five to seven competitors). Each card shows whether it is AI-generated or user-edited, with Edit and Refresh.

They are usually roughly right, not exactly right. Read all three and fix what is wrong. The typical corrections:

• The company is framed in the wrong category (a growth agency described as a marketing agency, a fintech described as a bank).
• A key competitor is missing, or a listed one is irrelevant.
• The market is wrong (data from another country or an old positioning).
• A discontinued product or an old price is still there.

Do this before enabling reports. A wrong framing here shows up in every ticket title.

Step 2: write the growth strategy

The strategy card can be generated once data sources are connected, and it improves as the agent sees your data. Then make it yours. Useful strategies are specific and short. Examples of what customers have written, generalised:

• "We do not want broad marketing. We want thought leadership and recommendations from partners. Do not propose paid social."
• "Target these 100 named accounts; they are flagged in the CRM. Everything else is secondary."
• "B2B in Germany. ICP is construction companies. Cost per lead no higher than 100 dollars."
• "Three business lines with three owners. Split insights and tickets by line."
• "We switched from consumers to businesses this year. Prior consumer data is not representative."

One customer changed a single sentence, from consumer to business focus, and the next report proposed redoing Google Ads, adding LinkedIn campaigns, and, because it reasoned about when housing-association boards meet, a seasonal email list. Rejected off-target tickets stopped appearing. That is the level of leverage this field has.

The quickest way to update it later is Home → Shortcuts → Give Strategic Direction. Type or dictate what changed ("we are pausing the German launch until Q1", "focus on monthly donors over one-time gifts this autumn"). The note is date-stamped, appended to the strategy, and the full strategy is regenerated in the background.

Step 3: fill the context tree

The context tree is a browsable set of documents the agent can read and search. Upload PDF, Word, HTML, Markdown or text (25 MB max; scanned PDFs need OCR first), or create nodes by hand with a path such as org/brand/tone-of-voice. The tree also exposes live views of tickets, conversations, meetings, site intelligence and the connected GitHub repository, so the agent can cross-reference.

What to put there, in rough order of value:

1. Compliance and restrictions. Restricted terms, regulated claims, legal review rules. A wealth-management client who was certain AI could never write compliant content loaded their rulebook here, reviewed the first batch, and now lets Cogny produce most of their content.
2. Brand voice and tone. Guidelines, example copy, words to avoid, language per market.
3. Measurement plan. What every event means, how conversions are defined, known tracking issues. The two reference non-profit workspaces both hold this: one as about 20 nodes (events, checkout steps, KPI definitions, dashboards, known problems, tools), the other as one uploaded 45,000-character document on day four of onboarding. It is the single upload that most improves data-quality and conversion tickets.
4. Ideal customer profile and named accounts. Who to go after, who to ignore, a historic customer list with lifetime value so the agent can reason about which segments to prioritise.
5. Commercial rules. Margins, minimum order values, budget ceilings, seasonality calendar, product priorities.
6. Research and strategy documents. Positioning work, past audits, agency reports.

The agent also writes here. Meeting recordings from the Tickets page (Review Meeting) save notes and create tickets. Reports write learnings into organisational memory. Keep the tree tidy: delete test uploads and stray meeting notes, because the agent treats everything it finds as true.

Step 4: core metrics

Define the numbers you run on. Covered on its own page: Core metrics.

How the agent uses context

Before a report runs, the rendered prompt includes the industry, company, competitor and strategy documents, the metric definitions, the organisational memory of previous insights, and any custom instructions. During the run the agent can browse and search the context tree and read specific nodes. Tickets are generated from the report with the same context, and the ticket agent checks past approvals, rejections and feedback before proposing something similar again.

Language is not a setting. The agent writes in the language you use and switches per market when the strategy says so.

Mistakes to avoid

• Typing rules into chat. They apply to that conversation only. Put them in the strategy.
• Never opening the context tree. Teams that skip it get generic content and tickets that ignore constraints nobody wrote down.
• Leaving the generated competitors unreviewed. Competitive angles in reports come from this list.
• Two businesses in one workspace. The context cannot describe both. Use one workspace per business model.
• Stale strategy. Revisit it when priorities change; the Give Strategic Direction shortcut takes a minute.