Files, research and planning in the AI build

Overview

When you start an AI-built campaign, Ken works through three preparatory steps before it touches any lead data: it reads your files, researches your business, and drafts a campaign plan. These steps give Ken the context it needs to write copy that sounds like you, target the right audience, and lead with the problems your prospects actually care about.

All three steps happen inside the campaign build workspace at /campaigns/build. The header shows an 8-dot step rail. Steps 1, 2, and 3 correspond to Files, Research, and Planning respectively. The right-hand panel updates live as each step completes.

Before you start

• You need an active Ken workspace. If you are setting up your account for the first time, complete workspace onboarding first.
• Your company website must be publicly accessible. Ken scrapes it automatically during the Research step. A site that is behind a login wall, a "coming soon" page, or a parked domain will not yield useful research.
• Files are optional but recommended. The more material you give Ken, the sharper the research profile and the more specific the campaign angle.

How to upload files, review research, and approve the plan

Step 1 - Upload your files

When you open the build workspace, the right panel shows "Your files". This is the file upload area for the current build.

1. Click anywhere in the dashed upload area, or drag files directly onto it.
2. Select one or more files. Accepted formats: PDF, DOCX, TXT, MD, CSV, JSON, HTML.
3. Each uploaded file appears in the list with a status badge:
   • "Reading..." - Ken is extracting the text (OCR for PDFs).
   • "Read" - the file has been processed and is ready.
   • "Couldn't read" - the file could not be parsed. Remove it and try a different format.
4. You can delete any file using the trash icon next to it.
5. If you have uploaded files in a previous session or in Settings, they appear here labeled "On file". You do not need to re-upload them.

When you have no files, the panel shows: "Drop a deck, one-pager, or customer list here, or click to browse. These inform every campaign you build."

The subtitle below "Your files" tells you how many files Ken has actually read - not just attached. A file that is still being processed is not counted until it completes.

Once you are satisfied, click "Start" in the chat to move to Research.

Step 2 - Review the research profile

While Ken researches your business, the right panel shows "Ken is researching your business" with a loading state. When research completes, the panel is replaced by a structured company profile with these sections (any section Ken could not populate is omitted):

• Company name, one-liner, and elevator pitch - a one-to-three sentence hook.
• "What you do" - a plain description of your product or service.
• "What you sell on" - the value propositions Ken will lean on.
• "Problems Ken will lead with" - each problem shows the pain, and the outbound angle Ken plans to open with.
• "Why you win" - competitive advantages and what makes them different.
• "Proof" - customer case studies and outcomes.
• "Your assets" - lead magnets, guides, or other resources.
• "Who you sell to" - the target audience archetype.
• "Context Ken personalizes with" - the company context saved to your workspace. Below this it reads: "Saved to your workspace and used across all your campaigns."
• "Sources" - the web pages and files Ken drew from.

Read through the profile. If anything is wrong or missing, you can tell Ken in the chat. The research profile is the source of truth for everything downstream - copy claims, personalization, and targeting all trace back to it.

Step 3 - Review and approve the plan

While planning runs, the right panel shows "Ken is planning your campaign". When planning finishes, four cards appear:

• "Who we're targeting" - a short description of the ideal customer profile.
• "Your offer" - the offer statement, the CTA type (Booked call, Product demo, Positive reply, or Free audit), and the goal metric.
• "Messaging angle" - the hook and framing the email sequence will lead with.
• "Goal" - the campaign objective.
• "Segments to target" - each audience segment Ken proposes, with a rationale for why it is a distinct group.

Review the plan. If you want to adjust the offer, change the targeting, or drop a segment, say so in the chat. Ken will re-plan. When you are happy with the plan, confirm it in the chat to proceed to the Ideal customer step (step 4).

How it works

File parsing and OCR

Uploaded files are stored at the workspace level, not per campaign. Every campaign you build from that workspace can read the same files. PDFs go through an asynchronous OCR pipeline. The status badge on each file polls the backend until the file either reaches "Read" or fails. If OCR is still running when you start the build, Ken waits for it to complete before using that file.

Research synthesis

Ken reads two sources in parallel: the files you uploaded, and your company website. For the website, a background worker fetches the pages most relevant to understanding your business - Home, About, Pricing, Case Studies, Services, and similar page types. It strips navigation, footers, scripts, and cookie banners before passing the readable text to the AI, so the model works from signal rather than chrome.

The research step synthesizes everything into the structured profile you see in the right panel. The synthesis follows a strict extract-don't-invent rule: every claim in the profile must come directly from your site or your files. If a section has no source material, it is omitted rather than filled with a guess. This is intentional - downstream copy is fact-checked against the research profile, so the profile must be trustworthy.

A campaign-agnostic company context string is generated at the end of research and saved to your workspace. This context covers your voice, positioning, and who you serve - without campaign-specific metrics or customer names. It applies automatically to every campaign you build, so you do not need to re-describe your business each time.

Campaign planning

Planning takes the research profile, your stated offer and goal, and decides: who to target, how to segment that audience, what messaging angle to lead with, and what the campaign structure looks like. The planner enforces a one-ICP, one-offer, one-campaign rule. If you have two distinct offers or two unrelated audiences, they become two separate campaign builds - not one mixed campaign - so the analytics stay readable and you can learn from each one independently.

The segment list in the plan becomes the scaffold for the Qualify and Segment steps that follow. Each segment gets its own strategy, email sequence, and AI personalization prompts later in the build.

Troubleshooting & debugging

Symptom: "Ken is researching your business" loading state does not go away.

• Likely cause: the website scraper could not reach the site, or all requested page types came back empty.
• What to check: confirm the website URL in your workspace settings is correct and publicly accessible. Parked domains, "coming soon" pages, and login-gated sites return no usable content. If the site is real and public, the scraper may be in a temporary backlog - wait a few minutes and refresh.

Symptom: The research profile looks thin - most sections are missing.

• Likely cause: the company website has very little text content (e.g., a single-page site with mostly images), and no files were uploaded.
• What to check: upload a one-pager, pitch deck, or any document that describes what you do and who you serve. Tell Ken in the chat what was missing and it can ask you questions to fill in the gaps directly.

Symptom: A file shows "Couldn't read" in the status badge.

• Likely cause: the file could not be parsed. This typically happens with scanned PDFs that have no embedded text layer, password-protected files, or very old document formats.
• What to check: convert the document to a plain PDF with embedded text, or copy the text into a .txt or .docx file and re-upload.

Symptom: A file stays on "Reading..." indefinitely.

• Likely cause: OCR for large or complex PDFs can take several minutes. The frontend polls the backend every few seconds and will update automatically. The poll stops after about 5 minutes if the backend never confirms completion. At that point the badge stays "Reading..." without failing.
• What to check: wait up to 5 minutes. If it still shows "Reading..." after that, refresh the page. If the file is very large, try splitting it into smaller parts.

Symptom: The plan shows the wrong offer or CTA type.

• Likely cause: Ken inferred the offer from your site and files. If your actual offer differs from what is on your site, the plan will reflect the site.
• What to fix: type your correct offer in the chat ("My offer is a free 30-minute audit, not a demo") and Ken will replan with that input.

Symptom: The "Segments to target" list is empty or has only one entry.

• Likely cause: the research profile did not surface enough distinct audience sub-groups, or the offer is highly uniform across all buyers.
• What to check: review the "Who you sell to" section in the research profile. If the target audience is described very broadly, tell Ken in the chat which sub-groups you want to reach - by industry, title, company size, or stage - and Ken will re-plan with those distinctions.

Symptom: Scraped pages appear in the "Your files" list with a globe icon and an orange "Scraped" badge, but the research profile is still thin.

• This is expected behavior. Scraped pages are included as source material and show up in the file list so you can see what Ken read. If the scraped content is sparse, supplement it with uploaded files.

Symptom: "On file" appears on files you did not upload in this session.

• This is expected. Files uploaded in a previous build session or added in Settings carry the "On file" label. They are already available to Ken and do not need to be re-uploaded.

FAQ

Do I need to upload files to use the AI build?

No. Files are optional. Ken will work from your website alone if no files are uploaded. Uploading a deck or one-pager gives Ken more material to work with and typically produces a sharper research profile and a more specific campaign angle.

What file formats are supported?

PDF, DOCX, TXT, Markdown (.md), CSV, JSON, and HTML. For PDFs, text-layer PDFs are processed immediately. Scanned PDFs go through OCR, which takes longer. Password-protected files cannot be read.

Can I upload files after the build has started?

Yes. You can attach files via the chat at any point during the build. Ken will acknowledge them. If you want the research profile to reflect new files, ask Ken to re-research after uploading.

Are my files shared across campaigns?

Yes. Files are stored at the workspace level. Every campaign you build from the same workspace has access to the same files. Deleting a file in the build also removes it from your workspace.

Why does the plan show segments I did not ask for?

Ken proposes segments based on the audience sub-groups it found in the research profile. If a segment is not relevant to your goal, tell Ken to remove it or replace it with a different cut of the audience. You can name the specific grouping you want - by industry vertical, company size, job function, or any other dimension.

Can I run two different offers in one campaign build?

No. The AI build enforces one offer per campaign. If you have two distinct offers or two unrelated audiences, start a separate build for each. This keeps the campaign analytics clean so you can compare results and learn from each independently.

Related articles

• Create your first AI campaign
• Ideal customer profile and lead targeting in the AI build
• AI build - email copy review
• Knowledge files and workspace settings
• Campaigns overview