Indie Strategist
Give AI enough context to be genuinely useful. Then install that inside your client's business, so both of you execute to the same standard and they can keep running the work consistently without you.
Kevin Whelan · September 2026
Context → Data → Method → Execution → Improvement
| 1 | Two problems. The model has no context, and the capability never gets installed in the client's business. |
| 2 | Who owns what. What you hand over, and what the agreement has to say. |
| 3 | Context. The nine areas worth writing down, and the guardrails that keep the output safe. |
| 4 | Data. Everything I integrate. |
| 5 | Method. Your own library of frameworks, processes, skills, templates and tools. |
| 6 | Execution. Three ways to sell the doing, what a retainer says, and what AI-first builds look like. |
| 7 | Improvement. Using every project to sharpen the process. |
| 8 | The tools I run, and what they cost. |
| 9 | What you can take away. |
Problem one
Most businesses ask for work, get something generic back, and blame the model.
To get output that actually works the way you want it to, you have to give it the necessary context.
Context about the business, and about how you work.
This is the foundation. Everything downstream gets better once it exists, and nothing downstream works properly until it does.
Problem two
Whatever gets written down usually lives in whatever tool the last consultant preferred, or in their head. The client ends up with output and no capability.
Aim at the opposite. Leave them able to run their own marketing, with the tools, systems and processes you built alongside them.
Make yourself a little redundant. There is always more work, and what they are paying for is your judgment. Deciding what to do next, iterating, pushing the thing forward. Not holding onto context or IP.
Install as much as you can in their business, and charge a premium for it.
Problem two
Handing over your life's work raises the obvious question. What stops them reselling it, or their next freelancer vacuuming it up.
The answer is the agreement, not holding things back. Mine now says roughly this.
Legal protection only works if you would actually use it. What it really gives you is the confidence to be generous, which is the whole point.
The shape of it
| Part | What it holds | How often it moves | Whose it is |
|---|---|---|---|
| Context | What's true here | Slowly | The client |
| Data | What's happening now | Constantly | The client |
| Method | How the work gets done | When you improve it | You |
| Execution | The work, running | Daily | Either of you |
| Improvement | What the work taught you | Every project | Shared |
Context, data and everything execution produces belong to the client outright. Method is yours, because it is the only part that isn't about one specific business. Execution runs on whichever side makes sense, and that choice is a commercial decision rather than a technical one.
Context
What everything is called and what counts. A lead, an opportunity, a customer, a cancellation. The counting rule, and where the number lives.
Where the business is going. Goals, positioning, the plan for the year, and what it has decided not to do.
What it sells. Every product and service in detail, what's included, the price list, packaging and discount rules.
Who buys and who else is selling. Customer profiles, segments, competitors, category research, objections, proof.
How it makes money. Revenue model, margins, what a customer is worth, what you can spend to win one, targets.
How it looks and sounds. Visual identity, style guide, writing voice, the asset library and the rules for using it.
How the business actually runs. Team and roles, who decides what, how the thing gets delivered, service standards.
Where the work and the numbers live. The tech stack, who owns which account, and links out to the KPI sheets and dashboards.
Deep research on the category, the competitors and the buyer. Sentiment analysis from reviews and listing sites. Sales call transcripts, summarised and anonymised.
Two borders worth fixing so nothing gets filed twice. Prices live in Offer and the economics behind them live in Money. Objections live in Market, because they're about the buyer.
Context
Accuracy above everything. Tell me when you are assuming, when you are making a judgment call, and when you are not sure. Pose it as a question rather than filling the gap.
Never throw a person under the bus, however bad something looks. Name the problem, not the people.
Security first, because the worst failure is one client's name surfacing in another client's report.
Customer and employee personal data. Credentials and API keys. Anything under someone else's NDA. Raw exports with names in them.
For the sensitive rest, link to the file instead of pasting its contents, use ranges where the exact figure isn't needed, and keep one client per workspace so nothing crosses.
Never fill a gap with a sentence that sounds right. A document that says "we don't know this yet" is safe to build on. A guess spreads into everything you make from it.
Context
Give it to someone who has never worked there. If they can do the work without asking you a question, it's finished.
Three that tell you quickly.
Whatever comes back thin tells you which of the eight areas to write next.
Data
Google, Meta, LinkedIn, Microsoft, TikTok, Amazon.
GA4, plus PostHog or Hotjar where they run it. Microsoft Clarity for recordings and heatmaps.
Search Console, Ahrefs, Google Business Profile.
Instagram, Facebook, YouTube, LinkedIn, TikTok.
HubSpot, Salesforce, Klaviyo, Kit, GoHighLevel.
Shopify, Stripe, PayPal, Amazon Seller Central.
QuickBooks, and whatever the bookkeeper actually uses.
Google Sheets, BigQuery, Snowflake, Postgres.
The industry system they run on, project management, call notes, email and calendar, design files, hosting.
Pulled, never typed. A number typed into a document is wrong by next week, and it's wrong quietly. Context holds what a lead means. Data holds the count.
Method
Build a library of processes and playbooks you can roll out across every project you work on.
All the things you already know how to do, written down once, refined, and then used as tools rather than remembered. That is the difference between being good at the work and owning something.
Ask for an audit with no method and you get the model's idea of one. Ask with a method and you get yours. Your format, your known traps already handled, your output already shaped.
Method
There is no blank page here. The list already exists, in whatever you use to track client work.
The first one is slow. The payback is on client three. That is the economics of an archive, and it is why the boring step is writing it down rather than doing it.
Method
A way of thinking about a problem. A lens you can point at any client.
A sequence someone follows to get a known outcome. Steps, order, who's needed.
A programme that runs over time rather than a one-off job.
A procedure the AI runs. The format, the data sources and the traps built in.
The shape of the finished thing, ready to fill.
Finished work to point at. Showing the standard beats describing it.
Plus tools, meaning something you built that runs. Scripts, generators, publishers, checkers. Sorted by what the thing is, not by what it's about, because a topic sort puts half of everything in two piles.
Execution
You hand over the template, the written process and a worked example. Their team runs it from then on.
Billed as a build with an end date. They keep a capability, which is easier to approve than more hours.
A monthly execution retainer where you ship the priority work yourself, alongside the advisory relationship.
Billed separately from strategy, so there is no monthly negotiation about which one you are doing.
You operate the processes on an ongoing basis. Closer to a marketing operations function than a project retainer.
Not settled yet, and not priced. Clients are already asking for it, which is usually the signal.
The open question is how the install itself gets sold. A fixed upfront project that sets everything up in ninety days, or rolling it out piece by piece as the client needs it. I am leaning towards rolling it out, because a big bang install teaches nobody anything.
Execution
"Strategy isn't your bottleneck. We know what should happen next. What slows things down is that the work sits waiting for someone to have time."
| Two projects a month | A page, a campaign, a new template, a proof of concept. A typical month, not a hard quota. |
| The website, managed | Small iterations and fixes are background work inside the fee, so the site never goes stale waiting for someone. |
| Processes, not one-offs | Where the win repeats, the project is the process. Written down, templated, proved once. |
| New systems, pushed to them | As you build new processes, they get deployed to the client rather than waiting to be asked for. |
| Where the line sits | Named in writing. Not an open task queue. Anything needing a person runs through their team. |
Today that runs at $2,500 a month on top of a base retainer of around $5,000. A few clients have moved onto it. A full audit engagement starts at $10,000. Building the whole AI system inside a client business is the one I have not priced yet.
Execution
What changes is who can build one, and how fast. These were single-operator builds, start to finish.
| A trafficked site, replaced | Page for page off an ageing CMS. Payload went 61.5MB to 9.1MB and the desktop performance score 33 to 93. |
| A 151-page site | Every word editable by the client, deploying the moment it's saved. |
| A nine-page rebuild | Matched to the original by measuring the rendered page rather than guessing at it. |
Weeks rather than quarters, with no developer in the loop to wait on. The client stops choosing between a cheap site and a good one.
Execution
The work is the template and the rules. Once those exist, writing the next send is the short part, and it is the part you can hand back.
Execution
Content was the first thing everyone reached for, and it is the thing to slow down on hardest.
Words carry the trust. Get them wrong at volume and you spend a year earning it back. The same goes for design direction, where a generated look that nearly fits is worse than no look at all.
The long tail of execution is the commodity. Automate that. Keep a human on the voice and the visual direction, and let the system carry everything underneath.
Execution
The client gets a link, not an attachment. It updates in place and their link still works, so nobody is reading version four while you're discussing version six.
Improvement
Every project you run is information about the process that produced it. Feed that back and the process gets sharper each time, which is what turns a pile of documents into something that compounds.
| From | To | What happens |
|---|---|---|
| Execution | Method | A step that didn't work gets fixed in the procedure. Versioned, so you can see what changed and roll it back. |
| Execution | Context | A fact you learned goes into a queue, and a person approves it before it lands. |
| Data | Context | Never. A reading is a measurement, not a decision. |
Nothing writes into context on its own. Without that rule the context layer fills up with claims nobody checked, and it stops being something you can trust.
The stack
| Claude | Where the work happens. Reads the context, runs the methods, writes the output. |
| ChatGPT | Image generation and a second opinion when an answer looks thin. |
| Perplexity | Deep research. It runs across the models and comes back with better sourcing. |
| Windsor.ai | One pipe for platform data. Ads, analytics, search, listings, social, CRM, commerce, finance. |
| Ahrefs | Keywords, backlinks, rankings, site audits. |
| Microsoft Clarity | Recordings and heatmaps. What people actually do on the page. |
| Granola | Call notes without taking them. They feed the context layer. |
| Higgsfield | Generated graphics and imagery for client sites, where stock or a shoot doesn't fit. |
| Trello | The playbook board. The index of every process. |
| Obsidian and Drive | Where the documents live. Plain text, owned outright. |
| Supabase | The warehouse. Data lands on a schedule and the database keeps each client apart. |
| Cloudflare and Netlify | Hosting for the sites and the reports, and the gate in front of them. |
| Kit | The newsletter. |
| Figma | Design files, read straight into the build. |
Roughly $300 to $500 CAD a month all in, and Claude is most of it because I burn credits building sites and email. The list moves. What matters is that the context, the method and the output sit in plain text I own, so swapping any line of it costs nothing.
Take this with you
Everything referenced in this talk, in one place.
| The handout | /handout |
| A starter library | /library |
| An example scope of work | /scope |
| A sample report | /report |
| The recording and transcript | /recording (private) |
Kevin Whelan · kevin@kvnw.com