← Back to Blog

What I've Learned in the First 3 Months of Building Stratiform AI

The honest account of building an AI content generation service for UK small businesses — including the proof of concept that proved the methodology works, and why 2026 is exactly the right moment.

D
David
Founder, Stratiform AI
30k+
Orders from AI workflow
50k+
Content views
#1
Bestselling titles
1hr
Per month to maintain

I didn't plan to write this post. But after three months of building, pitching, redesigning, and occasionally second-guessing everything, I think there's something worth saying — not to sell anything, but because building a business in public feels like the honest thing to do.

So here's what the first quarter of building an AI content generation service for UK small businesses actually looked like — where it came from, how I proved it worked before a single client came on board, and why the timing has never felt more relevant.

Where It Started

Stratiform AI didn't begin with a pitch deck. It started with an observation I'd made over years of producing high-level written outputs for senior government officials and ministers: most of the written work that businesses produce every day follows a pattern.

The structure changes. The subject changes. But the process — gathering information, organising it, drafting, reviewing, refining — is largely the same whether you're writing a property description, a service proposal, a vehicle listing, or a vendor update letter.

And yet businesses employ people to do it. People whose time costs far more than most owners stop to calculate.

"What if UK small businesses could produce professional written output at pace, to a consistently high standard, for less than it costs to produce in-house?"

That's still the model. Human supervised AI. The clue is in the name — AI does the heavy lifting while a human reviews every piece before it leaves the building.

Proving It Worked Before Anyone Paid for It

Before Stratiform AI existed as a business, I needed to answer a question I couldn't answer in theory. Could I build an AI content workflow that produced genuinely good output — consistently, at pace, with minimal ongoing effort — or was the promise of AI productivity largely overstated?

I chose Kindle publishing as a testing ground. Not because I had ambitions of becoming an author — but because it had everything I needed. A clear quality bar, measurable results, and a sector I knew nothing about going in. That last part was deliberate. If the AI content workflow worked in unfamiliar territory, it would work anywhere.

The principles I built around were straightforward. AI should never be the last set of eyes on anything. The workflow had to be designed around the output, not the tool. And the process had to be iterated relentlessly.

30,000+
Orders generated in 12 months
50,000+
Views on associated content
#1
Bestselling titles in category
1 hr
Per month to maintain everything

That last figure is the one worth sitting with. One hour a month — not because corners were cut, but because a well-designed AI productivity workflow runs almost entirely on its own once it's built properly. This wasn't luck. It was the direct product of treating AI content automation as a system to be designed rather than a tool to be used casually.

Building the Website. Then Rebuilding It.

One of the first things I learned this quarter is that your website is never finished on the day you think it is. I launched what I thought was a clean, functional site. Then I looked at it through the eyes of someone who didn't already know what Stratiform AI did, and realised it didn't answer the question that matters most: why pay for an AI writing service when AI tools appear to be free?

So I rebuilt it from scratch. The answer to that question is worth spelling out clearly. Free AI tools aren't free. Someone has to learn how to use them. Someone has to write the prompts, refine the outputs, and check the results. That someone's time has a cost — and in 2026, that cost is higher than it has ever been.

The Real Cost of Employing People in 2026

When I started reaching out to businesses in January, conversations across multiple sectors kept circling back to the same issue: the cost of employing staff in the UK has become genuinely difficult to manage.

True cost of a minimum wage employee · Per hour worked
National Living Wage (from Apr 2026) £12.71
Employer NI at 15% (threshold £5,000) £1.90
Holiday pay (12.07%) £1.54
Holiday cover £1.54
Sick leave provision (6%) £0.77
Training, recruitment & management overhead £1.00+
True fully loaded cost £19.46–£27.00

The British Chambers of Commerce found that 82% of firms said the NI increase alone would force them to rethink their plans. Business confidence fell to its lowest level in three years in Q4 2025. From 1 April 2026, Statutory Sick Pay also became a day-one entitlement — removing the buffer that previously limited short-term absence costs.

If employing someone costs £19–£27 per hour all-in, what is the real cost of using that person to write proposals, listing descriptions, service reports, and correspondence that human supervised AI document generation could produce in seconds at a fraction of the price?

What I've Learned About Which Sectors Benefit

Going in, I assumed car dealers would be the obvious starting point. High volume of listings, repetitive format, clear before-and-after opportunity. It remains a valid sector — but response rates in Q1 were lower than expected, partly due to cold email deliverability issues and partly because some dealers were already experimenting with generic AI tools.

What I didn't expect was how quickly the property sector came into focus. Self-employed estate agents running lean independent businesses are producing significant volumes of written content entirely themselves — property descriptions, viewing feedback letters, vendor updates — often in the evenings after a full day of client-facing work. One agent I researched had listed 74 properties in six months, all managed personally alongside valuations, viewings and negotiations.

Facilities management has also emerged strongly — not just for listing copy but for the broader written burden that comes with running multiple service lines. Quotes, proposals, method statements, risk assessments, service reports. All repeatable patterns. All time-consuming to produce from scratch.

The common thread

The ideal Stratiform AI client isn't defined by sector — it's defined by profile. Owner-operated. Lean team. High written output. Squeezed on time and increasingly squeezed on employment costs. The industry is almost secondary.

What's Working, What Isn't

✓  Working
Personalisation. Every message that generated a response was one where I'd done genuine research beforehand. Generic outreach returned nothing.
Before-and-after. Showing someone what their own content looks like transformed cuts through more than any abstract description.
The savings calculator. When someone runs their own numbers the conversation moves from "interesting" to "let me think about this."
Human supervision. Every time I explain that a human reviews every output, visible scepticism drops immediately.
—  Not working yet
Email deliverability. A meaningful number of cold emails landed in junk. Fixing SPF, DKIM and DMARC records is unglamorous but essential.
Volume. 15 emails per cycle isn't enough to draw conclusions. Q2 means two cycles per week across multiple sectors.
Patience. Building alongside a full-time job means the discipline of a proper weekly rhythm is still being developed.

Where We Are

One seriously interested conversation in property. One in facilities management. A redesigned website with a working savings calculator. A much clearer picture of who the ideal client is and what they need to hear.

Not a client yet. But the Kindle results prove the methodology works. The Q1 conversations prove the problem is real. And the employment cost data proves the timing is right.

The cost of employing people in the UK has never been higher. The case for reducing administrative overhead through AI document generation has never been more obvious. And the technology to do it properly — human supervised, quality-checked, consistent — is here.

"The question isn't whether AI can help. It's whether you trust the process behind it. The Kindle results are my answer to that question."

More to follow.

Start with one task. No obligation.

Most clients begin with a single clearly-defined admin pain point. Try it free — see the result — decide from there.

Try For Free →
← Back to all posts