A kitchen table in Egham — why I built Idukki

I spent the first eighteen years of my career deep inside enterprise software — SAP Fiori, UI5, S/4HANA, the architecture diagrams nobody asks to see but everyone depends on. Twelve of those years were at HCL AXON, on projects in nine countries: UK, Netherlands, Germany, Denmark, Switzerland, Singapore, Thailand, India. Mostly I was the person who got handed the brief "the user experience here is awful, fix it without breaking what 14,000 internal users already rely on". Then six years inside UK government, designing the systems the public actually touches — the ones you have probably sworn at while trying to renew something. That is the day job. It does not photograph well.

In July 2022 I started Idukki with a small team and a clear thesis: the place where commerce was being decided was shifting away from brand-controlled surfaces toward customer-controlled ones. Photos on a real person, fit checks on TikTok, three sentences in a review that mention how the linen breathes. If you sell things, that material is your most valuable inventory and almost nobody was treating it that way. We started in a kitchen table in Egham. Still mostly do.

The two jobs

Day job: SAP architect. Night job: founder. The Venn diagram of those two communities is, on a good day, approximately one person. The SAP world is enterprise procurement cycles, signed-off architecture decisions, audit trails and acceptance environments. The DTC world is "we want it live by Friday". I quite like both — the friction between the two has made me a better designer of either.

The day job taught me three things that bleed into Idukki:

• Reliability is not optional and it is not glamorous. Nobody writes a review of software because it did not break.
• Most enterprise UX problems are not design problems. They are data-model problems wearing a costume.
• The cost of explaining the thing is roughly the cost of building the thing. We undercount the first one.

The night job teaches you the opposite lesson on the same axis. DTC brands move at startup speed because the consequences of getting it wrong are reversible. Ship the widget. Watch the conversion graph. Roll it back if it lies. That feedback loop is faster than anything in enterprise and it sharpens you up in a way that twenty years of architecture reviews simply cannot.

Why UGC, why now

When we sketched Idukki in mid-2022 the UGC category was crowded with point tools. Galleries here, rights management there, shoppable video over there, review widgets somewhere else. Each one billed per impression. The maths added up to something embarrassing — a mid-size DTC brand could pay £4–8k a month across three vendors to do what is, structurally, one job: collect customer evidence, tie it to SKUs, make it useful on the storefront, prove the lift.

Our wedge was deliberately unromantic: one platform, one data model, every feature on every plan, billed once against your widget impressions instead of four times across four vendors — impressions in simple bands, so the bill stays forecastable. No "talk to sales" for the third widget. We were not the first idea in the space; we were the cheapest, most boring, most consolidated version of the idea. Three years on, that is still the bet.

The bigger shift since then is agents. ChatGPT, Claude, Perplexity, Gemini, vertical agents inside retailer apps — by 2026 a measurable share of buying decisions are being narrowed by software before any human eyeball lands on a PDP. Verified-buyer evidence is the substrate those agents trust. So the same UGC layer that drives your homepage gallery now also drives your visibility inside the AI shopper. Two surfaces, one input. That is the part I think a lot of brands have not yet internalised.

What we shipped first

Things I am quietly proud of, in roughly the order they shipped:

• A widget runtime under 40 KB that does not move LCP. I have an opinion about page weight that will outlast our friendship.
• A rights-management flow that gets a yes about 38% of the time without making creators feel solicited.
• A no-code embed path that means the marketing team does not have to file a ticket with engineering to ship a new layout.
• A reconciliation script that lets a CMO sit down with their CFO and defend the UGC line on a P&L. That one took longer than it should have.

Things we got wrong and had to redo:

• Our first AI tagging passed at 71% precision. We thought that was good. It was not. We rebuilt it on Claude vision; precision sits around 92% now.
• Our first pricing page had four tiers and a feature matrix nobody could self-serve. We collapsed it to simple impression bands with every feature included.
• We launched without a Klaviyo block. We assumed people would copy-paste images into emails. They did not. We shipped the block.

How I think about the product

I do not believe in feature parity for its own sake. Half the requests we get are for things competitors ship; most of those would dilute the core product. I would rather ship one extremely good lookbook layout than nine mediocre ones. I would rather have a billing model people can forecast — simple impression bands, every feature included, no per-source surprises — than one that hides the real cost behind a feature matrix.

I also do not believe that everything has to be AI. Some of what we ship is genuinely intelligent — the auto-tagging, the rights-DM template suggestions, the reviews summariser. Some of it is just clean engineering: a 37 KB runtime, a sensible event model, a JSON-LD shape that validators do not complain about. The boring stuff is half the product. People notice when it is missing and rarely thank you when it is there.

Still shipping. Still nervous.

I had a manager early in my career who used to say: "If you are not nervous before a release, you have stopped paying attention." Twenty years in, that is still true. Every Idukki release is a five-page checklist, a deploy, a five-minute breathe, a status check, and then back to the next thing. The nervousness is the bit that keeps the regressions out.

The other thing I have stopped apologising for: I am UK-based, on Windows, in a Mac startup world, vegetarian (negotiable for nobody), and a girl-dad whose evenings end whenever a four-year-old needs a story read. The startup playbook is full of opinions about what founder lives are supposed to look like. Most of those opinions were written by people who had not yet tried to do the thing while doing another full-time thing.

If you are thinking about this stack

Two suggestions, neither self-serving. First: pick one customer evidence surface that hurts the most — usually PDP or post-purchase email — and instrument it before you change anything. You cannot improve what you cannot see. Second: do not buy a UGC tool, a rights-management tool, a shoppable-video tool and a reviews tool separately. They are the same job. Buying them separately is how you end up with £8k/month and four dashboards that disagree.

If you want to compare the maths, the savings calculator is the brutally honest version (it tells you when not to switch). If you want the longer thesis, the State of UGC 2026 report has the benchmarks. If you want to argue with me, my LinkedIn DMs are open — I read all of them, I reply to most.

Built from a kitchen table in Egham. Still shipping. Still nervous. Coming up on year four. The rest is just timing.