Nish

I've been obsessed with productivity and entrepreneurship for over 15 years—testing every system on my own ventures, coaching business owners through growth plateaus, and running workshops for corporate executives. I applied the growth methodologies I've learnt to scale a luxury ecommerce brand from $2M to $6M, launch a consumer wearable to $3.5M ARR in 18 months, and grow a consumer electronics brand to $5M in 2 years. My background: Chartered Accountant, Former Consultant (EY), INSEAD MBA, two e-commerce exits. Now I work with entrepreneurs, helping them double their revenue while working fewer hours using battle-tested growth and productivity frameworks.

We built a backlink acquisition system this week for two of our own brands, and the most important decision was the target list, not the outreach volume. Generic DA scores pick sites that rank well on paper but have no influence over what AI search engines cite. In 2026, the sites worth targeting are the ones AI tools already treat as sources when answering questions in your category — they carry weight in both organic search and AI-generated answers. Here is how the system is wired: → Pull a list of domains that AI search engines already cite when answering questions in your category — these are the highest-value link targets → Cross-reference that list against your competitors' backlink profiles to find sites linking to them but not to you → Build one tracker row per target: domain, contact, outreach date, status, and whether a link landed → Run outreach in weekly batches of ten — small enough to monitor reply rates, large enough to build signal over 30 days → After each month, retire the target types that produced nothing and double down on the source categories that moved rankings or referral traffic The tracker turns a one-off effort into a compounding system. Without it, every quarter starts from zero. I put together a free Growth Dashboard Template — the input→output map, the weekly pull prompt, and the ranking method I use to decide which inputs to double. Here is how to grab a copy: → Follow so I can DM you → Comment "GROWTH" And I will send it over.

My video editor is in one country. My content coordinator is in another. I'm in London, usually in client work when the creative is actually being made. Which means I write briefs instead of running edit sessions. Yesterday I wrote one for a round of ad creative — three angles, the footage to pull from, what the on-screen copy might say for each, and the rule that ties it together: advance the message, don't repeat the opener. The whole document was about a hundred lines. It took ninety minutes to write. But now they can work for a week without needing a call. Every decision they'd otherwise interrupt me for is already answered. That's the leverage: not supervising execution, but front-loading the thinking so the execution doesn't need you. The quality of a brief is proportional to how much of your time the project eventually consumes. A weak brief means constant questions. A good brief means it runs without you.

Building out a paid social funnel for a brand this week. They had three campaigns running — cold audience, warm audience, retargeting — and each one was being measured on its own conversion rate. The cold campaign looked weak, because cold campaigns always look weak when you measure them on conversions. They're not supposed to convert. They're supposed to produce a warm pool that the retargeting layer works on. When you measure the wrong thing at each layer, you end up cutting the campaigns that are actually working. We rewired the measurement. Cold is scored on cost per warm engagement. The warm pool feeds retargeting. The email subscriber list syncs automatically so the paid journey and the email journey stay connected. Most paid social thinking treats each campaign as a standalone bet. The more useful frame is a pipeline: what is this layer qualifying for, and what does the handoff cost? Build the pipeline logic first. Then decide how much each layer needs to spend.

I was building a two-week itinerary for friends visiting from abroad — trips out of London, activities in the city, a cricket match at Edgbaston. When I fed the whole plan into Claude alongside my actual calendar, it flagged a clash I hadn't spotted. A client board meeting was already booked on a day I'd put my friends on a train. The cricket was non-refundable. The meeting date was fixed by someone else's schedule. There was a resolution, but it took some rearranging. The only reason I found it in time was because I'd given Claude the full picture — work and personal together — rather than asking about one part at a time. Most people use AI to answer questions about individual problems. The bigger return comes from using it to hold the whole picture at once: work, personal, other people's fixed dates. When everything is in one place, it finds the conflicts before you do.

A few weeks ago I completed on a flat I'd been buying for two years. The day after completion I was back in client meetings. What I hadn't fully registered: new landlords have 30 days from completion to protect the tenant's deposit in a government scheme. Miss it and you face fines and lose the right to serve notice — the scheme sends no reminders. So I gave Claude the full checklist: DPS account setup, proof of landlord status, deposit lodging, the Westminster council tax change. When the council's online form turned out to be a dead end — the guidance said one thing, the actual process was different — Claude found the email route and drafted the submission. The whole thing ran alongside a full client day. What would have been an afternoon of anxious tab-switching took about ninety minutes. Property has compliance deadlines just like a business does. Claude handles them well when you hand it the full picture upfront, not question by question.

Before redesigning a product page yesterday, I spent ninety minutes looking at product pages I had nothing to do with. Not competitors — pages I respected from adjacent categories. High-value products where the buying decision takes time and the page has to hold that process. I went through five of them systematically: where does the hero image sit, when does the first proof point appear, how is the price introduced, what does the call to action have to compete with. After ninety minutes I could describe our page's problems clearly. Before that study I had a vague sense something was off. After it I had a brief with specific changes. The study wasn't in the project plan. It took time I hadn't budgeted. But the redesign it produced was better than anything I would have built from instinct. Most redesigns start too early. The first hour doesn't belong on your page — it belongs on someone else's.

A brand we work with was boosting posts on Instagram. The budget was modest, the results were flat, and the instinct was to spend more. Before adding money, we ran an audit of everything they'd boosted in the last three months. Each post graded on three things: is the creative strong enough to stop a scroll, does the copy give a reason to click, and is the offer worth following through on. Most of them failed at least one. Several failed all three. The insight wasn't that the budget was too small. It was that they were amplifying posts that hadn't earned amplification yet. This matters because boosting feels like growth. You put money behind something and the numbers go up — reach, impressions, sometimes clicks. But if the underlying post isn't working, all you've done is show a weak message to more people. Run the audit before you run the budget. Grade your last ten boosted posts honestly. How many would you actually stop to read?

Most product and service pages describe what you built. They explain the feature, the methodology, the rationale — all of it written from the inside out. The buyer doesn't care about the inside. They want to know what they're getting, what changes, and why it matters to them specifically. Yesterday I rewrote a product page for a brand we work with. The original copy explained why the product was designed a certain way — useful for the team that built it, useless for the person deciding whether to buy it. The rewrite started with one question: what does a buyer want to know before they commit? Everything else followed from that. The features stayed, but the order changed, the language shifted, and the emphasis moved to outcomes instead of rationale. The gap between inside-out copy and outside-in copy is almost always the reason a good product page doesn't convert. Read your own pages the way a stranger would. If it answers questions you'd ask before buying, keep it. If it answers questions your team would ask, rewrite it.

We're planning a family trip that requires yellow fever vaccinations. Yellow fever shots can only be given at registered clinics, and there are apparently only a handful of them across London. The old approach: open eight browser tabs, cross-reference addresses, check each clinic's booking page, try to work out which one is nearest and actually has availability. Instead I asked Claude to sort it. It pulled the clinic list, cross-referenced it against Google Maps data for our postcode, and returned a ranked shortlist with distances and walk-in availability at each one. The whole thing took about twelve minutes. The time you save is the easy part to appreciate. The more useful insight is that most people still hand Claude single questions, when the bigger wins come from giving it an actual problem — one that would have meant five browser tabs and half an hour of your afternoon. Your AI system earns its keep when you stop asking it questions and start giving it jobs.

Most of the AI-generated content I read sounds like AI wrote it. The tell isn't the content itself — it's structural. One-word emphasis fragments, the "here's the thing" opener, the reversal for punch ("it wasn't the volume, it was the decision"): these patterns were trained in from the same wave of LinkedIn content, and now they're everywhere. Your readers recognise the pattern even when they can't name it. The fix isn't less AI — it's a blocklist. Before you ask the AI to draft anything, tell it exactly what constructions you want eliminated. Name the specific rhythms and phrases you want banned. The output stops reading like everyone else's output. Your voice isn't a style guide. It's an exclusion list.

Yesterday a system ran through four client websites without me. It scanned for broken links, rewrote underperforming page titles, fixed schema errors, and flagged a redirect chain that was leaking authority. All of it logged, all of it reversible, none of it waiting on a brief from me. I was writing a proposal. The site work happened in the background. This is what the leverage point actually looks like in practice — not AI writing your emails, but the ongoing unglamorous maintenance that keeps a business compounding, running on a schedule without you having to push it. The things that matter most for growth rarely feel urgent enough to prioritise. When they're systematised, they run whether you're at your desk or at school pickup.

Most productivity advice fails in one of two ways. It either asks for more discipline than a real life allows, or it hands you so many rules that running the system becomes the work. The 3-3-3 rule sits between the two. You take one working day and split it into three parts. Three hours on your most important deep work — the thing you'd want to point to at the end of the year, not the email that just landed. Three shorter tasks that have to happen today, one hour each. You get three slots, not six. Three maintenance items — the small things that keep a week functional, like the inbox, a tidy, or the ten minutes someone's been waiting for. So that's three hours of one thing, three quick tasks, and three bits of upkeep. It works because it's built around real limits, not aspiration. Three hours is about the ceiling for sustained deep work. Three live tasks is about what working memory holds before it turns into juggling. It also hands you a clean answer when someone drops a fourth thing on you: "I can do it tomorrow — my three for today are already set." It comes from Oliver Burkeman's Four Thousand Weeks. Pick tomorrow's three tonight, and protect those first three hours before anything else touches the day.

A 600-item inbox isn't 600 tasks. It's 600 decisions you've already put off. Every time you open it you re-read the same items and defer them again. That re-reading is the real cost — not the work itself, but the half-decisions you keep looping on. The 5 Ds kill the loop. Every task that lands gets one of five decisions within ninety seconds of you reading it. Do — under two minutes, handle it now. Defer — only you can do it, so put it on a real day, not "soon". Delegate — someone else can, so send it now with context and a deadline. Drop — it doesn't deserve doing by anyone, so delete it. Diary — not actionable yet, so park it in someday/maybe with a review date. There is no sixth option. Most people lean on Defer, because Drop feels like failure and Delegate feels like effort. So the inbox never clears. Audit your defer pile — how much of it should have been dropped or handed off on day one? Run it properly and the inbox empties in about twelve minutes a morning. Not because you're doing more, but because you're reading less.

Everyone has a CLAUDE.md file now. It tells the AI who you are, how you work, and the rules you care about. Mine carries my businesses, my writing voice, and how I want emails drafted — all genuinely useful. But on its own it just sits there. The AI reads it, understands you, and then waits. It knows how you work, but it doesn't know what to do today. So you still open the session and brief it from scratch. The fix was a second file next to it: a live task list I call actions.md. Every task is tagged three ways — who owns it (me, the AI, or both), roughly how much it matters, and what state it's in. Now the session starts differently. It reads the CLAUDE file for context and the actions list for the work, then carries on from where we stopped. It does the tasks tagged for it, and logs what it finished before it closes. Somewhere in there it stopped being an assistant I had to keep pointing at things. It started running the list like a project manager who already has it. The CLAUDE file is the context, and the actions list is the job. You need both of them. Set them up on one project and the next session opens already working, before you've said a word.

Before I let any automated process run live, I build in a shadow run first. It uses the same inputs and the same logic as the real thing. But instead of taking action, it writes what it would have done to a review file. You read that file, and if the decisions look right, you flip the switch. If they don't, you fix the spec first. Today I shadow-ran a marketing automation across four client accounts before going live. The shadow run caught two edge cases the spec had missed — things that would have been fine in testing but wrong in one specific client context. The go-live switch only got flipped once the shadow outputs looked clean. It adds one step to the process. In return, it removes the class of errors that are hardest to fix: the ones that ship confidently and then go wrong quietly. Any autonomous process worth trusting has to earn that trust first. Shadow mode is how you build it.

Before I let any automated process run live, I build in a shadow run first. Same inputs. Same logic. But instead of taking action, it writes what it WOULD have done to a review file. You read the review file. If the decisions look right, you flip the switch. If they don't, you fix the spec. Today I shadow-ran a marketing automation across four different client accounts before going live. The shadow run caught two edge cases the spec had missed — things that would have been fine in testing but wrong in one specific client context. The go-live switch only got flipped after the shadow outputs looked clean. It adds one step. It removes the class of errors that are hardest to fix: the ones that ship confidently and go wrong quietly. Any autonomous process worth trusting has to earn that trust first. Shadow mode is how you build it.

One of our product catalogues had a corruption problem. Unicode characters — the ones that show up as ??? in product names — scattered through 1,257 products. Years of imports, exports, encoding mismatches. The kind of technical debt nobody wants to touch. I pointed Claude at it. It catalogued every affected product, identified the corruption type for each one, generated the rollback data before making a single change, then applied the fixes in sequence. The part worth noting wasn't the volume. It was the rollback step first. Claude decided what needed capturing BEFORE it started — so the whole thing stayed reversible. That's not a find-and-replace. That's how you run data work you can trust. Unglamorous work. The kind of thing that sits on a list for months. Done in one session.

Everyone has a CLAUDE .md now. A file that tells the AI who you are, how you work, the rules you care about. Mine carries my businesses, my writing voice, how I want emails drafted. Useful. But on its own it just sits there. The AI reads it, understands you, and waits. It knows how you work. It doesn't know what to do today. So you still open the session and brief it from scratch. The fix was a second file next to it. A live task list. I call it actions .md. Every task tagged three ways: - who owns it — me, the AI, or both - roughly how much it matters - what state it's in Now the session starts differently. It reads the CLAUDE file for the context, the actions list for the work, and carries on from where we stopped. It does the tasks tagged for it. It logs what it finished before it closes. Somewhere in there it stopped being an assistant I had to keep pointing at things. It started running the list like a project manager who already has it. The CLAUDE file is the context. The actions list is the job. You need both. Set both up on one project and the next session opens already working, before you've said a word.

A 600-item inbox isn't 600 tasks. It's 600 decisions you've already put off. Every time you open it you re-read the same items and defer them again. That re-reading is the cost — not the work, the half-decisions you keep looping on. The 5 Ds kill the loop. Every task that lands gets one of five decisions within ninety seconds of you reading it. Do — under two minutes, handle it now. Defer — only you can do it, put it on a real day, not "soon". Delegate — someone else can, send it now with context and a deadline. Drop — it doesn't deserve doing by anyone, delete it. Diary — not actionable yet, park it in someday/maybe with a review date. No sixth option. Most people lean on Defer, because Drop feels like failure and Delegate feels like effort. So the inbox never clears. Audit your defer pile — how much of it should have been dropped or handed off on day one? Run it properly and the inbox empties in about twelve minutes a morning. Not because you're doing more. Because you're reading less.

Most productivity advice fails one of two ways. It asks for more discipline than a real life allows. Or it hands you so many rules that running the system becomes the work. The 3-3-3 rule sits between the two. You take one working day and split it into three parts. Three hours on your most important deep work. The thing you'd want to point to at the end of the year, not the email that just landed. Three shorter tasks that have to happen today. One hour each. You get three slots. Not six. Three. Three maintenance items. The small things that keep a week functional — inbox, tidy, the ten minutes someone's been waiting for. That's it. Three hours of one thing, three quick tasks, three bits of upkeep. It works because it's built around real limits, not aspiration. Three hours is about the ceiling for sustained deep work. Three live tasks is about what working memory holds before it turns into juggling. And it hands you a clean answer when someone drops a fourth thing on you: "I can do it tomorrow — my three for today are X, Y, Z." Comes from Oliver Burkeman's Four Thousand Weeks. Pick tomorrow's three tonight, and protect the first three hours before anything else touches the day.