{
  "version": "https://jsonfeed.org/version/1.1",
  "title": "Automancer — Field Notes",
  "home_page_url": "https://automancer.uk",
  "feed_url": "https://automancer.uk/field-notes/feed.json",
  "description": "Practical AI and workflow automation for UK small businesses. Cut admin, reduce errors, and improve operations with Automancer.",
  "language": "en-GB",
  "authors": [
    {
      "name": "Waseem Ilyas",
      "url": "https://automancer.uk/about"
    }
  ],
  "items": [
    {
      "id": "https://automancer.uk/field-notes/self-storage-software-what-to-check-before-you-sign",
      "url": "https://automancer.uk/field-notes/self-storage-software-what-to-check-before-you-sign",
      "title": "Self-storage software in the UK: what to check before you sign",
      "summary": "Before you renew a self-storage platform, check three things: how it charges you as you grow, what your customers can do online, and how you get your data out.",
      "content_text": "Self-storage is a good business partly because it is boring. Units, dates,\ndirect debits, the occasional padlock. The software you rent to run it is where\nthe interesting money quietly goes.\n\nThis is not a case study. We have not shipped a storage platform, and I am not\ngoing to pretend otherwise. It is the list I would work through if you asked me\nto look at yours.\n\n## What does a small self-storage operator actually need online?\n\nFive things. Live unit availability, so a customer can see what is free without\nringing you. Prices they can read without asking. A reservation that takes a\ndeposit or a first payment there and then. ID and agreement handling that\nproduces a signed, stored document without anybody finding a printer. And\nreporting you trust on a Monday morning: occupancy, arrears, who is due to\nmove out.\n\nEverything else on a vendor's feature list is a nice-to-have until those five\nwork properly. Dynamic pricing engines and marketing modules are very\npersuasive in a demo and worth nothing at all if your move-in still needs a\nphone call.\n\n## The monthly price is not the price\n\nStorage platforms tend to charge in one of three shapes: a flat monthly\nplatform fee, a monthly fee plus a percentage of the payments taken through\nthe system, or a per-unit fee that steps up in bands as you add units. Which\none you are on is in the contract, not usually on the pricing page.\n\nThe flat fee is predictable and the other two are a share of your growth. A\nper-transaction cut is the one that surprises people, because it grows exactly\nwhen the business is going well. Sell more, pay more, forever, for software\nthat costs the vendor no more to run at your higher volume than it did at your\nlower one.\n\nSo do not compare headline monthly fees. Rebuild the whole-year cost at the\noccupancy you expect to be at in two years, and again at the one you are\nfrightened of, then compare those. Vendors rarely make that easy to compute\nand they are not obliged to. It is not magic and it is not even hard: a\ncontract, a spreadsheet and half an hour.\n\n## Ask what happens when you want to leave\n\nThree questions, in writing, before you sign anything. The answers tell you\nmore than the demo does.\n\n- **Can I export my own data, whenever I want, in a format I can actually\n  use?** Usable means a spreadsheet or a data file with your customers, units,\n  agreements and payment history in it. A PDF report is not an export. If the\n  answer involves a support ticket, a fee, or a phrase like \"we can look at\n  that\", you now know what leaving costs.\n- **What happens to the price at renewal?** Ask for the uplift mechanism, in\n  the contract, not a reassurance on a call. A cap you can point to later is\n  worth having.\n- **What is the notice period, and what happens to live agreements and\n  in-flight payments if I serve it?** Your customers keep their units either\n  way. Somebody has to make that true, and you want to know who before it\n  matters.\n\nNone of this is hostile. It is the same thing any decent vendor has already\nthought about. But your customers will come to you when something goes wrong\nwith their data, not to the company whose logo is in the corner of the admin\nscreen, so the answers need to be yours to give.\n\n## Rent it, until renting stops making sense\n\nFor most independent operators with a site or two, renting is the right answer\nand I would not pretend otherwise. Somebody else handles the payment\nintegrations, the security patching and the phone-line when it breaks at 8pm,\nand that is worth real money.\n\nCustom software earns its place at a specific moment: when the percentage cut\nhas grown past what running your own would cost, or when the platform cannot\ndo the one thing your business actually depends on and the roadmap answer is\n\"not this year\". That is the calculation a UK plastics manufacturer went\nthrough before we built their [trade\nportal](/work/manufacturer-trade-portal). Account-specific pricing that used to\nlive in people's heads is now computed by a rules engine, with ordering,\nproduction and dispatch in one flow, Phase 1 signed off in April 2026.\nDifferent sector, same question: at what point does renting someone else's\napproximation cost more than owning the real thing?\n\nOurs are published, so you can do that sum without booking anything. An\nAutomation Opportunity Audit is from £450, and a System / Agent Build is from\n£4,500. They are on the [services page](/services) with everything else.\n\n## What I'd actually do about it\n\nFind your current contract and pull out three numbers: the monthly fee, the\npercentage cut if there is one, and the notice period. Most operators know the\nfirst, guess the second, and have never read the third.\n\nThen rebuild last year's total cost at next year's occupancy. If the answer\ngrows faster than your revenue does, you have found the thing worth\nnegotiating at renewal, and you have a year to do it in rather than a fortnight.\n\nWhile you are there, run the five-things list against your own website as a\ncustomer would. Availability, price, reserve, sign, report. If a prospective\ncustomer cannot get from \"have you got a 50 square foot unit\" to \"it's booked\"\nwithout speaking to a human, that is costing you more than any contract term\non this page.\n\nIf you want a second pair of eyes on it, that is what an Automation\nOpportunity Audit is for: from £450, credited in full against any project you\ngo on to do, and you keep the plan either way. Or just fill in [the\nform](/contact). Waseem reads it himself and emails you back, we promise a\nmeeting within one week, and a human makes every decision along the way.\n\n*We build software, we are not solicitors, and none of the above is legal\nadvice. It is how we read a software contract before recommending one.*\n",
      "date_published": "2026-08-24T00:00:00.000Z"
    },
    {
      "id": "https://automancer.uk/field-notes/new-business-website-legal-compliance-checklist",
      "url": "https://automancer.uk/field-notes/new-business-website-legal-compliance-checklist",
      "title": "New business, no website: the UK compliance checklist",
      "summary": "What a new UK company's website needs: company details in the footer, a real email address, an up-to-date privacy notice, and an honest cookie banner.",
      "content_text": "You can incorporate a UK company online, before lunch, without speaking to\nanybody. Working out what your website is then expected to carry takes rather\nlonger, which is why most new businesses skip it and hope.\n\nWe build sites for brand-new companies, so this is the list we work through\nevery time. It is not legal advice, and we are not the people to give you any.\nIt is what a working site needs in place before you start pointing traffic at\nit.\n\n## What does a new UK business website need to include?\n\nFour facts about the company: its registered name, the part of the UK where it\nis registered, its company number, and its registered office address. A real\nway to reach you: your name, a geographic address, and a working email address,\nnot only a form. A privacy notice, if the site collects any personal data at\nall, which a contact form does. And a cookie position that matches what your\nanalytics is actually configured to do. Worth knowing what is not on that list:\nnothing obliges a business website to publish terms and conditions. They are a\ngood idea, not a box to tick.\n\n## The footer nobody tells you about\n\nThe four company facts have to be available on the site. Nobody prescribes\nwhere, so a site-wide footer or a clearly signposted legal-information page\nboth do the job, as long as a human can actually read it without a magnifying\nglass.\n\nIt is tempting to file this under tidiness. It is not. Leaving it off is an\noffence in itself, and the sharper risk sits in the contracts you make while\nin breach: the other side can ask the court to throw out your claim on one. A\ncourt can still let the case through where the defendant was not actually\nprejudiced, so it is a risk rather than a certainty, but it is a risk you are\nrunning on your own unpaid invoices for no reason at all. No incantation\nrequired to avoid it: four facts, in a footer, ten minutes.\n\n## The requirement almost nobody has heard of: a real email address\n\nA commercial website is expected to publish a working email address, in a form\nthat is easy to find and stays put. A contact form on its own is generally not\ntreated as enough. The form is fine as the fast route, but a published address\nneeds to sit alongside it.\n\nThis is usually read to cover ordinary brochure sites too, not just online\nshops, which is why so many otherwise tidy small-business sites are quietly\noffside. It also has more teeth than it used to: since April 2025 the\nCompetition and Markets Authority can enforce these consumer-facing duties\ndirectly, after two decades in which more or less nobody did.\n\n## The data rules moved twice in 2026\n\nIf your site has an enquiry form, you are collecting personal data, and you\nneed a privacy notice that says who you are, what you are doing with the data,\nyour legal basis, who else sees it, how long you keep it, and what rights\npeople have. In plain language, where you collect the data, not buried three\nclicks away.\n\nOne line changed in June 2026. Your notice now has to tell people they can\ncomplain to you directly and not only to the Information Commissioner, and you\nhave to actually handle those complaints: make it easy to raise one,\nacknowledge it inside a month, and respond without dragging it out. If your\nnotice was written before this summer it is missing a line it is now supposed\nto carry. That is a five-minute edit, and it is the single most common thing we\nfind out of date on an existing site.\n\nCookies moved earlier in the year, so check the date on anything you read about\nthem, including this. You may have heard that analytics no longer needs\nconsent. It is narrower than that. The exemption covers analytics used only to\ncollect statistics about how your own site is used so you can improve it, where\nthe data is not passed on to anyone else, you tell visitors clearly what you\nare doing, and you give them a simple free way to say no. Analytics that feeds\nadvertising, measures conversions for an ad platform, or follows individual\nvisitors around still needs consent, so a default install with the ad features\nswitched on is not covered. The penalty ceiling for getting this wrong went up\nsharply in 2026, which is a good reason to know which of those two things your\ntag is doing.\n\n## From nothing to live and compliant, and why \"later\" is a bad plan\n\nWe took a brand-new UK travel agency from zero web presence to a live,\ncompliant marketing site: services sections, a tested working enquiry form,\nWhatsApp and phone contact wiring, and the required trading-disclosure footer.\nWe handled domain purchase, DNS, HTTPS hosting and interim email routing too,\nso they had a working front door and a working inbox from day one. That last\npart is not incidental. The email requirement above is impossible to satisfy if\nnobody has decided who reads the inbox yet. The write-up is in [the fast\nwebsites case study](/work/fast-small-business-websites), and it is hand-built,\nso it costs essentially nothing to run.\n\n\"We'll add the legal stuff later\" fails for a dull reason. Later is when you\nhave traffic. Disclosures, the privacy notice and cookie behaviour are cheapest\nwhile the site is being built, because they are decisions about structure.\nRetrofitting a consent mechanism onto a live site, or working out after the\nfact whose data you have been collecting since March, costs real money and\nproduces a worse result.\n\n## What I'd actually do about it\n\nOpen your own site and check three things. Are the four company facts in the\nfooter. Is there a real email address, not just a form. Does your privacy\nnotice mention the right to complain to you as well as to the ICO. If you\nincorporated recently, all three are probably missing, and all three are an\nafternoon's work rather than a project.\n\nLeave the cookie question until you know what your analytics is actually\nconfigured to do, because that determines the answer and nothing else does. And\nwhether you owe the ICO the annual data protection fee is worth ten minutes on\ntheir registration self-assessment rather than a guess in either direction.\n\nWe publish real from-prices for our automation work on our [services\npage](/services), starting with an Automation Opportunity Audit from £450, and\nwe would rather tell you one honest page is enough than sell you eight. If you\nwant a straight answer about your own site, fill in [the form](/contact).\nWaseem reads it himself and emails you back, we promise a meeting within one\nweek, and a human makes every decision along the way.\n\n*We build websites, we are not solicitors, and none of the above is legal\nadvice. It describes how we set up sites for new UK companies as at August\n2026. If your situation is unusual, or a decision here carries consequences you\ncare about, take proper advice on it.*\n",
      "date_published": "2026-08-17T00:00:00.000Z"
    },
    {
      "id": "https://automancer.uk/field-notes/credit-hire-website-compliance-trust-checklist",
      "url": "https://automancer.uk/field-notes/credit-hire-website-compliance-trust-checklist",
      "title": "Credit hire firms: the website compliance and trust checklist",
      "summary": "What a UK credit hire website needs: the registration details that belong in the footer, the trust signals that win the enquiry, and the SEO basics templates skip.",
      "content_text": "Someone has just been in a crash that wasn't their fault, and over the next hour\nthree different people will tell them to be careful who they trust. Then they\nfind your website.\n\n## What does a credit hire firm's website actually need?\n\nA credit hire website needs four things: a plain explanation of how a non-fault\nreplacement vehicle works and who pays for it, visible proof that you're a real\nregistered UK company, the registration details that belong in the footer, and\nstructured data so search engines know you're a local business serving a specific\narea. Design comes after all four, not before.\n\nThat list is short because the job is narrow. Nobody browses credit hire sites\nfor pleasure. Your visitor arrived from a search or a garage's recommendation,\nthey have one question, and they'll decide in about a minute whether you look\nlike a firm that answers the phone.\n\n## Trust does more work here than design does\n\nIn most trades a website's job is to look credible. In this one it has to defuse\nsuspicion that arrived before the visitor did. Your customer has probably already\nbeen offered a courtesy car by the other side's insurer, warned about hire\ncharges by somebody, and asked to sign an agreement on the worst day they've had\nall year. None of that is your fault, and all of it is yours to answer.\n\nOne thing I won't do here: tell you what your obligations are on the hire\nagreement itself. That's your solicitor's territory and it moves. This is about\nthe shop window.\n\nWhat the shop window has to carry:\n\n- **A named, registered company.** Registered name, company number, where it's\n  registered, registered office. Not a first name and a mobile number.\n- **Plain words about money.** Who pays, when, and what happens if the claim\n  isn't accepted. Vagueness on this point reads as a catch, because the reader has\n  already been told there's one.\n- **A phone number a human answers**, in the header, tappable, plus whichever of\n  WhatsApp or SMS your customers actually use. Nobody fills in a contact form\n  from the hard shoulder.\n- **A place.** People want to know roughly where the vehicle is coming from and\n  whether you cover their town.\n\n## The footer most small sites get wrong\n\nHere's the list we work through when we build a site for a UK limited company:\nthe company's registered name, its registered number, the part of the UK where\nit's registered, and the address of its registered office. It's a footer, not a\nproject, and leaving it off costs you credibility long before it costs you\nanything else. If your own site is missing any of it, that's worth a conversation\nwith your accountant or solicitor as well as with whoever built the site.\n\nThe footer matters twice as much in this sector. A customer who has just been\ntold to be careful about credit hire will go looking for exactly this sort of\ndetail, and an honest firm with nothing in its footer gives them no way to tell\nthe difference. We built that footer into a new travel agency's site when we took\nthem from zero web presence to live, because a brand-new company is precisely the\nsort of business nobody has ever thought to mention it to. It's the cheapest\ncredibility you'll ever buy.\n\n## Getting found: the boring layer that decides it\n\nThe part of your site that determines whether you turn up for \"accident\nreplacement vehicle near me\" is invisible on the page. It's structured data, a\nsitemap and a robots file, and templates routinely skip all three.\n\nStructured data is the closest thing web development has to an incantation: a\nblock of text no visitor ever sees, which tells a search engine, and increasingly\nan AI assistant answering the question instead of a search engine, precisely what\nyour business is, where it operates and how to contact you. It's not magic. It's\na small, well-formed block of machine-readable text, and its power comes entirely\nfrom the fact that most sites can't be bothered to include it.\n\nFor a local credit-hire firm we built a sharp single-page marketing site with the\ntechnical fundamentals baked in: structured data so search engines understand the\nbusiness, plus the SEO basics, sitemap, robots and the rest, done properly rather\nthan skipped. Live, fast and findable. It's hand-built and costs essentially\nnothing to run, with no monthly platform fee quietly draining the account. The\nwrite-up is in [the fast websites case study](/work/fast-small-business-websites).\n\nFast and cheap didn't mean amateur. It meant we left out the parts that weren't\nthe job, rather than the parts that were.\n\n## The checklist: run this against your own site today\n\n1. **Trading disclosures in the footer of every page.** Registered name, company\n   number, place of registration, registered office address. Open your own site\n   and look. Most people find nothing.\n2. **One clear sentence explaining what credit hire is** and who pays, near the\n   top, before any form.\n3. **What happens if the claim isn't accepted**, in your own words, rather than\n   left for the customer to imagine.\n4. **A tappable phone number in the header**, plus the messaging channel your\n   customers actually use.\n5. **Your town and service area written in the page text**, not only inside a map\n   embed. If the area exists nowhere in words, \"near me\" searches can't match it.\n6. **Structured data describing you as a local business**: name, address, phone,\n   hours, area served. Ask whoever built the site to show you it's there.\n7. **A sitemap and a robots file.** Two files. Check that\n   yoursite.co.uk/sitemap.xml actually loads.\n8. **Test your own enquiry route** from a phone you don't normally use, and time\n   how long the reply takes. That number is your real conversion rate.\n\n## What I'd actually do about it\n\nIf you're missing items 1, 6 and 7, you don't need a new website. You need an\nafternoon of someone competent, and the work is cheap because it's small and well\nunderstood.\n\nIf you're missing 2 and 3, the fix is writing rather than code, and it's the\nhighest-value writing in the business. Answer the money question before the\ncustomer has to ask it, because being asked to trust someone who won't answer it\nis exactly what they were told to watch for.\n\nAnd if you have no site at all, don't start by shopping for a design. Write down\nthe three things a visitor must be able to do, which here is usually understand\nwhat credit hire costs them, believe you're real, and reach you in one tap. Get\nthat live, then improve it with real traffic in front of you.\n\nWe publish real from-prices for our automation work on our [services\npage](/services), starting with an Automation Opportunity Audit from £450, and\nwe'd rather tell you a single page is enough than sell you eight of them. If you\nwant a straight answer about your own site, fill in [the form](/contact). Waseem\nreads it himself and emails you back, we promise a meeting within one week, and a\nhuman makes every decision along the way.\n\n*We build software. We are not solicitors, and none of this is legal advice. For\nthat, ask someone qualified to give it.*\n",
      "date_published": "2026-08-10T00:00:00.000Z"
    },
    {
      "id": "https://automancer.uk/field-notes/small-business-website-cost-2026",
      "url": "https://automancer.uk/field-notes/small-business-website-cost-2026",
      "title": "How much should a small business website cost in 2026?",
      "summary": "A small business website should be a one-off cost, not a monthly rent, and not a five-figure project. What you are really paying for, and what to check.",
      "content_text": "Ask five web agencies what a small business site costs and you will get five\nnumbers, four of which only arrive after a discovery call. Here is the honest\nversion, including the part of the bill most quotes quietly leave out.\n\n## How much should a small business website cost in 2026?\n\nA marketing site for a small UK business should be a one-off cost rather than a\nmonthly rent, and it should not be a five-figure project. The number you are\nquoted for the build is only half the question. The other half is what the site\ncosts you every month for the next three years, and that second figure usually\ndecides whether you got a good deal.\n\nA site built once and hosted for essentially nothing will cost you less by year\ntwo than a cheaper one with a subscription bolted to it, and the gap widens\nevery month you keep trading. So the useful question is not \"what does a\nwebsite cost\". It is \"what does this website cost me to keep\".\n\n## The two bad options most small businesses are offered\n\nThe first is a web agency with a five-figure quote and a three-month timeline.\nFor a business selling complex products online with stock, payments and a\nwarehouse behind it, that can be entirely fair. For a new company that needs a\ncredible front door and a working enquiry form, it is a lot of money and a lot\nof waiting for something that should exist by the end of the week.\n\nThe second is a drag-and-drop builder that charges you rent forever and still\nlooks like everyone else's. It gets you live quickly, which is real value. But\nyou are renting the shop window, the lights stay on only while you keep paying,\nand the platform owns the thing your customers find you through.\n\nNeither is a scam. They are just both priced for someone who is not you.\n\n## What you are actually paying for\n\nMost of the cost difference between a good small site and a bad one is invisible\non the homepage. Three things are worth paying for, and templates routinely miss\nall three.\n\n**The footer.** Four details belong in it: registered name, company number, place\nof registration and registered office address. It is the list we work through on\nevery limited company site, and a new business is exactly the sort of business\nnobody has ever mentioned it to. It is a footer. It takes minutes to do properly\nand it is embarrassing to be missing. Check any quote includes it.\n\n**The technical fundamentals.** Structured data so search engines and, these\ndays, AI assistants can tell what your business actually is and where it\noperates. A sitemap and a robots file, done properly rather than skipped. None\nof this is glamorous and none of it shows up in a design mock-up, which is\nprecisely why it gets dropped.\n\n**A front door that works.** An enquiry form that has been tested by an actual\nhuman sending an actual enquiry. Contact routing that reaches you on the channel\nyour customers use. A domain, DNS and HTTPS that someone else set up so you\nnever have to learn what a nameserver is.\n\n## A real example: from nothing to live, in days\n\nWe took a new UK travel agency from zero web presence to a live marketing site:\nservices sections, a tested working enquiry form, WhatsApp and phone contact\nwiring, and the trading-disclosure footer. We handled the domain purchase, DNS,\nHTTPS hosting and interim email routing too, so they had a working front door\nand a working inbox from day one.\n\nFor a local credit-hire firm we built a sharp single-page site with structured\ndata so search engines understand the business, plus the SEO basics done\nproperly rather than skipped.\n\nBoth are hand-built and cost essentially nothing to run. No monthly platform fee\nquietly draining the account. The full write-up is in [the fast websites case\nstudy](/work/fast-small-business-websites).\n\nThere is no sleight of hand in that. Hand-built static pages have no platform\ntax to pay, so the running cost collapses to a domain renewal once a year. It is\nnot magic, it is just not paying rent on something you could own.\n\n## Five questions to ask before you accept a website quote\n\n1. **What does this cost me in year two?** Add up the build fee plus 24 months\n   of every recurring line on the quote. Compare that total, not the headline.\n2. **Who owns the site if I leave?** If the answer involves exporting to a zip\n   file that does not run anywhere else, you are renting.\n3. **Are the trading disclosures included?** If the person quoting you does not\n   know what you mean, that is your answer about the rest of the detail.\n4. **Is structured data, a sitemap and a robots file in scope?** These are cheap\n   to include and expensive to retrofit into a template.\n5. **When does it go live?** Days and weeks are reasonable for a marketing site.\n   Three months usually means you are queued behind bigger clients.\n\n## What I would actually do about it\n\nIf you are a new business with no site at all, do not start by shopping for a\nprice. Start by writing down the three things a visitor must be able to do:\nusually understand what you sell, trust you enough to enquire, and reach you.\nAnything that does not serve one of those three is scope you are paying for and\ndo not need yet.\n\nThen get it live and improve it with real traffic in front of you. A modest site\nthat exists today beats a better one that launches in November, because the\nbetter one is still a guess until someone visits it.\n\nWe publish real from-prices for our automation work on our [services\npage](/services), and we would rather tell you a site is a small job than sell\nyou a big one. If you want a straight answer about your own site, fill in [the\nform](/contact). Waseem reads it himself and emails you back, we promise a\nmeeting within one week, and a human makes every decision along the way.\n\n*We build software. We are not solicitors, and none of this is legal advice. For\nthat, ask someone qualified to give it.*\n",
      "date_published": "2026-08-03T00:00:00.000Z"
    },
    {
      "id": "https://automancer.uk/field-notes/b2b-trade-portals-what-to-build-first",
      "url": "https://automancer.uk/field-notes/b2b-trade-portals-what-to-build-first",
      "title": "B2B trade portals: what to build first, and what to leave for phase 2",
      "summary": "Phase 1 of a B2B portal should be the smallest thing your team can use on a Monday and trust. What to build first, what to defer, where to draw the line.",
      "content_text": "Every custom software project hits a moment where the wish-list meets the\ncalendar. How you handle that moment decides whether you get a working system\nin a few weeks or a half-built one next year.\n\n## Why big-bang launches go wrong for small teams\n\nA big-bang launch is the one everybody pictures. Months of building, one\ngo-live date, the old way switched off on a Friday and the new way switched on\nby Monday.\n\nLarge organisations can just about absorb that. They run both systems side by\nside for a quarter, they have someone whose actual job is training, and if it\ngoes badly there is a fallback and a budget line for it. A team of twelve has\nnone of those things. Nobody has a spare afternoon to test a system that is not\nlive yet, so the testing happens on real orders in front of real customers.\nThere is no parallel run, because the same six people would have to do\neverything twice. And the review arrives in one enormous lump: you are asked to\napprove a system you have never used, on the day you start depending on it.\n\nThe quieter failure is what happens during the silent months in between.\nNothing to react to means every requirement stays a guess, and guesses drift.\nBy the time it lands, half of what you asked for in February is not what you\nneed in September, and nobody found out in time to say so.\n\n## What should go in phase 1 of a B2B portal build?\n\nPhase 1 should be the smallest version of the system your team could use on a\nMonday morning and trust with real work. Not a demo and not a prototype: a\nusable foundation that runs your main flow end to end, from the thing coming in\nto the thing going out.\n\nThat means a thin slice through the whole process rather than a thick layer of\none part of it. A polished customer login with no order queue behind it is a\nnice screen. An order that can be placed, priced, made, dispatched and turned\ninto a draft invoice is a business running on software, even with half the\nreporting still missing.\n\nThe test is blunt. Can one of your people take a real job all the way through\nwithout leaving the system to finish it off in a spreadsheet? If yes, that is a\nfoundation, and everything after it is an improvement on something that already\nworks. If no, it is not phase 1, it is a fragment, and you will be running two\nsystems until phase 2 lands.\n\n## What we deferred on a real build, and why it wasn't a corner cut\n\nFor a UK plastics manufacturer and trade supplier, we built a trade portal\ncovering the order lifecycle: account-specific pricing, catalogue, ordering,\nproduction records, dispatch records and invoice drafts, with per-customer\npricing computed by a rules engine rather than remembered. Phase 1 was signed\noff in April 2026 as a usable foundation, running as a hosted preview behind\nsecure access, staged for client review.\n\nDeeper accounting integration and full invoicing were deliberately scoped into\nphase 2.\n\nWe drew that line where the system stops depending on someone else's system,\nrather than where the work got hard. Order to dispatch sits entirely inside the\nclient's own four walls, so we control the rules, the data and the edge cases.\nPosting finished invoices into an accounting package means living with a third\nparty's API, its authentication model and its opinions about what a valid\ninvoice looks like. Build that first and you are integrating against a process\nyou have not finished designing. Build it second and you are integrating a\nshape you already know is right.\n\nDeferring is not delivering less. It is being wrong earlier and more cheaply,\nwhich is most of what good delivery actually is. The full write-up is in [the\ntrade portal case study](/work/manufacturer-trade-portal).\n\n## Four questions to phase your own project\n\n1. **Which single flow, if it worked properly, would change your week?** That\n   is the spine of phase 1. Everything else is scenery until the spine holds\n   weight.\n2. **Which parts depend on someone else's system?** Accounting packages,\n   couriers, payment providers, anything with an API you do not control. Push\n   them right, unless the integration is the entire point of the project.\n3. **What can you load with real data on day one?** That trade portal was\n   seeded from the client's own price-list workbook, so it stood up loaded with\n   around 74 customers, 888 products and 1,263 price rules rather than three\n   tidy fake ones. Fake data hides exactly the problems real data finds.\n4. **What are you keeping just in case?** Every brief has items nobody can\n   describe an actual Monday morning for. Those are not phase 2. They are phase\n   never, and saying so out loud is the cheapest decision you will make all\n   project.\n\nAnswer those four and you have a scope. Answer none of them and you have a\nwish-list, which is a different document that costs a lot more.\n\n## What to do with this on your own project\n\nWrite down your phase 1 in one sentence, in the form \"someone in the office can\ndo X from start to finish without leaving the system\". If you cannot finish\nthat sentence, the scope is not ready, and no amount of quoting will make it\nready. If you can, you have something a developer can price honestly and you\ncan check when it arrives.\n\nThere is no sleight of hand in any of this. Phasing well is just refusing to\nguess for six months at a time, then testing what you built against real data\nbefore anyone's Monday depends on it.\n\nIf you would rather work through it with someone who ships this sort of thing,\nthat is what an [Automation Opportunity Audit](/services) is for: we map how\nyour business actually runs and hand you a ranked, costed plan for what to fix\nfirst, from £450, with the fee credited against a Sprint (from £1,950) or a\nBuild (from £4,500) if you go on to do one. Fill in [the form](/contact) and\nWaseem reads it himself, then emails you back. We promise a meeting within one\nweek, and a human makes every decision along the way.\n",
      "date_published": "2026-07-28T00:00:00.000Z"
    },
    {
      "id": "https://automancer.uk/field-notes/manufacturing-order-processing-phone-and-memory-pricing",
      "url": "https://automancer.uk/field-notes/manufacturing-order-processing-phone-and-memory-pricing",
      "title": "Manufacturing order processing: getting off phone-and-memory pricing",
      "summary": "Trade pricing kept in someone's head is a single point of failure. What a B2B trade portal for a UK manufacturer actually needs, and in what order.",
      "content_text": "Ring most UK trade suppliers with an account and you get a person, not a\nsystem. They know who you are, they know your pricing, and the order is on\nits way before you have finished the sentence. That is genuinely good\nservice. It is also one head cold away from being no service at all.\n\n## What actually goes wrong when the pricing lives in someone's head\n\nPer-customer trade pricing is a relationship asset. A builder gets a better\nrate because they have bought every month for years, and someone in the\noffice knows that. Nothing about the arrangement is written down anywhere a\ncomputer could check it.\n\nThe failure modes are boring and predictable, which is exactly why they get\ntolerated for years. The person who remembers is off, so an order sits until\nThursday or goes out at the wrong number. The account list grows past what\none memory can hold, and quotes start drifting. A price gets given wrong and\neats the margin on that job quietly, because nobody reconciles quoted price\nagainst list price afterwards. Then that person retires, and a good chunk of\nyour commercial policy retires with them.\n\nNone of that is an argument for stripping the relationship out of the sale.\nIt is an argument for writing the relationship down, so the system knows what\nthe person knows and the person is freed up to do the part only a human can.\n\n## What does a B2B trade portal for a manufacturer actually need?\n\nA trade portal earns its keep when order, production, dispatch and invoicing\nrun as one flow rather than five disconnected tools. In practice that means:\naccount-specific pricing computed per customer, a catalogue those customers\ncan browse themselves, ordering and quick reordering, a staff-side order\nqueue, production records, dispatch records, and a draft invoice that falls\nout of the end of the process instead of being typed up from scratch.\n\nThe word carrying the weight there is \"one\". Most manufacturers already have\nevery item on that list. They just have them in six places: an inbox, a\nwhiteboard, a price-list spreadsheet, an accounting package, a delivery book\nand someone's memory. Every join between those six is a re-key, and a re-key\nis where the number goes wrong.\n\n## The multi-unit problem generic e-commerce doesn't solve\n\nIf you sell building products, you do not sell in units. You sell in feet,\nmetres, square feet, square metres and each, and the same product moves\nbetween them depending on who is buying and what they are doing with it.\n\nOff-the-shelf e-commerce platforms are built for a world where a product has\na price. Yours has a price per unit type, per customer, which is a different\nshape entirely. Force it into a shopping-cart platform and you end up\nmaintaining a spreadsheet alongside the system that was supposed to replace\nthe spreadsheet.\n\nThere is a second thing generic platforms rarely handle well: what happens\nwhen the sheet price moves. If a supplier puts up your raw material cost, you\nneed to reprice in bulk and have every account's specific arrangement follow\nalong, not hand-edit hundreds of rules and hope. Add cross-border orders into\nIreland and the gap between \"an online shop\" and \"a trade system\" is fairly\nobvious.\n\n## What we built for one manufacturer\n\nFor a UK plastics manufacturer and trade supplier, we replaced informal\norder-taking with a full trade portal covering the whole order lifecycle.\nCustomer side: a public brochure site, secure passwordless login for trade\ncustomers, and a portal showing their own account-specific pricing, the\ncatalogue, ordering and quick reorder. Staff side: a workspace for managing\ncustomers, products and price rules, an order queue, production records,\ndispatch records and invoice drafts.\n\nAt the centre sits a multi-unit pricing matrix covering feet, metres, square\nfeet, square metres or each, with bulk reprice, plus handling for cross-border\norders into Ireland. The demo was seeded from the client's own real price-list\nworkbook, so it stood up loaded with around 74 customers, 888 products and\n1,263 price rules on day one rather than three tidy fake ones.\n\nPhase 1 was signed off in April 2026 as a usable foundation: order to dispatch\nworking end to end, running as a hosted preview behind secure access, staged\nfor client review. Deeper accounting integration and full invoicing were\ndeliberately scoped into Phase 2, so the client got something real early and\nwe built the rest on proven ground.\n\nPer-customer pricing is now computed rather than remembered. That is the whole\ntrick, and it is not much of a trick: it is a rules engine, written down\ncarefully, tested end to end. Not magic. Just very good engineering. You can\nread the full write-up in [the trade portal case\nstudy](/work/manufacturer-trade-portal).\n\n## Where to start if this sounds like your business\n\nYou do not have to commission a portal to make progress this quarter. Start\nwith three questions, in this order.\n\n1. **Where does your pricing actually live?** If the honest answer is \"in\n   Dave's head and a spreadsheet only he opens\", that is your single point of\n   failure, and it is a bigger commercial risk than any software decision you\n   are weighing up.\n2. **How many times does one order get typed in?** Count the re-keys between\n   the enquiry landing and the invoice going out. Count them honestly,\n   including the ones that feel too small to count. Each is a place a digit\n   can slip.\n3. **What is the smallest useful first phase?** Not the finished system. The\n   smallest thing your team could genuinely use on a Monday morning and\n   trust. If nobody can name it, the scope is not ready yet.\n\nThose three answers will tell you more than any vendor demo will. If you would\nrather work through them with someone who builds this sort of thing for a\nliving, an [Automation Opportunity Audit](/services) is exactly that: we map\nhow your business actually runs and hand you a ranked, costed plan for what to\nfix first, from £450, with the fee credited against any project you go on to\ndo. No funnel, no discovery-call theatre. Just a straight look at where your\norders are leaking time.\n",
      "date_published": "2026-07-20T00:00:00.000Z"
    },
    {
      "id": "https://automancer.uk/field-notes/five-signs-spreadsheet-problem",
      "url": "https://automancer.uk/field-notes/five-signs-spreadsheet-problem",
      "title": "Five signs your business has a spreadsheet problem, not a software problem",
      "summary": "Five signs a UK small business has outgrown its spreadsheets: disagreeing numbers, a single point of failure, weekly re-keying and a 'do not touch' tab.",
      "content_text": "Most businesses do not decide to run on spreadsheets. They arrive there one\ntab at a time, because a spreadsheet is the fastest way to solve today's\nproblem, and today's problem always wins. That is fine, right up until the\nspreadsheet stops being a tool you use and becomes a thing you maintain.\nHere is how to tell which side of that line you are on.\n\n## How do I know if my business is too reliant on spreadsheets?\n\nYou have a spreadsheet problem, not a software problem, when the spreadsheet\nhas stopped recording how the business runs and started dictating it. The\ntell is friction: work now bends around the file's limits instead of the\nfile serving the work. The five signs below are the specific shapes that\nfriction takes. If two or more sound like a normal Tuesday in your business,\nthe spreadsheet is no longer saving you time. It is quietly charging rent.\n\n## Sign 1: the same number lives in three places, and they disagree\n\nAsk three people in your business for the same figure, your current\nheadcount, this month's revenue, how many orders are open, and see if you\nget the same answer. In a lot of small businesses you do not, because that\nnumber lives in a rota, an accounting package and a spreadsheet, and nobody\nappointed one of them as the truth.\n\nThis is the reconciliation problem, and it is the most common hidden cost in\nsmall-business operations. Nobody notices until the numbers are wrong in a\nmeeting, in front of someone who matters. The fix is not a better\nspreadsheet. It is deciding which system is the source of truth and making\nthe others agree with it on purpose, not by hope.\n\n## Sign 2: one person is the only one who knows how it works\n\nThere is a spreadsheet, and there is a person who understands the\nspreadsheet, and the business runs on the overlap. When that person is on\nholiday, a normal task becomes a research project. When they leave, a\nworking process leaves with them.\n\nThat is not a criticism of the person. They built something genuinely\nuseful. But a business should not have a single point of failure sitting in\none head and one file, because that is the one part of your operation you\ncannot audit, back up, or hand over. Software that anyone on the team can\nuse turns private knowledge into a shared process. That is most of the value\nright there, before you automate a single thing.\n\n## Sign 3: there is a tab called \"DO NOT TOUCH\"\n\nEvery long-lived spreadsheet grows a tab like this. A hidden sheet full of\nlookup tables and formulas that took an afternoon to get right and would\ntake a week to rebuild. A red cell nobody is allowed near. A macro someone\nwrote in 2021 that still runs and nobody remembers why.\n\nA \"DO NOT TOUCH\" tab is a confession. It means the spreadsheet is now doing\na job that needs the discipline of actual software, versioning, testing, a\nrecord of what changed and who changed it, but has none of the safety rails\nthat come with it. The work has outgrown the tool. The warning label is the\nspreadsheet telling you so.\n\n## Sign 4: you re-key the same data into a second system every week\n\nIf a person on your team spends part of every week copying data out of one\nplace and typing it into another, that is not a workflow. It is a job for a\nmachine that has been given to a human. It is slow, it is boring, and it is\nwhere errors breed, because a person re-keying the same fields for the\nhundredth time is exactly when a digit slips.\n\nThe cost is easy to miss because it is spread thin. Ten minutes here, twenty\nthere, never big enough on any single day to fix. Add it up across a year\nand it is often a meaningful slice of someone's salary spent moving data\nthat never needed a human to move it. This is usually the first thing worth\nautomating, not the biggest system, the most repeated re-key.\n\n## Sign 5: you cannot get an accurate number, right now, for something that matters\n\nSomeone asks how the business is doing today. Not last quarter, once the\naccounts are done. Today. And the honest answer is \"give me a couple of days\nto pull it together\". Revenue, occupancy, hours logged, stock on hand, the\nnumber that tells you whether this week is going well. If you cannot see it\nwithout a manual assembly job, you are flying on last month's instruments.\n\nA business should be able to answer its own most important question on any\nordinary day, not just at month-end. When it cannot, that is rarely because\nthe data does not exist. It is because the data is scattered across files\nthat were never built to give you a live view, only a snapshot someone has\nto reassemble by hand each time.\n\n## What to actually do about it\n\nNotice that none of these five signs is \"the spreadsheet is bad\". Spreadsheets\nare brilliant. They are the right tool for a genuinely surprising amount of\nwork, and swapping one out for software you do not need is its own expensive\nmistake. The signs above are not \"you failed at spreadsheets\". They are the\nspreadsheet honestly telling you it has been promoted past its job.\n\nSo do not start by buying software. Start by auditing:\n\n1. **Map how the business actually runs**, tools, handoffs and re-keys\n   included, not how the org chart says it runs.\n2. **For each step, ask who does it, how often, and what it costs** in time or\n   money when it goes wrong.\n3. **Rank by frequency and pain, not by how interesting it is to fix.** The\n   dull, daily, error-prone re-key beats the exciting rebuild almost every\n   time.\n\nThat order matters, because the point is never to have the most software. It\nis to stop paying, in time and mistakes, for work a machine should be doing.\nSometimes the answer really is a better spreadsheet. Sometimes it is one\nsmall automation. Occasionally it is a proper custom system. You cannot know\nwhich until you have looked, honestly, at where the friction actually is.\n\nIf you would rather not do that first pass alone, an [Automation Opportunity\nAudit](/services) is exactly that working session done rigorously: we map how\nyour business runs and hand you a ranked, costed plan for what to fix first,\nwith the fee credited against any project you go on to do. You can also read\n[how we work](/about), agents do the boring work, a human makes every\ndecision and takes every call, before you decide whether any of it is for\nyou. No funnel, no countdown, just a look at where your week is actually\ngoing.\n",
      "date_published": "2026-07-13T00:00:00.000Z"
    },
    {
      "id": "https://automancer.uk/field-notes/cqc-compliance-evidence-stop-scrambling",
      "url": "https://automancer.uk/field-notes/cqc-compliance-evidence-stop-scrambling",
      "title": "CQC compliance evidence: how to stop scrambling before an inspection",
      "summary": "How UK care providers stop scrambling before a CQC inspection: turn scattered training records and QA signals into one source of truth.",
      "content_text": "Every care provider knows the feeling. An inspection is coming, and\nsomeone loses three days pulling training certificates out of four\nsystems, chasing refresher dates, and rebuilding a picture the business\nshould have been able to show at any moment. That scramble is not a sign\nyour care is poor. It is a sign your evidence lives in too many places.\n\n## The real problem is where your evidence lives, not whether you have it\n\nIf you are scrambling before a CQC inspection, the fix is not more effort\nthe week before. It is a single source of truth you can point an inspector\nat on any ordinary Tuesday. Most providers already do the caring well.\nWhat they struggle to do is produce the evidence quickly, because it is\nspread across an outsourced training provider, a rota spreadsheet, an\nincident log, and one manager's memory of who is overdue on their\nmoving-and-handling refresher.\n\nThat is not a failing on your part. It is what happens when a sector this\nheavily regulated gets served by tools that were each built to do one\nslice of the job and none of them built to talk to the others.\n\n## What a single source of truth for training actually looks like\n\nTo prepare training records for a CQC inspection without the last-minute\npanic, you need one system that tracks three states for every member of\nstaff automatically: what is mandatory, what is due for a refresher, and\nwhat is overdue. Not a spreadsheet someone re-reads and colour-codes by\nhand the week before, but a live record that already knows the answer.\n\nMost off-the-shelf learning platforms do not do this natively. They will\nhost courses and mark completions, then leave the compliance question, the\n\"who is out of date, right now, and on what\" question, back on your plate.\n\nWe built exactly this layer on a real project for a ~600-person\nsupported-living care provider. We replaced outsourced training and manual,\nspreadsheet-based refresher-tracking with a self-hosted learning management\nsystem, plus a custom compliance layer to track mandatory, refresher and\noverdue training the off-the-shelf tool could not handle on its own. It was\nseeded across all three of the provider's branches with courses loaded.\nThe point of that layer is simple: it turns scattered inspection evidence\ninto one place you can actually point an inspector at.\n\n## What a QA oversight dashboard surfaces that a manual review can't\n\nA quality dashboard earns its keep between inspections, not just before\nthem. Done properly, it stops being a report you assemble and becomes a\nview that is already assembled, showing you the exceptions before a\nregulator does.\n\nOn the same build, the QA and management oversight dashboard is live. It\npulls care-quality signals from multiple systems into one view: KPI cards,\nbranch comparisons, a support-worker priority watchlist, a repeat-concern\nwatchlist for service users, medication review, and a filterable review\nqueue. Instead of hunting through separate systems for the thing that\nneeds attention, a manager opens one screen and sees it. That is the\ndifference between evidence you have to go and find and evidence that\ncomes to you.\n\n## The honest bit: \"staged for go-live\" is not the same as \"live\"\n\nHere is the part most vendors skip. Not all of this happens overnight, and\npretending it does would be its own kind of dishonesty.\n\nOn that provider's build, the QA dashboard is genuinely live and in daily\nuse. The learning management system and its compliance layer are seeded\nacross all three branches and staged for go-live, which is a specific\nthing, not a softer word for \"done\". It means built, courses loaded, and\ntested on staging, waiting on the client's sign-off before it touches real\nstaff records. Care records are not a place to move fast for its own sake.\nEvery lane of this work was scoped, built on staging, reconciled against\nthe client's real exports, and board-gated before it went anywhere near\nlive data.\n\nThere is no filing cabinet that quietly organises itself the night before\nan inspection. There is just one system, kept current on purpose, so the\nanswer is already there when someone asks.\n\n## What I'd actually do about this\n\nIf inspection prep is a recurring fire drill in your business, work in this\norder:\n\n1. **Put training compliance in one place first.** Mandatory, refresher and\n   overdue, tracked automatically. This is the evidence an inspector asks\n   for most predictably, so it is the highest-value thing to stop tracking\n   by hand.\n2. **Add the oversight view second.** Once the underlying records are\n   trustworthy, a dashboard that surfaces exceptions is mostly plumbing,\n   not a separate project.\n3. **Insist on honest delivery status.** \"Live\", \"staged for go-live\" and\n   \"in build\" mean different things. A supplier who blurs them before an\n   inspection is not one you want near your care records.\n\nIf you cannot tell where your own evidence gaps are, that is what an\n[Automation Opportunity Audit](/services) is for: a working session that\nmaps how your business actually runs and gives you a ranked, costed plan\nfor what to fix first. You can see the fuller care build behind this post\non our [care provider transformation](/work/care-provider-transformation)\nwrite-up. No obligation to do any of it with us, and the audit fee comes\noff if you do.\n",
      "date_published": "2026-07-06T00:00:00.000Z"
    },
    {
      "id": "https://automancer.uk/field-notes/automation-for-care-providers-where-to-start",
      "url": "https://automancer.uk/field-notes/automation-for-care-providers-where-to-start",
      "title": "Automation for care providers: where to start",
      "summary": "A plain-English order of operations for automating a UK supported-living or domiciliary care business, based on real delivered work: canonical records first, then scheduling, then QA.",
      "content_text": "Supported-living and domiciliary care runs on paperwork most other sectors\ndon't have to think about: rosters, supervisions, medication records,\ntraining compliance, incident logs, mileage claims, timesheets. If you're\nrunning a CQC-regulated care business on spreadsheets and a patchwork of\noutsourced platforms that don't talk to each other, you're not doing it\nwrong. You're doing what almost every provider your size does. The\nquestion isn't whether to automate. It's what to automate first, and in\nwhat order, without putting real care data at risk while you do it.\n\n## Why care providers can't just buy off-the-shelf software\n\nMost business software assumes clean, stable data: one customer record,\none address, one relationship. Care providers don't have that luxury. A\nservice user has a care plan, a funding body, a next of kin, incident\nhistory and a medication schedule that all need to line up. A support\nworker has qualifications, DBS checks, supervision dates and shift\navailability that all need to line up too. Generic CRM or rostering tools\nhandle a slice of that and leave the rest in spreadsheets, which is how\nmost providers end up with the same person's details spelled three\ndifferent ways across four systems.\n\nThat's not a software failure. It's what happens when a sector with this\nmuch regulatory complexity gets served by tools built for simpler\nbusinesses.\n\n## Start with canonical records, not the flashy dashboard\n\nIf you're planning where to automate first, resist the pull toward the\nmost visible thing (a nice-looking dashboard) and start with the least\nvisible thing: one clean, canonical record per person. Every downstream\nautomation, from scheduling to compliance tracking to QA reporting, is\nonly as trustworthy as the record it's built on.\n\nWe proved this out on a real build: a single roster import for a\n~600-person supported-living provider turned messy client CSVs into\n**294 carers and 293 service users** as clean, canonical, auditable\nrecords. Identity matching (comparison logic that decides \"same person\"\nacross systems) **auto-resolved 274 of 418** review candidates, so a human\nonly had to look at the ones that were genuinely ambiguous. Get this layer\nright and everything you build afterward gets easier. Skip it and you're\nautomating on top of duplicate data, which just makes the mess move\nfaster.\n\n## Then scheduling, because it compounds\n\nOnce you have canonical records, scheduling is usually the next highest-\nvalue target, because it's the thing that repeats constantly and fails\nquietly. Supervisions and spot checks are a good example: they're a\nregulatory requirement, they happen on a cycle, and tracking that cycle by\nhand across a growing workforce is exactly the kind of job a machine does\nbetter than a person with a full diary. On the same build, **588\nsupervision and spot-check schedules were created automatically** once the\ncanonical records existed underneath them.\n\n## QA and compliance evidence come last, and they come easier than you'd think\n\nOnce records and scheduling are automated, a QA or compliance-evidence\nlayer stops being a separate project and becomes mostly plumbing: pulling\nsignals that already exist into one place a manager or an inspector can\nactually look at, instead of hunting through separate systems for\nexceptions. Do this step first, before the record and scheduling layers\nare solid, and you're just building a nicer window onto the same mess.\n\n## A three-question checklist before you automate anything in a regulated care business\n\n1. **Is this a canonical record, or a report built on top of one?**\n   Automate records first. Reports second.\n2. **What happens to this data if the automation gets it wrong?** If the\n   answer touches medication, safeguarding or a regulator's evidence\n   trail, it needs to be staged, tested against real exports, and signed\n   off by a human before it goes anywhere near live service-user data.\n   Never move fast for its own sake here.\n3. **Can you point to a verified backup and a tested restore, not just a\n   backup policy on paper?** Infrastructure discipline matters more in\n   care than almost anywhere else. On our own build, the documented\n   production infrastructure audit came back with zero failures and zero\n   warnings across every layer, including a genuinely restored backup, not\n   a hopeful one.\n\n## It's not magic. It's the right order of operations\n\nNone of this is a secret trick. It's the same discipline you'd want from\nany engineer working near care records: get the foundation right, stage\neverything, reconcile against real data, and let a human sign off before\nanything touches production. That's how we approached the whole\ntransformation for [a ~600-person supported-living care provider](/work/care-provider-transformation),\ndelivered as roughly eighteen governed workstreams rather than one risky\nbig-bang rebuild.\n\nIf you're looking at your own care business and can't tell where to start,\nthat's exactly what an [Automation Opportunity Audit](/services) is for:\na working session that maps how your business actually runs, and a ranked,\ncosted plan for what to automate first. No obligation to do any of it with\nus, and the fee comes off if you do.\n",
      "date_published": "2026-07-04T00:00:00.000Z"
    },
    {
      "id": "https://automancer.uk/field-notes/what-does-an-ai-agent-actually-cost",
      "url": "https://automancer.uk/field-notes/what-does-an-ai-agent-actually-cost",
      "title": "What does an AI agent actually cost? A 2026 UK price breakdown",
      "summary": "Real published UK prices for AI agent and automation projects: audits from £450, not the five- or six-figure guess most vendors make you wait for.",
      "content_text": "Someone pitches you an \"AI agent\" for your business. You ask how much it\ncosts. You get \"it depends\", a discovery call, and a number that only\nappears after you've sat through a sales deck. Here's the honest version\nof \"it depends\", with real prices attached, not a funnel.\n\n## The short answer\n\nFor a UK small business, a genuinely useful automation or AI agent project\ntypically starts in the low thousands, not the tens of thousands. The\nrange depends entirely on scope: a single automated workflow can be live\nin around two weeks from roughly £1,950. A proper custom system or AI\nagent built to last starts from £4,500. If you don't yet know what's worth\nautomating, an Automation Opportunity Audit starts the whole process for\n£450, and that fee comes straight off if you go on to do the work with us.\nThose are our own published prices, not industry averages, but they're a\nreal, current UK data point for what this actually costs when a vendor\ntells you the truth up front.\n\n## Why \"it depends\" isn't a dodge, but it's also not an excuse\n\nThe honest reason cost varies is that \"AI agent\" covers wildly different\namounts of engineering. A script that reads an inbox and drafts a reply\nfor a human to approve is a different job from a system that ingests\nmessy client data, matches identities across records, and writes back into\nthree other platforms. Vendors who say \"it depends\" and then refuse to\ngive you *any* number until you've sat through a call aren't protecting\nyou from a bad estimate. They're protecting their pricing from comparison.\n\n## What actually moves the price\n\n- **How much human review stays in the loop.** More judgement calls kept\n  with a person costs more to build (proper approval flows, audit trails)\n  but usually costs less in risk. We build this in by default, not as an\n  upsell: a human makes every decision and takes every call on anything\n  that matters.\n- **How many systems it has to talk to.** One clean workflow between two\n  tools is a Sprint-sized job. A system that has to reconcile data across\n  four platforms and keep them in agreement is a Build.\n- **Data quality going in.** Clean source data is cheap to automate on top\n  of. Messy, duplicated, inconsistent data (which is most small\n  businesses, honestly) needs a records layer built first, which is real\n  work, not padding.\n- **What \"properly built\" actually includes.** Audited, monitored, backed\n  up and tested infrastructure costs more up front than a demo that only\n  has to survive one meeting. It's also the difference between something\n  that's still running in six months and something that quietly breaks the\n  first busy Monday.\n\n## What that looks like against a real number\n\nWe built a self-service AI product for a diversity & inclusion consultancy\nthat replaced a **£300–400 consultant engagement** with a **£10–80\nself-service task, completed in minutes instead of hours**. That's not a\ncost estimate for a generic AI agent. It's the real economics of one\n[shipped product](/work/debiaser-ai-product), included here because it's a\nuseful sanity check: a well-built AI product doesn't just move a task from\na person to a machine, it can fundamentally change what the task costs the\ncustomer, because the delivery model changed, not just the labour.\n\n## A quick way to sanity-check any AI agent quote you're given\n\nAsk three questions before you sign anything:\n\n1. **Can you show me this running in production somewhere, today, not in\n   a slide deck?** If the answer is no, you're paying for a prototype, not\n   an agent.\n2. **What happens when it gets something wrong, and who signs off before\n   it acts?** If there's no human in the loop for anything that touches\n   money, customers, or compliance, that's a red flag, not a feature.\n3. **Is the quote a real number, or a range that only becomes real after a\n   sales call?** A vendor who can't give you a ballpark before a meeting\n   usually has a reason not to.\n\n## Our own answer, published, not pitched\n\n- **Automation Opportunity Audit:** from £450. Fee credited against any\n  project you go on to do.\n- **Workflow Sprint:** from £1,950. 1–3 automations live in around two\n  weeks.\n- **System / Agent Build:** from £4,500. A proper custom system or AI\n  agent, built to last.\n- **AI Ops Partner:** from £495/month, cancel anytime. We keep it running\n  and keep automating the next thing.\n\nFull detail on what's included at each level is on our [services and\npricing page](/services). If you're not sure which one fits, start with\nthe Audit. That's exactly what it's designed to answer.\n",
      "date_published": "2026-07-04T00:00:00.000Z"
    }
  ]
}