# Lurto

> AI automation agency for US and UK businesses. Custom AI solutions in production: AI agents, document generation, voice receptionists, CRM and ERP integrations.

## About
Lurto is an AI automation agency operated by Meridal Group LLC (Sheridan, Wyoming, United States). It builds and repairs custom AI systems that survive production: AI agents and LLM workflows, document generation from a client's own records, custom voice agents including a production voice AI receptionist for the US market, RAG knowledge assistants, CRM and ERP integrations, the n8n/Make/Zapier workflows around them, and eval and reliability harnesses. Every engagement is a written scope with a price agreed before any work, and clients own all accounts and code from day one.

Prices are published in US dollars, pounds sterling and euros. Sterling and euro figures are fixed list prices, not live conversions.

## Services and prices
- AI Rescue (/services/rescue): from $750 / from £600 / from €700, 1-2 weeks for fixes, 2-4 for rebuilds. diagnostic from $500, credited in full. Your AI build broke, stalled, or never worked right. We find the root cause and fix it: dead n8n workflows, hallucinating agents, voice bots that drop calls.
- Free consultation (/contact?intent=consult): Free / Free / Free, 30 minutes, proposal in 2 business days. no obligation, no paid discovery. Not sure what to automate yet? Spend 30 minutes telling us how the work runs today. Within two business days you get a written proposal with a fixed price, or a plain answer that automation will not pay back for you.
- Build Sprint (/services/build): $3,000-$25,000 / £2,300-£19,500 / €2,800-€23,000, 2-6 weeks. price agreed in writing before we start · 50% upfront, 50% on delivery. One named outcome, shipped to production on a written scope, price and date. You own everything we build.
- Care & Scale (/services/care): $1,000-$3,000/mo / £800-£2,300/mo / €900-€2,800/mo, ongoing. monthly rolling, 30 days notice. AI systems rot without maintenance: APIs change, models drift, edge cases appear. We monitor, fix, improve, and show you the numbers every month.
- Fractional AI Engineer (/services/fractional): $6,000-$12,000/mo / £4,700-£9,400/mo / €5,500-€11,000/mo, ongoing. 8-20 hours per week · 3-month minimum. A senior AI engineer embedded in your team for 8 to 20 hours a week. Strategy and shipping from the same person.

Build sprint menu:
- Single workflow automation: $3,000-$5,000 / £2,300-£3,900 / €2,800-€4,600, 2 weeks. One system talking to another with AI in the middle: intake, enrichment, routing, notifications.
- Multi-system automation: $6,000-$12,000 / £4,700-£9,400 / €5,500-€11,000, 3-4 weeks. CRM + finance + comms working as one pipeline. The work your team keeps doing by hand.
- Custom AI voice agent: $6,000-$20,000 / £4,700-£15,600 / €5,500-€18,400, 3-4 weeks. Inbound answering that books real appointments and writes back to your CRM. Built on the voice stack we run in production.
- RAG knowledge assistant: $5,000-$10,000 / £3,900-£7,800 / €4,600-€9,200, 2-3 weeks. An internal assistant grounded in your documents, with citations and access control.
- Multi-agent platform build: $15,000-$25,000 / £11,700-£19,500 / €13,800-€23,000, 4-6 weeks. Several agents and systems working as one operation: orchestration, human approval gates, monitoring, and a runbook your team can actually operate.
- Eval & reliability harness: $5,000-$15,000 / £3,900-£11,700 / €4,600-€13,800, 2-4 weeks. Regression tests, monitoring, and a failure taxonomy for an AI system you already run. Proof it works, release after release.

Rescue bands:
- Written diagnostic: from $500 / from £400 / from €450, 2 business days. We read the workflows, logs, and prompts. Written root-cause report and a fixed repair quote. Credited in full against the fix, and refunded if it cannot name a fixable cause.
- Small fix: $750-$1,500 / £600-£1,200 / €700-€1,400, 2-5 days. One clear failure: a broken integration, a bad prompt, a missing retry. When the diagnostic shows a minor break, you pay the minor price.
- Standard rescue: $1,500-$4,000 / £1,200-£3,100 / €1,400-€3,700, 1-2 weeks. Where most rescues land. Several failure points or a fragile architecture: root-cause fixes, error handling, alerts, and a trail you can audit.
- Rebuild: $4,000-$10,000 / £3,100-£7,800 / €3,700-€9,200, 2-4 weeks. Only when repairing costs more than starting over. The diagnostic says so in writing, with both numbers, before you decide.

Care tiers:
- Care: $1,000/mo / £800/mo / €900/mo. Up to 2 systems monitored. Fixes within 48h. One small improvement per month. Monthly report.
- Care+: $1,800/mo / £1,400/mo / €1,700/mo. Up to 5 systems. Fixes within 24h. One workflow-sized improvement per month. Monthly report plus a written strategy review.
- Scale: $3,000/mo / £2,300/mo / €2,800/mo. Whole-stack ownership. Priority same-day response. Continuous improvement backlog worked every month. Monthly written strategy review plus quarterly roadmap.

Third-party running costs are paid by the client directly to the vendor at cost, with no markup. They are quoted in dollars because the vendors bill in dollars:
- Automation platform (n8n cloud / Make / Zapier): $20-$80/mo
- AI model usage (OpenAI / Anthropic APIs): $10-$150/mo
- Voice minutes & phone numbers (if voice): $0.10-$0.20/min
- Hosting (if self-hosted n8n / custom app): $10-$50/mo

## Guarantees
- G/01 30-day bug warranty: Anything we shipped breaks within 30 days, we fix it free, and it is never billed as a change request.
- G/02 Reply within one business day: Every email answered within one business day. Care clients: 24-48h fix SLA.
- G/03 Scope before invoice: Every offer has a written scope and a price you see before any work starts. No 'book a demo to find out'.
- G/04 You own everything: Accounts, code, workflows, docs: all in your name from day one. Fire us anytime and keep it all.
- G/05 Diagnostic or free: If the written diagnostic cannot name a fixable root cause, it is refunded. A report that says nothing is not worth paying for.

## Frequently asked, with the answers as published
Q: What does an AI automation agency actually do?
A: It takes the work your people do by hand and puts it into production as something that keeps running without them. For us that is four jobs: rescuing a build that broke, building one named outcome to a written scope, running what ships, and standing in as the senior engineer when you need one. Working out where automation would pay, and where it would not, comes first and costs nothing: a free consultation ending in a written proposal. The model is usually one step inside a workflow rather than the whole system: it reads, sorts, drafts or answers, and the workflow decides what happens next.

Q: Why publish pricing when most agencies do not?
A: Because hidden pricing is a sales tactic. You should be able to know what this costs before spending an hour on a call. Putting the number on the page also keeps us honest about scope: the price is the price.

Q: What does the free consultation cost, and what is the catch?
A: Nothing, and there is no catch. It is a 30-minute call where you explain how the work runs today. Within two business days you get a written proposal with a fixed price and a date, or a plain note that automation would not pay back for you. There is no paid discovery phase and no obligation to go ahead.

Q: Someone already built us an automation and it keeps breaking. Can you fix it instead of rebuilding?
A: Usually yes: that is the Rescue service. The written diagnostic, from $500, tells you the root cause and a written price to repair it. We only recommend a rebuild when repairing genuinely costs more, and the diagnostic will say so in writing.

Q: Do we own what you build?
A: Yes, entirely. Everything runs on accounts in your name, code is handed over with documentation, and third-party costs are paid by you directly at cost, with no markup. If we stop working together, everything keeps working.

Q: Where is the agency based?
A: The company behind the site is Meridal Group LLC, registered in Sheridan, Wyoming. US clients get a W-9 and a domestic invoice in dollars, engineering coverage across US business hours, and a fixed-scope statement of work that fits a standard vendor onboarding.

Q: Can you work white-label for my agency?
A: Yes. We are used to building behind the scenes, and your client does not need to know we exist.

Q: Are these 'AI agents' or workflows?
A: Both words appear on this site, and they mean different things. A workflow is a fixed sequence with AI at specific steps; an agent decides which step comes next. Most business jobs need the first, because it is deterministic and auditable. We use an agent only where the job genuinely needs one, as in the document and stock systems on the work page, and we name the outcome either way.

Q: What happens after launch?
A: 30 days of included support, then an optional Care plan from $1,000/mo: monitoring, fixes within 24-48h, one improvement per month, and a monthly report with the numbers. Monthly rolling with 30 days notice.

## Work: 5 systems in production today
### Regulatory document drafting agent
Client: Hydrogeology engineering firm (Engineering)
Problem: Regulatory well-project documents took an engineer hours of copy-paste drafting per project, from templates plus site data.
Built: A production document agent behind an authenticated web app: engineers submit project data, the agent drafts the full regulatory document set in the firm's own format, with background job processing and status polling.
Result: The regulatory document set for a well project now drafts itself from the project data | Drafting that took an engineer hours now takes minutes of review | Client came back and commissioned a second system
What was hard: The documents had to come out in the firm's own template, not a lookalike, so the checks compare every draft against real past submissions. Drafting takes long enough that it cannot happen while somebody waits on a page, so it runs in the background and the page reports where it has got to.

### Warehouse stock intelligence agent
Client: Same engineering firm, second contract (Operations)
Problem: Stock answers lived in an ERP nobody wanted to query. Finding what is in the warehouse meant asking the one person who knew.
Built: A second production agent for the same client: natural-language answers over live warehouse data, with its own access-controlled portal alongside the document system.
Result: Second contract from the same client, scoped in a fraction of the time | Both systems run side by side on shared infrastructure | Deployed with one-command deploy scripts and monitoring
What was hard: Stock lives in an accounting system with no clean API, so the agent answers from a mirrored, timestamped copy and always says how old the numbers are. A confident wrong quantity is worse than no answer.

### Google Ads operations agent
Client: Digital marketing agency (Marketing)
Problem: Campaign management work was eating senior time: repetitive checks, reporting, and client-by-client housekeeping across accounts.
Built: An agent that automates the agency's Google Ads operations workflows, with per-client metadata in Supabase and containerized deployment on the agency's VPS.
Result: Routine campaign operations run without a human driving them | Second phase commissioned after the first shipped | Runs alongside other automation services on shared infra
What was hard: Every client account is different, so the rules live as per-client metadata rather than in the code, and the agency's own people edit them without a deploy.

### AI receptionist for US small businesses
Client: Voice AI product, US market (Voice AI)
Problem: Calls that arrive after hours or mid-job go to voicemail, and most callers never leave one. The brief was a receptionist that answers, books the appointment and writes it to the CRM, built to survive real telephony rather than a demo.
Built: A receptionist that answers around the clock, replies to website chat, makes outbound follow-ups, books into the calendar and writes the caller and the outcome into whatever system the business already runs on.
Result: Production voice infrastructure we now build client agents on | Voice eval harness catches regressions before customers do | Answering live calls every day, on real telephony
What was hard: A voice agent's mistakes leave with the caller instead of sitting on a screen, so the dangerous failure is a confident sentence with nothing behind it. Finding those meant building something that calls the system and listens, rather than reading logs and hoping.

### Multi-phase AI content production pipeline
Client: International publishing client (Publishing)
Problem: A multi-stage book production process needing orchestration across drafting, review, and delivery workflows. Too complex for one prompt, too manual as-is.
Built: A pipeline split into separate stages, each one able to run, fail and be retried on its own, driven from a chat window the client already had open all day.
Result: Three phases delivered and extended across follow-on work | Client operates the pipeline day to day without us | Survived platform updates thanks to documented update cycle
What was hard: The tools it is built on keep upgrading underneath it, so every stage has a written update routine and a way back, and the client runs both without calling us.

Clients are described by industry until each agrees in writing to be named.

## Fix library: symptom pages, each with a real failure trace and an importable workflow
### My automation stopped working: how to find what changed
/fix/automation-stopped-working (n8n, updated 2026-08-26)
When an automation stops working and nobody touched it, the change still has a date on it: the date just belongs to somebody else. Open the earliest failed execution rather than the newest and read three fields: the HTTP status code, which node last executed, and the timestamp. The gap between the last green run and the first red one is usually about an hour wide, and it names the cause: a rotated key, a token expired by a security policy, a renamed CRM field, a retired API version, or volume that has started hitting a rate limit.

### Workflow runs but nothing happens: find the silent failure
/fix/workflow-runs-but-nothing-happens (n8n, updated 2026-08-26)
A green execution means only that no node threw an exception. It says nothing about whether any work was done. Open the run and read the small item counter above each node: the point where a node takes items in and emits none is where your data disappeared, and every node after it shows as not executed rather than failed. Four shapes cover almost every case: items filtered to zero after a field rename, an error returned inside a 200 response, Continue On Fail turning an error into ordinary data, and work landing in a sandbox account nobody is looking at.

### OAuth stopped working after weeks: token expiry, traced and fixed
/fix/credentials-expired (Any platform, updated 2026-08-26)
An OAuth connection that worked for weeks and now returns 401 has almost never broken inside your workflow. A refresh token reached the end of its life or was revoked somewhere else, a password change, an offboarding, an admin tidying third-party apps, or a security policy firing on a risk signal. The credential record on your side still looks connected, because what died is at the provider's end. Reopening the credential and clicking through the consent screen again is usually the entire fix, and no amount of reading the workflow will show you anything.

### AI agent not calling tools: read the n8n trace before you fix it
/fix/ai-agent-not-calling-tools (n8n, updated 2026-08-26)
An agent that ignores its tools is almost always a plumbing problem visible in the trace within ten minutes, not a stupid model. Read the per-run log for three things: how many tool calls happened, the arguments the model sent, and the raw value that came back. Two shapes cover most of it: the tool never ran because its description was empty or generic, so the model could not tell it was relevant; or the tool ran, returned the right data, and the result was trimmed out of context before the next turn. Stop editing the system prompt and go read what the model was handed.

### Voice agent books the wrong time: timezones, DST and double bookings
/fix/voice-agent-books-wrong-times (Any platform, updated 2026-09-01)
Four distinct faults produce a voice agent booking the wrong time, and they need different fixes: a UTC offset stored instead of the timezone, so every future booking made before a clock change lands an hour out; the hour that does not exist on spring-forward and the hour that happens twice on fall-back, both of which most date libraries accept without a word; check-then-book, where availability was read half a minute before the write went in; and a confirmation composed from what the agent intended rather than from what the booking system actually returned.

### Chatbot goes off script: leaked prompts, invented discounts, and how to stop it
/fix/chatbot-goes-off-script (Any platform, updated 2026-09-01)
A chatbot goes off script through four routes, and adding a line to the prompt closes only the exact sentence in the screenshot: the instructions come back out, because a refusal written against the word prompt is routed around by asking for a translation or a summary; an instruction arrives inside retrieved data; the model commits the company to a price or policy nothing retrieved; or it simply agrees with an angry customer. Three of the four are fixed architecturally rather than textually, anything you would not want published does not belong in the prompt, and anything that changes state belongs behind a tool your code approves.

### Make scenario burning operations: where they go and how to stop paying for them
/fix/make-scenario-operation-limits (Make, updated 2026-09-01)
Make charges an operation per module run, so a scenario that loops fifty times over an empty result set has failed at nothing and spent fifty operations. Five patterns account for almost all runaway consumption: a polling trigger that finds nothing (every minute is 43,200 operations a month), an iterator that charges per module per record, a retry loop against an endpoint that moved, a scenario nobody turned off, and a data store used as a cache. Find them by sorting scenarios by operations consumed and asking what business outcome each of the top ten produced this month.

### RAG assistant answering from old documents: stale indexes, orphans and drift
/fix/rag-assistant-answers-from-old-documents (Any platform, updated 2026-09-01)
An assistant quoting an old policy has no error state: retrieval succeeded, the answer was grounded, the citation is real, and every dashboard is green. Four faults produce it and they present identically, the document changed and nothing re-indexed it, the document was withdrawn at source and the vectors stayed, chunks from the old version survive alongside the new so retrieval returns both, or the index is fine and the right chunk simply sits below the cutoff. Diagnose them in that order, because chunks from withdrawn documents poison every retrieval experiment you run.

## AI automation agency price index (/tools/agency-price-index)
Every figure below was read from the agency's own public website on 2026-08-22, with no call booked. Of the 20 agencies measured besides us: 4 credit the entry fee against the build, 7 name repair of an existing system as an offer, and 2 removed their public prices during 2026. Median flat-fee entry point across the 12 that price one: $2,000.

### Lurto (US), this is us
URL: https://lurtoagency.com/pricing
Prices public: full
Published: Free consultation: a 30-minute call, then a written fixed-price proposal within two business days. Written diagnostic from $500, credited in full and refunded if it names no fixable cause. Rescue $750 to $10,000. Build Sprints $3,000 to $25,000 fixed. Care & Scale $1,000 / $1,800 / $3,000 a month. Fractional AI Engineer $6,000 to $12,000 a month.

### The AI Consultancy (UK)
URL: https://theaiconsultancy.ai/pricing
Prices public: full
Published: Diagnostic audit from £495. Readiness Sprint from £3,500, 2 weeks, 50% credited within 90 days. Discovery and Pilot from £15,000, typically £15k to £35k. Build and Embed £40k to £90k for SMEs, £90k to £250k enterprise. Day rate £950 to £1,500. Fractional CAIO £3,000 / £5,000 / £8,500 a month. Chatbots £3k, £8k, £25k. Voice £5k to £12k. Production Clinic rescue: £350 to £750 quick fix, £1,250 to £3,500 deployment rescue, £8k to £25k refactor. Support £950 to £3,500 a month. Payment 30/40/30.

### Startrise AI Labs (US)
URL: https://startrise.io
Prices public: full
Published: Around 25 fixed SKUs. AI Assistant from $900. Voice bots $2,400. Workflow automation $2,500. Vibe code cleanup $2,500. Monitoring and guardrails $2,500. AI project rescue $3,500. AI agents $3,500. Security audit and red-teaming $4,500. Control centers $5,000. MVP $7,500. Compliance readiness $7,500. SaaS build $15,000. Automatic SEO $299/mo. Care plans from $2,500/mo. Human-agent teams $6,000/mo.

### CloudNSite (US)
URL: https://cloudnsite.com
Prices public: full
Published: Current State Assessment $999, credited against builds over $12,000 within 30 days. Defined automation build from $8,000. Managed care from $1,500/mo. AI readiness and governance sprint $7,500. Fractional AI office $6,000 to $8,000/mo. On-site activation $30,000 plus travel. Hourly $150.

### The Automation Agency (UK)
URL: https://automation-agency.co.uk
Prices public: full
Published: Simple automation £350. Single AI workflow or agent £750. Chatbot £750. Internal tool or dashboard £1,500. Process Audit £1,500 fixed, credited in full within 60 days if a build follows. Automation build from £3,000 over 2 to 6 weeks. Multi-agent from £8,000. Retainer £1,500/mo rolling.

### Palavir (US)
URL: https://palavir.co
Prices public: partial
Published: Custom market report from $1,500. Consulting discovery $2,000. Audit $2,500 single workflow, $5,000 multi-workflow. Build $25,000 fixed in 10 working days. Consulting retainer $7,500 to $10,000/mo. Data products $99 to $499/mo on a separate page.

### Zaps Studio (Global)
URL: https://zapsstudio.com
Prices public: full
Published: Free 30-minute audit. Workflow Rescue Sprint from $1,000 over 2 to 4 weeks. Revenue Operations System from $3,000 over 4 to 6 weeks. Automation Partner from $500/mo. Prices switch to sterling for UK visitors at roughly 0.8.

### LeftClick (US)
URL: https://leftclick.ai/pricing
Prices public: partial
Published: Free funnel audit. Engagements start at $5,000. Most projects $10,000 to $50,000 fixed, no hourly billing. Retainers offered but not priced publicly.

### Justin McKelvey (US)
URL: https://justinmckelvey.com
Prices public: full
Published: Free 20-minute repo audit. AI Readiness Assessment $2,500 flat over 2 weeks, credited in full within 90 days. Claude install from $4,500 plus optional $1,500/mo retainer. Vibe Code Rescue $25,000 to $50,000. $15,000/mo and $2,000/hour also listed.

### Aegis Designs (US)
URL: https://aegisdesigns.io
Prices public: partial
Published: Build plus subscription tiers. Anchor $1,500/mo plus a $5,000 to $7,500 build. Growth $3,200/mo plus $7,500 to $15,000. Scale $6,500/mo plus $15,000 and up. All month to month. A second page lists $150/hour and $750/mo, which does not reconcile with the first.

### MQLFlow (UK)
URL: https://mqlflow.com
Prices public: full
Published: Automation strategy from £3,200. Automation set-up from £800. AI agents from £4,000. Live dashboards from £2,400. Personalised outreach from £1,200. Retained services from £200/mo. Day rate £800.

### XRAY (US)
URL: https://xray.tech
Prices public: full
Published: $250/hour on demand. Hour blocks $1,000 for 4h, $2,500 for 10h, $5,000 for 20h. Design Sprint $15,000. Full service $15,000/mo. Zapier Solution Partner.

### 2V Automation (US)
URL: https://2vautomation.ai
Prices public: removed
Published: Public pricing removed during 2026. Previously $5,000 to $20,000 per project over 3 to 6 weeks, retainers from $1,000/mo, annual engagements described as mid five to low six figures.

### AY Automate (US)
URL: https://ayautomate.com
Prices public: removed
Published: Public pricing removed during 2026. Previously discovery from $4,500, builds $8,000 to $35,000, retainers $500 to $3,000/mo, fractional at $6,500 / $9,500 / $15,000 a month, first month fully refundable.

### Buldrr (Global)
URL: https://buldrr.com
Prices public: full
Published: Starter $400 to $1,200. Growth $1,500 to $4,500. Scale retainer $1,200 to $8,000/mo with queued requests.

### Podlevskikh Automation (Global)
URL: https://auto.podlevskikh.com
Prices public: full
Published: Free diagnosis within 24 hours. Rescue $250 to $800 with 48 to 72 hour turnaround. New builds $500 to $3,500. Retainer from $300/mo. Solo operator.

### Axivon (UK)
URL: https://axivon-ai.com
Prices public: full
Published: Revenue Leakage Review £495, fully credited. Business Brain Programme £1,995 over 90 days. First implementation £4,000 to £12,000. Support £300 to £900/mo. Day rate £550.

### Devaland (Global)
URL: https://devaland.com
Prices public: full
Published: Paid pilot from $2,500. Fixed builds $8,000 to $25,000. Fractional from $4,000/mo. Day rate from $750. Publishes a VAT number, an EU business register entry and a NATO NCAGE code.

### Aloa (US)
URL: https://aloa.co/ai/pricing
Prices public: full
Published: Flexible consultancy from $3,000. Proof of concept $20,000 to $30,000 over 6 to 8 weeks. Production ready $50,000 to $150,000 over 3 to 4 months. Enterprise $150,000 to $300,000 and up. Publishes a blended rate comparison: $150 to $200/hour against Big Four at $400+.

### Winder.AI (UK)
URL: https://winder.ai
Prices public: full
Published: £150 to £400/hour. Strategy sprints £10k to £50k. Proof of concept £15k to £40k. Pilot or MVP £40k to £120k. Production mid-market £120k to £400k. Enterprise and regulated £400k to £1.5m. First-year operations 15 to 25% of build cost.

### Blue Orange Digital (US)
URL: https://blueorange.digital
Prices public: full
Published: Assess $25,000 to $50,000 over 2 to 4 weeks. Deploy $150,000 to $500,000 over 3 to 6 months. Scale $500,000 and up. Advisor, described as fractional AI leadership, $15,000 to $30,000/mo.

## Writing
### We put an AI receptionist on our own phone line first. Here is what broke.
/blog/voice-receptionist-on-our-own-line (Case study, 2026-09-08)
We built a voice receptionist and pointed our own business number at it before selling one to anyone, so that every failure landed on us rather than a client. It answers at any hour, books into a real calendar, writes the caller into the CRM and leaves a summary in the morning. Two things it taught us: in week two it double-booked a slot because availability was checked ninety seconds before the write, and our eval harness once graded the agent on a transcript it never heard, so a change that scored better made the calls worse.

### A document agent for a drilling firm: 168 regulatory documents and a second contract
/blog/regulatory-document-agent-case-study (Case study, 2026-09-07)
A hydrogeology engineering firm produced its regulatory document sets by opening the last similar project and changing the site data, hours per project, done by a qualified engineer. The system now drafts the full set in the firm's own templates in minutes, and the engineer reads and corrects it. As of August 2026 it has produced 168 documents and holds 349 boreholes on file. The hard part was never the model: it was making the file open in Word looking exactly like theirs.

### The Monday spreadsheet: ad reporting for a skincare brand that writes itself
/blog/meta-ads-reporting-case-study (Case study, 2026-09-05)
Every Monday somebody copied numbers out of Meta Ads Manager and a skincare brand's store analytics into a spreadsheet, for about an hour. A nightly n8n pipeline with a Postgres store now pulls both, reconciles spend against revenue onto the same row, and sends a one-page summary with the slipped ad sets marked. It removed about four hours a month, but the change that mattered was that a bad ad set gets noticed the day it slips instead of at month end. There is no model in the nightly path at all.

### The work somebody in your company does by hand every week, and which to automate first
/blog/work-done-by-hand-every-week (Starting out, 2026-09-04)
A job worth automating has four marks: it happens the same way every time, it moves information between two systems that do not talk, nobody ever decided it should work this way, and it is done by someone hired for something else. Three out of four makes it a candidate; all four usually makes it the first one. Then do the arithmetic on paper: times per week, minutes each, what an hour of that person's time costs, and ask the one question the arithmetic misses: what happens when the job is done wrong?

### What an AI automation actually is, explained without the vocabulary
/blog/what-an-ai-automation-is (Starting out, 2026-09-02)
Most of what is sold as AI automation is a workflow: a fixed sequence of steps a computer runs, with one model step in the middle that reads, sorts, drafts, extracts or answers. The model does not decide: the workflow decides what happens next, and anything that touches money or goes out to a customer waits for a person to approve it until the error rate has been measured on real work. An agent is a model allowed to choose its own next step, which is impressive in a demo and expensive at three in the morning.

### Make.com consultant cost in 2026, with a rate card and n8n math
/blog/make-com-consultant-cost (Pricing, 2026-08-26, updated 2026-09-01)
A Make.com consultant costs $85 to $150 an hour, or $650 to $1,100 a day, for an independent specialist in the US or UK, the band most successful projects come from. Fixed-price scenario builds run $350 to $1,500 for a single scenario and $4,000 to $15,000 for a multi-scenario system. The platform bill decides the five-year cost more than the build price does: at 10,000 runs a month the same 20-module scenario costs roughly $500 on Make and roughly $40 on self-hosted n8n.

### Fractional AI engineer cost in 2026 and the break-even math
/blog/fractional-ai-engineer-cost (Pricing, 2026-08-26, updated 2026-09-01)
A fractional AI engineer costs $6,000 to $12,000 a month for a named engineer at 8 to 20 hours a week. The published market runs from $3,000 to $30,000, and the spread is mostly advisors against builders. A fully loaded full-time senior hire is $13,500 to $15,000 a month, so per hour fractional is 20 to 60% more expensive. Above roughly 25 hours a week of sustained work, hire the person. Below about 15 hours a week, fractional wins on total cost even at the higher hourly rate.

### Why AI automations die in production, and the six usual causes
/blog/why-ai-automations-die-in-production (Engineering, 2026-08-22, updated 2026-09-01)
Working AI automations die in six ways, and the model is almost never one of them: credentials expire or get revoked, runs succeed while doing no work, an API changes shape on the other side, rate limits and quotas hit at real volume, Continue On Fail gets used as error handling, and nobody is named to watch. Each has a signature visible in an execution list without opening anything: a run that normally takes twelve seconds finishing in 400 milliseconds did nothing at all, and it is sitting there marked as a success.

### AI automation agency pricing 2026, compared across 20 price lists
/blog/ai-automation-agency-pricing-2026 (Pricing, 2026-08-18, updated 2026-09-01)
AI automation agency pricing in 2026, read off 20 agencies' own websites in August 2026: paid audits cluster at five anchors, £495, $999, £1,500, $2,500 and £3,500. Fixed mid-market builds run £350 to $15,000. Retainers run $200 to $8,500 a month, and day rates £550 to £1,500. Compare like with like and the gap between the cheapest and most expensive published build is roughly seven to seventeen times, not the hundredfold spread the internet repeats.

### Evals for business automations: 30 real cases beat any spot check
/blog/evals-for-business-automations (Engineering, 2026-08-15, updated 2026-09-01)
An eval set is thirty to a hundred real cases from your own operation, each paired with the outcome a competent employee would expect, pushed through the live configuration by a script every time anything changes. It runs in minutes and costs a few dollars. Wired into the deploy: no prompt change, model upgrade or workflow edit reaches production until the run scores at or above the previous baseline: it turns we think the update is fine into the update passed all but two cases, and here are the two.

### n8n workflow keeps failing in production: find the cause upstream
/blog/n8n-workflow-keeps-failing (Rescue, 2026-08-12, updated 2026-09-01)
A red node in n8n shows where a failure surfaced, not where it started: the cause usually sits a few nodes upstream, or outside n8n entirely. Work four checks in order: read the full error body for the status code, compare the runs either side of the failure, pin the input data and run the node on its own, and find the date the failures started. Then harden so it stays fixed: retry on 429 and 5xx but never on 400 or 401, set an error workflow that alerts a person who reads it, and make reruns safe with an existence check or an upsert.

### What a paid AI opportunity audit buys you, and when to skip it
/blog/what-is-an-ai-opportunity-audit (Buying guide, 2026-08-08, updated 2026-09-01)
A paid AI opportunity audit is a short, fixed-scope engagement producing four things: a workflow map showing how work actually travels through the business, a ranked list of opportunities with the arithmetic shown next to each build quote, a 30/60/90 day sequence, and fixed prices in dollars rather than estimates in hours. Published fees ran $999 to $7,500 in the US and £495 to £3,500 in the UK in August 2026. The fair version credits the whole fee against a build started within 90 days. We do not sell one: our first step is a free consultation that ends in a written proposal with a fixed price. If the deliverable is a slide deck about AI trends, you bought a brochure.

### Your chatbot is making things up because it has nothing to read
/blog/chatbot-making-things-up (Rescue, 2026-08-05, updated 2026-09-01)
A chatbot invents answers because a language model does not look anything up: it continues text, and it cannot say I don't know because it has no way of knowing that it does not know. Confidence is a feature of its writing style. What fixes it is changing what the model can see and what it is permitted to say: retrieval over your own documents with an index that actually gets rebuilt, citations on every answer, a refusal path that hands off to a person, and fixed templates for prices and policies so the digits never pass through the generative step at all.

### 10 questions before hiring an AI agency, with the bad answers
/blog/questions-before-hiring-ai-agency (Buying guide, 2026-08-01, updated 2026-09-01)
Ten questions separate agencies that run production systems from agencies that run demos, usually inside one call: who owns the accounts and the code, what happens when it breaks at 2am, whether there is a price before the call, what is in scope and what counts as a change, whether they run evals or vibes, the monthly running cost of their last project, whether the bug warranty exists in clause form, what offboarding looks like, whether you can speak to a client from a year ago, and what work they turn down.

### AI voice agent cost in 2026, from $49 a month to a $15,000 build
/blog/ai-voice-agent-cost (Pricing, 2026-07-28, updated 2026-09-01)
An AI voice agent costs about $0.05 a minute to run, roughly $47 a month for 300 three-minute calls on OpenAI Realtime with LiveKit and Twilio, at rates read off vendor pricing pages in August 2026. A cascaded Deepgram and ElevenLabs pipeline is about $10 cheaper at that volume, and platform products land between $78 and $294 a month for the same 900 minutes. The headline per-minute figure a platform prints is a component price, not a bill. Building the agent is a separate line entirely: $6,000 to $20,000 fixed.

### Zapier vs Make vs n8n in 2026: where each one wins on cost and ops
/blog/zapier-vs-make-vs-n8n-2026 (Engineering, 2026-07-25, updated 2026-09-01)
Pick by volume and by who owns the instance. Under roughly 2,000 runs a month with simple flows and no technical staff, Zapier: the premium buys reliability you never think about. Branching logic at moderate volume on a tight budget, Make. High volume, custom logic, AI-heavy workflows or data residency, self-hosted n8n, but only if someone competent owns the instance, and that someone can be a vendor and cannot be nobody. At 10,000 runs a month of five-step workflows the same logic costs $150 to $450 on Zapier, $30 to $100 on Make and $25 to $40 self-hosted.

### Why your voice agent drops calls and double-books, with the fixes
/blog/voice-agent-drops-calls (Rescue, 2026-07-22, updated 2026-09-01)
Every broken voice deployment we have scoped fails in one of five ways: latency long enough that callers assume the line is dead, barge-in configured wrong in one direction or the other, a check-then-book race that double-books a slot, a confirmation spoken before the booking system ever replied, and background noise wrecking names and phone numbers. Aim for under 1.5 seconds from the caller finishing to the system starting to speak, and score twenty real calls against the transcript instead of judging the system by a demo from a quiet office.

### Why AI projects fail: five causes behind the 42% abandonment rate
/blog/why-most-ai-projects-fail (Buying guide, 2026-07-18, updated 2026-09-01)
S&P Global's 2025 survey found 42% of companies had abandoned most of their AI initiatives, with the average company scrapping 46% of its pilots. The causes are boring and they repeat: no named owner, failure that is invisible by default, the wrong thing automated, a tool pushed past its limits, and nobody counting value monthly. The projects that survive share a shape: small, short, and owned by somebody who does not get paid until it works.

### Freelancer disappeared and the automation broke: get access first
/blog/freelancer-disappeared-broken-automation (Rescue, 2026-07-15, updated 2026-09-01)
When the builder stops replying, accounts come first and fixes come last. List every account the system could touch, automation platform, hosting, domain and DNS, the model provider, databases, CRM, calendar, email, and ask one question of each: can someone in your company log in today? Then add visibility before changing anything: execution logging, failures routed to a channel someone actually reads, and a written baseline of runs and failures per day. Then document each workflow in a paragraph. Only then stabilise, smallest change first.

### The monthly running costs of AI automation, and how to cap them
/blog/running-costs-nobody-mentions (Pricing, 2026-07-10, updated 2026-09-01)
AI automation running costs come from four meters: the automation platform at $20 to $80 a month, model usage at $10 to $150, voice minutes at $0.10 to $0.20 each, and hosting and storage at $10 to $50. A single-workflow deployment lands at $40 to $150 a month all in; a multi-system deployment with voice reaches $300 to $500. Every one of those meters sits in a dashboard you should be able to open yourself, because every account should be in your name at the provider's list price.

### Automation handover checklist: 10 things you hold at the end
/blog/automation-handover-checklist (Engineering, 2026-07-08, updated 2026-09-01)
A proper automation handover moves ten things onto your side, ordered by how much it hurts to be missing them: ownership of every platform and vendor account including billing, credentials and API keys on one page with the human behind each OAuth consent, environment variables and container-only secrets, workflow exports in a git repository you control, an error workflow whose alert reaches a living person, execution history and its retention setting, one page of architecture plus a runbook, test data and a safe way to run end to end, vendor and API dependencies with versions and deprecation dates, and a recorded call in which somebody walks one real failure end to end.

### Production-ready AI checklist: six things a demo never proves
/blog/production-ready-ai-checklist (Buying guide, 2026-07-03, updated 2026-09-01)
Production-ready should mean a set of properties you can verify without being technical, not that the demo worked while someone was watching. There are six: reliability, meaning retries, timeouts, dead-letter handling and idempotency; someone finds out when it breaks; quality expressed as a score against real cases rather than a mood; documentation a stranger could operate from; every account and repository in your company's name; and after launch, a named watcher plus a monthly report with numbers in it. Send the list with your request for proposal and ask which items are included at the quoted price.

### Fix or rebuild a broken AI automation: most repairs are under $4,000
/blog/fix-or-rebuild-broken-ai (Rescue, 2026-06-30, updated 2026-09-01)
Fix when the failure has an address: when someone can say this node, this prompt, this webhook and explain the chain from cause to symptom. Rebuild when the foundation is wrong: a no-code tool pushed past its envelope, no data model, spaghetti nobody can trace, or a system built in the wrong shape entirely. The prices say the same thing. Small fixes run $750 to $1,500, most rescues $1,500 to $4,000, and a rebuild $4,000 to $10,000, so a rebuild recommendation is a claim that repairing your system costs more than $4,000, and it should arrive on paper with reasons attached.

### What a care retainer should include, and what it should cost
/blog/what-a-care-retainer-should-include (Engineering, 2026-06-26, updated 2026-09-01)
A care retainer is worth paying for when it contains four things: monitoring that pages a human, including an alert when runs quietly drop to zero; a fix window written into the contract rather than promised on a call; one named improvement shipped every month; and a one-page monthly report the owner reads in three minutes. Ours are $1,000, $1,800 and $3,000 a month, inside a market that runs $500 to $8,500. Skip it if you have a single workflow with no AI in it, or a technical person in-house willing to own the system.

### Hourly vs fixed price for AI automation: why fixed usually wins
/blog/hourly-vs-fixed-price-ai-work (Pricing, 2026-06-24, updated 2026-09-01)
Fixed price is the right default for AI automation work, for a reason that has nothing to do with cost certainty: a vendor cannot quote a fixed price without first defining the scope, and vague scope is one of the main causes of abandoned AI projects. Hourly billing also pays a vendor more for taking longer and nothing for being clever. Hourly earns its place in two cases, true research, and untangling an undocumented legacy system, and both are better bought as a small paid discovery phase that turns the unknowns into a scoped fixed build.

### Why we publish prices, and the best argument against it
/blog/why-we-publish-prices (Buying guide, 2026-06-19, updated 2026-09-01)
We publish prices because hiding them moves the shopping cost onto the buyer: a form, a wait, a 45-minute discovery call that is mostly questions about your budget, times the four or five agencies you should be comparing. And a price cannot be published until somebody has defined what it buys, which is the same scoping discipline that decides whether the project succeeds. The honest caveat is that we publish partly because we are small: two agencies in our survey removed their price lists during 2026 as they moved upmarket, and the arithmetic behind that is sound.

### Agency vs freelancer vs in-house for AI automation, with 2026 costs
/blog/agency-vs-freelancer-vs-inhouse (Pricing, 2026-06-15, updated 2026-09-01)
A fully loaded in-house automation engineer costs $13,500 to $15,000 a month in the US, and every other option is measured against that number. A freelancer is cheapest per project at $2,000 to $8,000, but carries the continuity risk that started half the broken systems we have scoped for rescue. An agency quotes $3,000 to $15,000 for the same scope, and the premium buys a fixed price, a warranty that survives delivery, and more than one person who understands your system. Hire in-house only when the automation queue is genuinely full time.

## Contact
- Email: sales@lurtoagency.com
- Company: Meridal Group LLC, Sheridan, Wyoming, United States. UK address: 71-75 Shelton Street, Covent Garden, London WC2H 9JQ. Serving US and UK clients.
- Primary way in: the written diagnostic at https://lurtoagency.com/diagnostic. No call required.
- Nothing built yet: a free 30-minute consultation, booked at https://lurtoagency.com/contact, with a written fixed-price proposal within two business days.