Questions buyers ask before hiring us
Scope, price, ownership, data and search, answered the way we answer them on a scoping call. Each answer is short enough to forward to finance or IT before the first meeting.
Updated
Working with us
How a project starts, what we need from your team and what happens after launch.
7 questions
What happens in the first 2 weeks?
It starts with a 30-minute scoping call, followed by a written scope and a fixed price. If you go ahead, the first 1 to 2 weeks are the Map stage: we interview the people who do the work, watch the tools they use and pull a sample of real data. You finish with a process map, a ranked list of opportunities with cost and effort, and the confirmed scope and acceptance test for the build.
What happens if the pilot does not work?
We stop, and you keep everything we learned. During Map we agree a pass mark with you, such as an accuracy rate on last month's invoices or the share of support questions answered correctly. The Prove stage runs a pilot on your real data for 1 to 3 weeks and measures it against that bar. If it misses, you get a written no-go decision with the reasons, and you never pay for a full build that was not going to work.
What kind of companies do you work with?
Companies with roughly 10 to 500 staff, usually led by an operations, finance, sales or marketing head who owns a process that takes hours every week. You do not need an in-house data team. We do not build consumer apps, run cybersecurity audits or sell general digital marketing. If that is what you need, we will say so on the first call and point you to someone who does that work.
Do you only advise, or do you build what you recommend?
We do both. Strategy work ends in a fixed-price build proposal, and we build what we recommend: the automations, agents, knowledge systems and websites in it. Having to make our own advice work keeps it realistic. If you only want advice, the AI readiness audit and fractional AI lead engagements stand on their own, and our written specs are detailed enough for your own developers to build from.
Will you tell us when AI is the wrong tool?
Yes. If a rule-based automation in n8n, Make or Zapier does the job, we build that instead, because it is cheaper to run and easier to debug than anything involving a language model. A model is worth its cost where inputs are messy: free-text emails, scanned documents, phone calls, questions about your own policies. Sometimes the right fix is a process change or a better spreadsheet, and the audit will say that in writing.
How much time will our team need to give?
Plan on 2 to 3 hours a week from one process owner during a build, plus a 45-minute demo of working software each week. During Map, each person who does the work spends 1 to 2 hours with us in interviews and screen-shares. Someone with admin rights needs to grant access to your tools in the first week. At handover, the people who will use the system get hands-on training.
What happens after the system goes live?
You choose between a monthly care plan and a clean handover. Care plans cost $1,500, $3,000 or $6,000 a month and cover monitoring every run, fixing failures, tuning prompts and a short monthly report. If you would rather run it yourselves, you get documentation, a runbook and training at handover, and we stay on call for questions. Either way the system sits on your accounts, so nothing depends on us staying involved.
Pricing and contracts
Fixed prices per phase, who owns the work and how changes get approved.
6 questions
How does your pricing work?
Each phase of a project has its own fixed price, set in the written scope before anything starts, and it holds unless you approve a change. The figure comes from the band published on that service page, and the scope names what placed your project where it sits in the band. Ongoing work, such as care plans and SEO retainers, is a flat monthly fee.
Who owns the code, workflows and prompts?
You do. The work lives in your n8n or Make workspace, under your OpenAI or Anthropic API keys, in your cloud and in your code repository. Code, workflows, prompts, test sets and data all belong to you, and the contract says so. Nothing sits behind a platform we control or a license we charge for. If you end the relationship tomorrow, you keep a working system and the documentation to run it.
Are AI model and software costs included in the price?
No, and that is deliberate. Usage from OpenAI, Anthropic, Twilio, vector databases and similar vendors is billed to your own accounts at cost, with no markup from us, so you see every invoice directly. The proposal includes an estimate of monthly running costs at your expected volume. You know the fixed build price and the likely ongoing cost before you sign anything.
What moves a project to the top or bottom of its price band?
Mostly the number of systems a build connects and how messy the inputs are. A workflow that moves clean web form entries into HubSpot sits near the bottom of its band. One that reads scanned PDFs, writes to your ERP and routes exceptions to a reviewer sits near the top. Human approval steps, audit logs and regulated data push the price up further. The written scope names each driver, so you can drop one to bring the price down.
Do you sign NDAs?
Yes. Send your NDA before the scoping call and we will sign it within 1 working day, or we can send our own mutual NDA. When a project touches personal data, we also sign a data processing agreement and follow your vendor and security requirements. During scoping we ask for samples in place of full exports wherever a sample is enough, so less of your data leaves your systems before there is a contract.
What if our requirements change mid-project?
We write the change down, price it and wait for your approval before doing the work. Small changes are billed at $175 an hour against a written estimate. Bigger ones usually fit better as a second phase with its own fixed price. Because you see working software at the weekly demo, most changes surface early, when they cost an hour to fix and not a week.
AI, data and security
Where your data goes, which models we use and how answers stay accurate.
6 questions
Do we need clean data before we start?
Not for most automation work. Workflows that read emails, forms or invoices cope with the mess you already have, and we build checks for the cases that break them. A knowledge system that answers from your documents (RAG) is different: it needs those documents in one place we can read, with outdated versions removed. The Map stage measures how much cleanup that takes and puts a price on it before you commit.
Will our data be used to train AI models?
Not through anything we build. We use the business API tiers from OpenAI and Anthropic, and both state that they do not train on API inputs or outputs by default. Consumer chat apps have different terms, so client data never goes through personal accounts. Data stays on your accounts, each person gets only the access they need, API keys live in your secrets manager, and personal details are masked before they reach a model wherever the task allows.
Which AI models do you use?
Whichever passes the tests on your data at the lowest running cost. We usually build with models from OpenAI or Anthropic through their APIs, and sometimes with Google Gemini. When data cannot leave your environment, we run open-weight models in your own cloud. The model sits behind a single interface in the code, so switching providers later is a configuration change plus a rerun of the test set.
How do you stop the AI from making things up?
By narrowing what it is allowed to do and testing it on your real cases. Knowledge assistants answer only from your retrieved documents, cite the source, and say so when they cannot find an answer. Agents call defined tools instead of writing free text into your systems. Before launch we run a test set built from past examples, and anything high-stakes, such as a refund or a contract clause, goes to a person for approval.
Can you work with our IT and security team?
Yes, and we prefer to meet them in week 1. Before the build they get architecture notes: which systems connect, what data moves where, and which accounts and permissions we need. Access goes through your SSO and normal approval process, with one named account per person. Your team can review code and workflows at any weekly demo. We do not run security audits, so your security team or auditor signs off on anything sensitive.
Can you work with regulated data such as health or financial records?
Yes, within limits we agree in writing. Regulated data stays in your environment, and we only send it to a model vendor that will sign the agreement your rules require, such as a business associate agreement for US health data. Where possible we mask identifiers before a model sees them, and every action is logged. We are not lawyers or compliance auditors, so your compliance lead approves the design before the build starts.
Search and GEO
How search and AI-answer work is measured, and what nobody can promise.
5 questions
How is GEO different from SEO?
SEO gets your pages ranked in search results. GEO (generative engine optimization) gets your company cited inside AI answers from ChatGPT, Perplexity and Google AI Overviews. The two overlap heavily: Google says AI Overviews draw on its normal ranking systems, so technical SEO still comes first. GEO adds pages written to be quoted, with direct answers, named sources and specific numbers, plus the same facts about your company everywhere it appears online.
How long until we see results from search work?
Technical fixes usually show in Google Search Console within 4 to 8 weeks, and new pages on competitive queries take 3 to 6 months to settle. AI citations can appear sooner, because ChatGPT and Perplexity search the live web when they answer. A brand-new domain is slower on both counts. Each monthly report shows where you rank and which AI answers cite you, so you can see progress before traffic moves.
Can you guarantee rankings or AI citations?
No, and be wary of anyone who does. Google, OpenAI and Perplexity decide what they show, and they change their systems without notice. What we commit to is the work and the reporting: an agreed list of technical fixes, an agreed number of pages written or rewritten each month, and a monthly report on rankings, AI citations and leads from search. If the numbers stall for a quarter, the report says so and proposes what to change.
How do you measure visibility in AI search?
With the platforms' own data first, then our own checks. Google Search Console counts AI Overviews and AI Mode traffic in its Performance report. Bing Webmaster Tools has an AI Performance report showing citations in Copilot answers and the queries behind them. We tag referral visits from chatgpt.com and perplexity.ai in your analytics, and each month we run a fixed set of buyer questions through the main assistants and record who gets cited.
Do we need llms.txt or special schema to appear in AI answers?
Not for Google. Its documentation says AI Overviews and AI Mode need no special files or markup beyond normal SEO. We still add an llms.txt file and markdown versions of key pages, because some AI agents read them and they cost little to produce. Organization, Service and Article markup helps search engines identify your company and what you sell, provided it matches what the page visibly says.
Ask the question this page does not answer
Bring it to a 30-minute scoping call, or send it to [email protected].
We reply within 1 working day with times for a call.