AI agency in London that ships working software, not pilots

Three lanes, all production-grade: on-site agents that qualify inbound enquiries, AI features inside products you sell, and internal automation when repeat ops justify the build. We use Anthropic Claude and OpenAI API models, tier chosen per latency, cost and task, not logo. Embeddings and RAG when search over your docs is part of the brief.

We build on-site AI agents for modern companies: trained on your offer, scoped to qualify inbound enquiries, with CRM or email handoff. This is agency-led development (scope, build, deploy, monitor), not a self-serve widget. Product detail and pricing bands on AI chatbots & agents; this page is the London agency view.

Buyers search chatbot; we often deliver an agent. Same surface (on-site chat), tighter job: qualify, capture structured data, hand off. A chatbot that only answers FAQs is cheap to build and easy to ignore. An agent asks what your sales team would ask and filters tyre-kickers before a human picks up. We use both words deliberately: chatbot is what you Google; agent is what we ship when the brief is pipeline, not novelty.

From £20/mo for an on-site chatbot/agent on your marketing site. Scoped automation and in-product AI from £3,000/sprint. See pricing.

NextFlow CRM (see the full NextFlow case study) includes AI-drafted replies inside the unified inbox. A service-business CRM we built for a client: lead capture embeds, quotes, payments and pipeline kanban. Before AI replies, owners rewrote the same email ten times a week; now drafts land in-thread for review and send. That is the pattern we replicate: AI where it removes repeat cognitive work, with a human approve step. See web app development for product builds.

Also shipped for clients: CourseOS (AI course structuring from video links). Insighto is our own behavioural analytics product: AI-assisted session replay and drop-off detection.

Scoped engineering when chat is not enough: CRM hooks, webhook sync, document routing, internal tools. Quoted per engagement after we know volume and ROI, typically from £3k/sprint inside a wider web app or integration brief. We do not sell vague “AI transformation” retainers.

Full rate card: Pricing. Model API costs (OpenAI / Anthropic usage) are either included in the retainer band or passed through at cost with caps, stated in writing before sign-off.

Named live work: client builds including NextFlow CRM and CourseOS, plus Insighto (our behavioural analytics product). Client AI chatbots run under NDA; ask on a call. Full write-ups: Results.

You own the prompts, flows and custom code. On project completion or retainer exit, we transfer repositories, prompt libraries and integration configs to your accounts. We do not hold hostage models or training data.

Data handling: Visitor chat content is processed via API calls to Anthropic/OpenAI under their enterprise terms. We configure zero-retention where available. Your CRM and site data stay in your stack (HubSpot, Salesforce, custom DB, etc.). We do not train foundation models on your conversations. Sub-processors listed in our Privacy policy; DPA available for clients on request.

API keys: Preferably in your org’s OpenAI/Anthropic account so billing and limits stay visible to you. We can host during build; production should be yours.

Skip the chatbot if:

• Traffic is too low: under ~500 relevant sessions/month, you will not learn fast enough; fix positioning and CRO first.
• The offer changes weekly: without a stable brief, prompts become whack-a-mole; nail the offer, then automate qualification.
• You need legal precision every time: regulated advice (financial, medical, immigration) needs human sign-off architecture, not a friendly widget pretending certainty.
• The real problem is speed-to-lead: if nobody replies to form fills for 48 hours, an agent will not fix a broken sales process; fix humans first.
• You want “AI” for the pitch deck: investors and boards see through it; build where repeat questions or drafting actually cost you hours.

We will tell you this on the first call. Better to earn trust than sell a retainer you will cancel in ninety days.

Studio at 66 Paul Street, London EC2A 4NA (Shoreditch). Most AI work is remote: async updates, shared staging, video when useful. London clients often kick off in person at the studio or a nearby meeting point; we reply within one UK working day, with 24h target on active retainer work.

AI agency UK: same team, same pricing, same process whether you are in London, Edinburgh, Manchester or Bristol. No separate “UK AI agency” URL; this page covers both London intent and UK-wide delivery. For the web-and-design view of the same studio, see London web design agency or UK-wide delivery.