AI & Automation
Custom GPT and RAG chatbots that answer from your own data, hand off to humans, and are tested before they go live.
Quick answer
WebStackRank builds a custom AI chatbot for one use case from USD 2,000 (≈ £1,512) at a fixed price: a knowledge base built from your documents, one channel (your website, WhatsApp or Slack), a hand-off to your team, and delivery in about 3 weeks. Model usage is billed by the AI provider to your account.
Prices
| Build | From | In GBP | Included | Timeline |
|---|---|---|---|---|
| AI chatbot, one use case | USD 2,000 | ≈ £1,512 | Knowledge base from your documents (RAG), one channel (site widget, WhatsApp or Slack), human hand-off, question-set test before launch, basic analytics | ~3 weeks |
| WhatsApp Business API setup (if WhatsApp is the channel) | USD 1,100 | ≈ £832 | Account setup, one number, message templates, CRM hook | by scope |
| Chatbot inside a custom web app or portal | USD 3,500 | ≈ £2,646 | Web app with auth, dashboards and API, then the chatbot scoped on top | ~5 weeks |
| Multi-tool / agentic assistant | quoted on scope | — | Tools that read and write in your systems, review queue, evaluation harness | 4–12 weeks |
50% to start, 50% at launch · you own the code, prompts and test set · itemised proposal within 24 business hours.
AI model usage fees (billed by OpenAI, Anthropic or Azure to your account), WhatsApp conversation fees charged by Meta, third-party SaaS licences, content writing for the knowledge base, and integrations outside the scope.
We price one use case and one channel at a time, reuse a tested RAG and hand-off architecture, and work as a small founder-led team from Dubai. You don't pay for a six-month pilot. If the first bot works, the next use case is quoted on top.
Most "Custom GPT" projects in 2026 are a system prompt pasted into the OpenAI Custom GPT builder, branded with a logo, and shipped with no plan for accuracy, no plan for evaluation, and no plan for what happens when the model is wrong. That works for an internal toy. It does not work when a customer-facing assistant gives a wrong policy answer, or when an internal ops copilot hallucinates a SKU that does not exist.
Our chatbot engagements treat the assistant as a piece of software, not a prompt. That means: retrieval augmented generation over your real knowledge base so the model answers from your data; tool use and function calling so the assistant can read live systems (your CRM, your order DB, your scheduling system) instead of guessing; evaluation harnesses with a ground-truth question set so accuracy is measured, not assumed; and human-in-the-loop review on every workflow where a wrong answer has a real cost.
We build on OpenAI or Anthropic models (chosen by testing both against your workload) and host the orchestration on your infrastructure — Laravel, Next.js, or a managed orchestration platform — so your data does not leave a system you do not control. If your data must stay in a region, we can deploy on a cloud region you choose, subject to the provider's availability. We document the data flow in the scope.
The deliverable is a chatbot plus the surrounding software that makes it reliable: the knowledge ingestion pipeline, the eval suite, the review dashboard, the observability that flags when accuracy drifts, and the documentation for your team to extend it. You can swap the underlying model, swap the vector DB, even swap us out, and the system still works.
What We Build
Answers visitor questions from your pages and policies, captures leads and hands off to your team.
The same knowledge base on WhatsApp, through the WhatsApp Business API.
Answers staff questions from your SOPs, contracts and training PDFs.
Drafts replies for your support team in your brand voice with reference to order history. Human reviews and sends.
Qualifies inbound leads against your ideal customer profile, drafts follow-ups, syncs to CRM.
Extracts structured data from invoices, contracts and RFPs. Posts drafts to your accounting or ERP for approval.
Answers buyer questions on product pages from your spec sheets and policy docs.
Single use case from USD 2,000 (≈ AED 7,345). Multi-tool assistants are quoted on scope within 24 business hours. WhatsApp as the channel: WhatsApp Business API setup.
Custom GPT
A custom GPT is a version of ChatGPT configured with your instructions and files. For internal-only use cases with a small audience, the OpenAI Custom GPT product is genuinely fine and we can configure one for you. For production systems with measured accuracy, integration into your existing tools, or customer-facing surfaces, it is a starting point — the production system is an application calling the model's API directly, hosted on your infrastructure with your data controls.
Deliverables
Same components, every engagement. The size of each grows with the use case.
We sit with the people who do the work, not just the manager, and write what success looks like in measurable terms.
Documents chunked, embedded and indexed in a vector store (e.g. pgvector on your Postgres, Pinecone or Weaviate).
Functions the assistant can call against your live systems — read-only by default, write actions gated by human approval.
A ground-truth question set from your real work. Automated regression tests on every model or prompt change.
Where humans see assistant drafts, edit, approve and send.
Cost per session, latency, accuracy on the test set and user feedback (e.g. Helicone or Langfuse).
Architecture diagram, runbook for prompt edits, runbook for adding new tools, hand-over recording.
When the use case demands it, every assistant reply lands in a queue for human review before sending.
Process
How long does it take to build an AI chatbot? About 3 weeks for one use case on one channel; 4–6 weeks for a multi-tool production assistant; 8–12 weeks for agentic systems.
Shadow the people who do the work, define the ground-truth questions, write the scope.
Your documents ingested; a baseline assistant answering a first test set.
The chatbot on your site, WhatsApp or Slack, with hand-off to your team.
Accuracy measured on the question set and tuned; refusal rules for out-of-scope questions.
Go-live with observability and a spending alert on model usage.
Tech Stack
Frameworks compared in depth: how to build a custom AI chatbot for your business.
| Layer | Default | When we deviate |
|---|---|---|
| Model | Current OpenAI GPT and Anthropic Claude models, chosen per use case by testing both on your question set | A cloud-hosted model in a region you choose when data must stay there |
| Vector store | pgvector on Postgres | Pinecone or Weaviate when the collection is very large or latency matters |
| Orchestration | Laravel jobs + queues; Next.js API routes for synchronous flows | A workflow engine for complex multi-step agents |
| Eval | Custom harness with ground-truth questions in Postgres | A hosted eval tool on larger deployments |
| Front-end | Your existing site or a dedicated dashboard | Slack or Teams when staff already work there |
Accuracy and privacy
Answers come from your documents through retrieval, with citations to the source; refusal rules cover questions outside the knowledge base; a test set is run before launch and after every change; and human review sits on any answer with a real cost.
We sign up for the model provider's API terms in your account, set retention, redact personal data where the use case allows, and tell users they are talking to an AI. Under GDPR and UK GDPR we document the data flow and the provider's data processing terms in the scope.
Yes: a website widget, WhatsApp through the Business API, Slack or Teams. One channel is in the base price; each extra channel is quoted.
We are a small team: a good fit for one department or one use case at a time, not for large enterprise programmes that need a big vendor team.
We are a Dubai studio and work remotely with US and UK companies, priced in USD with a GBP reference. City guides: AI chatbot company London.
How to choose
Hiring developers suits a long roadmap of AI features with someone in-house to manage them. A chatbot development company suits a defined use case that has to work at a fixed price, with testing and hand-over included. A freelancer suits a prototype.
Six checks, in writing: the test set and accuracy method; who owns the prompts and data; how hand-off to humans works; a running-cost estimate; where data is processed; and what happens when the model changes. Use the same checks on us.
Adjacent Services
Frequently Asked Questions
From USD 2,000 fixed for one use case on one channel. Model usage is billed separately by the provider.
About 3 weeks for one use case; 4–6 weeks for a multi-tool assistant.
For internal-only use cases with a small audience, the OpenAI Custom GPT product is genuinely fine and we can configure one for you. For production systems with measured accuracy, integration into your existing tools, or customer-facing surfaces, the Custom GPT product is a starting point — the production system is a Laravel/Next.js app calling the model API directly, hosted on your infrastructure with your data controls.
Current OpenAI and Anthropic models, tested on your question set before we lock the choice.
We measure accuracy on a question set built from your real workflow before launch and share the score. Customer-facing answers below the agreed threshold go to human review.
Yes via function calling — read-only by default, write actions gated by human approval. We do not ship fully autonomous write-action AI for customer-facing use cases.
Yes. Full IP transfer on final payment — repository, prompts, test set, infrastructure-as-code, and documentation. You can extend it, replace us, or replace the underlying model without our involvement.
Model usage depends on volume and is billed by the provider to your account. We estimate it from your expected traffic in the proposal and set a spending alert at launch.
Yes. English and Arabic are tested in our builds, and other languages are supported by the models and tested on your question set.
Yes, to email, your helpdesk or WhatsApp, with the conversation history attached.
SaaS AI products are great for cases that fit their feature set. A custom chatbot makes sense when you need integration with your specific systems, your specific knowledge base, your specific tools, and your specific accuracy bar — all of which SaaS products generalise away from.
Case studies
Published case studies from our own products — real screens, verified status, honest cost ranges.
Share your highest-friction workflow and get an itemised proposal within 24 business hours.
How would you like to reach us?
Thanks for chatting!
It was lovely to connect. Come back anytime — our starter packages and their 3 months free maintenance are on the pricing page whenever you're ready. See you again!
See the pricing page
