Skip to content
TRDThe Research Desk, home

04 · AI & Integrations

AI automation where it is reliable, with checks where it isn't

Language models are good at reading messy input, classifying it and drafting a first response. They are not good at being the only safeguard on money or commitments. We design AI features with that line drawn clearly, so they can save time without adding new risk.

Free first conversation · No obligation · You own the code

§ 01Is this you?

You're likely in the right place if…

  • 01Your team reads and re-types information from emails, PDFs or forms.
  • 02Enquiries need sorting and routing before anyone can act on them.
  • 03You receive calls or messages that follow the same few patterns.

Sound familiar? Tell us which one on WhatsApp (opens in a new tab) and we'll suggest where to start.

§ 02Deliverables

What we build

  1. 01

    Document and email extraction

    Pull structured data from invoices, forms, CVs or emails into your systems, with confidence checks.

  2. 02

    Classification and routing

    Tag, prioritise and route enquiries or tickets to the right person or workflow.

  3. 03

    Chat and voice agents

    Assistants that answer routine questions and take structured requests over chat, WhatsApp or phone, handing off to humans cleanly.

  4. 04

    Drafting assistants

    First drafts of replies, summaries and reports that staff review before sending.

§ 03Approach

How we build it

  1. Human in the loop

    Low-confidence results and anything involving money or commitments go to a person for review.

  2. Measured, not assumed

    We test against real samples of your data before rollout and show you the error cases.

  3. Your data, your keys

    Model provider accounts are in your name, with data-retention settings chosen deliberately.

Typical stack

  • OpenAI API
  • Anthropic API
  • n8n
  • Node.js
  • Python
  • Supabase
  • Twilio
§ 04FAQ

Questions about ai automation

01Will AI replace our staff?

That's not how we design these systems. The aim is to remove repetitive reading, sorting and typing so your team spends its time on decisions and customers.

02Is our data used to train AI models?

We use providers and settings where your data is not used for training, and set up the accounts in your name so you control retention and access.

03How accurate is AI extraction?

It depends on your documents. We test on real samples before rollout, show you the errors, and route low-confidence results to a person for review.

04Can an AI agent answer customers on WhatsApp or the phone?

Yes, for routine questions and structured requests, with a clear handover to a human for anything else. We define what it may and may not answer before it goes live.

05What does AI automation cost to run?

Model usage is billed by the provider per request. We estimate the monthly cost from your volumes before building, and you pay the provider directly.

06How is the price decided?

After a short study of your process we send a written scope with fixed deliverables, timeline and price. Nothing is charged before you approve it; payment is typically 50% to start and 50% on delivery.

07Who owns the work when it's finished?

You do. Code, accounts and documentation are handed over in your name once the project is paid.

Next step

Ready to talk about ai automation?

Tell us which systems need to talk to each other, or where AI could take work off your team. We'll say honestly whether it's reliable enough.

Free first conversation · No obligation · NDA on request