Skip to content

AI and agents

AI assistants, agents and automations that answer from your own content, act through your existing systems and hand decisions to people.

A woman on a phone call at her desk with a laptop open

DelacoTech builds AI into the work your organisation already does: answering customer questions from approved content, drafting proposals and reports, processing documents, and watching data for problems a person should see.

Each system starts with a defined set of tasks and the sources it may use. Models run server side, tools are limited to what the task needs, and anything that commits your organisation to a price, a date or a decision goes to a person first. The project assistant on this website is one of these systems.

Customer assistants
Website and WhatsApp assistants that answer from your published content, ask follow-up questions and prepare a brief for your team.
Internal agents
Agents that search your records, draft proposals, summaries and reports, and leave the final call to staff.
Workflow automation
Scheduled and event-driven jobs that move data between systems, send alerts and remove repeated manual steps.
Forecasting and anomaly detection
Models that predict output and flag readings that fall outside the expected range, as built for Croft Group's farm platform.
AI features in products
Chat, search and summarisation inside an existing product, such as the property chat assistant in Scansan's Property Analysis Assistant.

How the work runs

  1. 1

    Define

    List the questions and tasks the system should handle, the ones it must refuse, and who reviews its output.

  2. 2

    Ground

    Choose the approved sources and system connections it may use, and nothing beyond them.

  3. 3

    Build

    Connect a server-side model through a gateway, add tools, guardrails, logging and spending limits.

  4. 4

    Test

    Run it against real past enquiries or records, fix the failures, then launch with monitoring in place.

Try it

The project assistant on this siteis one of these systems

It reads only DelacoTech's published solutions, case studies and hardware list, asks follow-up questions, and writes a brief that a person on our team prices.

Open the assistant
Example conversation with this site's assistant
  1. We run six pharmacies. Customers keep calling to ask if a drug is in stock. Can they check on WhatsApp instead?
  2. Checked DelacoTech solutions: AI and agents, Data
  3. Yes. That is a WhatsApp assistant reading your stock records. Do all six branches share one stock system, or does each keep its own?
  4. Each branch has its own spreadsheet.
  5. Then the first step is a shared stock record the assistant can read. I've added that to your brief for the team to price.

Project brief sent to DelacoTech

WhatsApp stock assistant for six pharmacy branches

Croft Group's farm platform, one morning

Sample data, modelled on the pipeline we built for Croft Group

  1. INFarm recordsInput

    412 entries from 4 houses: eggs, feed, deaths

    Waiting0.00 sec
  2. ETClean and loadETL job

    3 duplicates removed, 2 missing feed weights filled

    Waiting0.00 sec
  3. FCForecastModel

    House 3 expected today: 3,450 eggs

    Waiting0.00 sec
  4. ADAnomaly checkRule

    House 3 recorded 2,860, 17% under forecast

    Waiting0.00 sec

OUTHouse 3, eggs per day

2,8003,2003,600Day 1Day 7Day 14
Recorded Forecast, 4% range

Alert to farm manager · 07:04

House 3 laid 2,860 eggs, 17% under forecast.

Feed intake in the same house fell two days in a row. Check water lines and bird health before the next collection.

Croft Group case study

Related work

Questions

  • Which AI models do you use?

    We are not tied to one provider. We route requests through a gateway, so a system can use Anthropic, OpenAI, Google or other models and switch when a better or cheaper model is available.

  • Will an assistant make promises to my customers?

    No. We set the system to refuse prices, delivery dates and commitments, and to pass those to your team with a written brief.

  • What does it cost to run?

    Model usage is charged by the model and the number of tokens processed. We set spending limits on the project and estimate monthly usage from your expected volume before launch.

  • What happens to our data?

    The system only reads the sources you approve. Requests run from your server, not the visitor's browser, and we agree retention and logging rules with you before launch.

More in Digital Solutions

What do you want to build?

Our project assistant asks a few questions and writes a brief. A person at DelacoTech replies with scope, price and dates.