Services
Explore
Free AI assessmentAnswer 15 questions about one process and see where AI could help.
Start an assessment02 · Agent Deployments
AI agents that handle the steps your team repeats
We build and test AI agents that take on repeated steps in your workflows, in the systems you already use, with a person approving their work.
- Claude
- OpenAI
- Outlook
- HubSpot
- Xero
See one run
An agent at work
Type a task and the agent works through the steps in the tools your team already uses, then stops for a named person to approve the result.
- Tested with your team before wider use
- A named person signs off each result
- Every action recorded
Example · every action recorded
Agents we have built
Agents built for client tasks
Each tab is an agent we built for a client, shown as it runs. Sample records based on each project.
Allbright Law · Contract review against the firm’s playbook
The agent checks incoming contracts against the firm's own playbook, flags deviations with a risk rating and writes a structured summary. A fee-earner validates it before anything goes to the client.
- Read the contractClauses extracted and labelledDocuments
- Check clauses against the playbookThe firm’s own positionsPlaybook
- Flag deviations with a risk ratingHigh, medium or lowModel
- Prepare a structured summaryEach point linked to its clauseDocuments
SolarDirect · Replying to new enquiries and qualifying them
The agent replies to each new enquiry, asks about property, roof, energy use and budget, and scores the lead. Qualified prospects reach the sales team with a brief, and the rest go into a follow-up sequence.
- Reply with the qualifying questionsProperty, roof, energy use, budgetEmail
- Read the customer’s answersChecked for missing detailsEmail
- Score the leadAgainst the agreed criteriaCRM
- Write a brief for the sales teamPre-filled from the customer’s answersCRM
Countrywide · Booking viewings and chasing feedback
The agent confirms applicant availability, books viewings against the negotiators’ diaries, sends reminders and access instructions, and asks for feedback within two hours of each appointment. Vendors get a weekly summary.
Viewing coordination agentCountrywide viewing coordination: the agent reads a viewing request, finds a free slot in the negotiator’s Outlook diary, texts the applicant a confirmation and asks for feedback after the viewing
White Rock · Briefing packs before every client review
Twenty-four hours before each client review, the agent pulls current portfolio data, flags positions outside agreed tolerance bands and drafts an agenda from the client’s objectives and last recorded circumstances.
Client review briefingsWhite Rock client review briefing: 24 hours before a review the agent pulls the portfolio, flags two positions outside their bands, reads last year’s notes and drafts the pack for the adviser
bloobloom · Product descriptions, emails and captions from a spec sheet
The workflow takes a product specification sheet and drafts descriptions, email sequences and social captions in the brand’s tone of voice. The team reviews each draft and keeps editorial control.
Product content workflowbloobloom product content: the agent reads a product sheet, applies the brand voice and drafts the Shopify page, launch email and Instagram captions for the marketer to approve
A plumbing and heating business · From quote request to booked job
Customers submit job details through a form. The agent checks engineer availability by location, prices the job from the business’s rate cards and sends the quote. Accepted quotes are booked into the engineer’s diary.
- Check engineer availability by locationFrom the job management systemJobs
- Price the job from the rate cardsThe business’s standard ratesRate cards
- Send the quoteWith the available slotsEmail
- Customer accepts the quoteReply or link in the emailEmail
- Book the engineer and confirmAdded to the engineer’s diaryDiary
Sparkles · Quotes, invoices and Xero
AI workflows follow up quotes, chase overdue invoices and reconcile Xero by service. The owner of the Llanelli cleaning company got over 15 hours a week back.
Quotes, invoices and XeroSparkles invoices and Xero: the agent reads the bank feed, matches payments by cleaning service in Xero, finds an overdue invoice and drafts a reminder for the owner to approve
Overview
What Agent Deployments means in practice
You get an agent doing one repeated task, measured against how it runs today.
An agent is software that reads what comes in, decides what to do with it within the rules we agree, and does it. We agree what it can access, which actions a named person approves and how unusual cases reach that person, then test it on your own records before wider use. We choose the model for each task on performance, cost and data handling. Where ordinary rules-based automation would do the job, we say so.
What you receive
- 01A tested workflow in use
An agent carrying out an agreed workflow after testing, starting with a pilot.
- 02Measured results
Time spent, error rates and volume handled, compared with the baseline recorded before the pilot.
- 03Monitoring and logs
Dashboards, alerts and audit logs that record the agent's actions and decisions.
- 04Handover
The code, documentation and operating knowledge, so your team can run and change the agent.
Each agent works within agreed limits and approval points. A written proposal sets out the pilot, its price and the timeline before work starts.
Talk to usBuilt for daily use
Monitoring, error handling and the route to a person go into each agent from the start.
Connected to your systems
Agents read from and write to your existing tools through their application programming interfaces (APIs), so they work on your real records.
A person approves the work
Each agent has agreed points where a named person reviews or approves its work. The agent handles routine cases and sends unusual ones to that person.
After the pilot
Results measured against the baseline
Before the pilot we record how the task works today. The pilot review compares the agent with that baseline and recommends whether to continue.
Example document. Names and figures are illustrative.
Client work
Agent Deployments for clients
Projects we have delivered, with the results measured in each.
Allbright Law · Legal
First-pass contract review down from an average of 72 minutes to 11
Agent DeploymentsSolarDirect · Consumer services
First response from 4 hours to under 90 seconds
Agent DeploymentsWhite Rock · Financial services
Client review preparation time cut by about two-thirds
Agent DeploymentsCountrywide · Estate agency
22% more viewings handled without adding headcount
Agent Deploymentsbloobloom · Retail
Campaign output tripled with the same team and no extra budget
Agent DeploymentsSparkles · Cleaning services
Over 15 hours a week back from quote, invoice and Xero admin
Agent DeploymentsHow it works
How we deliver Agent Deployments
Step 1 of 4
Choose the workflow
We choose a workflow with you and map its inputs, outputs and unusual cases.
What is an AI agent?
Software that uses a language model to carry out agreed steps, such as reading documents, searching records, drafting messages or updating a system. We limit what it can access and which actions need a person's approval. Its output still needs checking, especially on customer or financial records.
How long does a deployment take?
It depends on how many systems the agent connects to and how complex its decisions are. You get an estimated timeline, with its assumptions, before work starts.
How are security and data protection handled?
Each agent gets access controls, audit logging and data handling rules, and where needed runs on your own servers or a private cloud. We agree the rules that apply with you, including UK GDPR and any sector requirements, and the checks before anything goes live.
Which platforms and models do you use?
We choose the model for each task, for example Claude (Anthropic), OpenAI models, Google Gemini or open-source models, on accuracy, cost, speed and privacy needs.
How do agents connect to our systems?
Through APIs, webhooks and direct database connections, depending on what each system supports. The connection work is planned and priced before the build starts.
Who looks after the agent once it is live?
We can monitor and improve it through Managed Services, including model updates. Or we hand it over with documentation and training so your team runs it.
What does an agent cost to build and run?
The build price is set out in the proposal. Running costs, mainly model usage and hosting, depend on volume, and we estimate the cost per task before you commit.
What happens when an agent gets something wrong?
We test each agent on agreed examples before it goes live. In use, unusual or uncertain cases go to a named person, and agreed actions need their approval. These checks reduce errors but do not remove them, so we review the output regularly and fix the rule when we find a mistake.
Also from Reformat Labs
Explore our other services
For teams starting with AIAI WorkshopsPractical sessions where your team uses AI on its own tasks and learns what needs checking.For deciding what to buildAI ConsultancyA written plan of which tasks AI can take on in your business, what each would cost and which to start with.For software built around your processApplication DevelopmentSoftware designed and built around your processes, users and existing systems, then tested with the people who will use it.For tools your team already relies onManaged ServicesSupport, monitoring and improvement for the AI tools and applications your team relies on, with a monthly review.Talk to us
Which task does your team repeat most?
Bring one task your team repeats every day and how it runs now. We will tell you if an agent suits it, or if ordinary automation would do the job.