Business analysis · Agentic operating model · Nov 2026

A business where AI agents do the routine work, and people do the relationships.

A department-by-department plan for PowerXpress: what each AI agent does, the tools it uses, which are free and which are paid, what it costs to run, and where a person still signs off.

8
Departments covered, with a governance layer across all of them
–
Automated jobs described in this plan
~440 h
Staff hours a month freed (sum of the department estimates)
–
Recommended running cost a month (software, AI and upkeep)

Costs are indicative GBP estimates. Prices change, so check each supplier before you buy. Time and revenue figures are targets, not promises.

Senior business analyst view

What I found, and what it means for the plan

PowerXpress is unusual: the hard part is already built. The gap is not software, it is the glue between the apps and a stream of leads big enough to feed them.

Strengths to build on

  • A working stack: Lead Hub CRM, Quote Desk with live Magento prices per customer group, Stripe pay links, Marketing Hub.
  • Two stores and three price groups already modelled, so agents can quote correctly.
  • Product Asset Library, Installer Intelligence and Price Desk already exist as data sources.

Gaps that cost money now

  • Reactor chat uses the wrong price groups on its live host and is not on the site.
  • Chat leads go to email and Sheets, not the CRM.
  • No reviews on site, an empty delivery line on product pages, a hero link to the wrong store.
  • A stock feed failed silently for two weeks. Nothing watches the feeds.

The honest constraint

  • Real order volume is low, around 8 a month in the last review. Agents multiply demand, they do not create it.
  • So the first money goes on lead flow, speed and measurement. Heavy back-office automation waits until volume justifies it.
  • Start on the Lean tier and move up when the numbers earn it.

On "no human intervention"

For standard orders this is realistic: find, qualify, contact, nurture, quote, negotiate inside a fence, take payment and create the order, all without a person. The fence is set in software, not in the AI's judgement: price floors, discount ceilings, spend caps and a kill switch. Anything outside the fence, such as a big commercial job, a non-standard system, credit terms or a refund, is prepared by the agent and closed by a person. That is where the largest deals are won anyway.

How the model works

One brain, many agents, clear limits

Every agent is a Claude-powered worker that uses your existing apps as tools through one shared layer (the PowerXpress Brain). It reads data, drafts or acts, and writes everything to an audit log.

L0

Suggests

Shows a recommendation. A person does the work. Used for new, risky tasks.

L1

Drafts

Prepares the email, quote or order. A person clicks approve.

L2

Acts within limits

Sends and updates on its own inside server-side limits. Exceptions go to a person.

L3

Runs by itself

Fully automatic, with random sampling and a weekly scorecard.

Tools the agents use

Lead Hub, Quote Desk, Marketing Hub, Magento, Stripe, Pipedrive, CloudTalk, Gmail, Brevo and WhatsApp, each through a narrow, permissioned API.

Rules live in code

Price floors, discount bands, send limits and budgets are enforced by the server. An agent cannot talk its way past them.

Everything is visible

Audit log, approval queue, agent scorecards and a kill switch in the Digital Sales Hub.

Department 1 · Sales

Find, win and close: the full sales journey

Nine stages from a stranger to a repeat customer. Tap a stage to see what the agent does, the tools, and where a person steps in.

Straight-through rule (zero-touch orders)

A standard system inside the price book, under a set order value, with a verified address and no special requests goes from first contact to paid order with no person involved. Anything above the threshold or off-book is prepared by the agents and closed by a person within the hour. The threshold is a decision for you (suggested starting point: £5,000 for homeowners, £10,000 for trade).

Department 2 · Marketing and your agency

Supplying the right information to marketing

Agencies and marketers do their best work when they get the right facts at the right time. Eight automatic feeds, plus one pack every Monday, so nobody has to ask or dig.

The Monday agency brief (08:00, automatic)

  • What worked last week, by campaign, with leads, quotes and paid orders (not just clicks).
  • Budget moves proposed, with the data behind each one.
  • Top five content ideas drawn from real chat questions, searches and call themes.
  • Offers and stock to push, and stock to avoid promoting.
  • Claims register: what you may and may not say this week.

The closed loop that makes ads smarter

Today ad platforms only see clicks. The agent sends them what really happened: lead created, quote sent, order paid, with staff orders removed. Google and Meta then optimise towards paying customers, and the agency sees true cost per sale.

Google offline conversions: freeMeta Conversions API: free

Marketing channel agents

Every department

The minor details, department by department

Each job reads: when it starts, what the agent does, which tools it uses, and the autonomy level. Open a department to see its jobs.

Tool stack

Which tools are free, which are paid

Many of the building blocks are things you already pay for or can use free. Only a handful of new subscriptions are needed.

ToolWhat it does for the agentsStatus£ / month

"Already have" means you pay for it today, so it is not added to the plan. Pay-as-you-go items depend on volume. Ad spend is separate and not included anywhere on this page.

Expenses

What it costs: build once, then run

Three tiers. Lean proves the model with the fewest new tools. Recommended is the full plan. Scale adds more volume and tooling for the group.

One-off build

Item£

Monthly running cost

Item£ / month

Updated from the earlier growth plan: the monthly list now includes lead intelligence, SEO tooling, scraping and call transcription. Existing subscriptions (Pipedrive, CloudTalk, Google Workspace, Magento hosting) and Stripe card fees are already paid and are not counted. All figures are estimates to firm up with quotes.

Break-even

What you need to believe

Instead of a promise, a calculator. Move the sliders to see how much extra monthly sales the plan has to produce to cover its running cost and pay back the build.

Hours are valued at £22 an hour. They only count as money if the time is redeployed or a hire is avoided.

Guardrails

How it stays safe, legal and on-brand

Roadmap and decisions

Where to start

Decisions needed from you