PowerXpress has already built its own sales stack: the Digital Sales Hub, the Lead Hub CRM, the Quote Desk with Stripe payments, the Marketing Hub and the Reactor AI chat. What's missing is one AI brain that works across all of them, and more customers going into them. This plan adds 18 AI agents that sell, market, fulfil and report around the clock, plus a personal Chief-of-Staff agent for every top manager, so leaders who also run other businesses can steer PowerXpress in 10 minutes a day.
The figures are 12-month targets, not guarantees. Phase 0 measures today's baseline, and every target is then tracked against it.
Four more pages build on this plan: the full agentic business model with tools and costs, the sales and marketing detail, the other departments, and the interactive Company Map.
Findings, operating model, sales journey, free vs paid tools, expenses and a break-even calculator.
Open →
Lead sources, a lead scorer, outreach playbooks, negotiation, and what marketing and the agency receive.
Open →
Customer care, resellers, operations, finance, intelligence and people, plus the leakage control tower.
Open →
The cinematic tour of the company, with every department's workflows animated.
Open →
Customers don't want to buy an inverter. They want lower bills and energy independence. The winner in UK solar and storage e-commerce is the one that turns a confusing technical purchase into a 3-minute guided decision, for both homeowners and installers. AI agents make that possible at scale.
A guided system designer and a 24/7 expert in place of 10 category pages. It sizes the system, checks compatibility, prices it and handles the objection.
Instant quotes, the right stock at the right price, rebate tiers that pull them up, and repeat orders predicted before they ask.
Agents run the routine. Managers get one brief a day and approve only what matters, across every business they're involved in.
I read the code of the six PowerXpress apps. All of them run on Cloudflare Workers, talk to Magento, and already have APIs, sign-ins, roles and audit logs. The agents don't need a new platform. They need one shared "brain" that can use these apps' APIs.
Your Marketing Hub's own first-week review found 0 lead events tracked in 30 days, 296 paid visits with 0 conversions recorded, purchase events counting staff test orders (28 vs about 8 real), and 18 leads in the Hub (17 from Meta). With volumes this low, agents that save staff time matter less than agents that bring in, track and convert more demand. That's why the roadmap puts measurement, speed-to-lead and quote-to-cash first.
These are the figures in the Marketing Hub first-week plan. Some may already be fixed, and Phase 0 re-measures them.
From the repositories: lead hub, quote desk, marketing hub (+UI), shop capture and chat. The Digital Sales Hub portal's own code isn't in these repos, so it's described from its links to the other apps.
Every feature across your apps and the website, marked by status. ✅ Have = built and live. 🔧 Improve = built but unfinished, switched off or not connected. ➕ New = recommended add-on. P1 = do first.
Status comes from the code in your repositories (lead hub, quote desk, marketing hub, shop capture, chat) and the live site check on 14 Nov 2026. Items marked "verify" need confirming in Magento or the admin.
I went through the homepage, a category page, two product pages and the Installed Solutions page in a real browser. Agents will send traffic to these pages, so they need to convert first. The Installed Solutions page is already excellent (0% VAT, MCS installers, free survey form). Most of the problems are on the homepage and the product pages.
P1 = losing sales or trust today. "Verify" means I saw a symptom from outside and someone should confirm it in Magento. The fixes are Page Builder blocks, settings or catalogue data, and you have the conventions for all of them already.
Each stage has an agent that owns it, and every sale feeds data back to make the next one cheaper.
SEO + AI-search (ChatGPT, Gemini, Google AI Overviews), Shopping ads, marketplaces
24/7 Concierge sizes the system, answers questions on site and WhatsApp
Instant quote, bundles, finance, cart and quote recovery
Stock, pallet booking, proactive tracking, install scheduling
Reviews, referrals, rebate tiers, reorders, PowerX app upsells
Every outcome flows back to the shared memory, so the agents improve every week
Pick a flow and press play. Each step shows which agent acts, which of your systems it uses, and where a person approves (amber). The console on the right is what the agent would log.
🗺️Watch the Company Map"Fully automated" means agents do the routine work end to end and people handle exceptions, approvals and relationships. These are 12-month targets for the share of routine tasks agents handle.
Independent sales agents, partner websites (such as the TJ Digital / plug-insolar arrangement), electricians and referrers get their own login. They create quotes from your live catalogue at the right store price, send a branded quote with a Stripe pay link, and earn commission automatically. PowerXpress stays the seller of record: it owns the customer, takes the payment and fulfils the order.
Homeowners. Inc VAT. Fully installed packages at 0% VAT. Resellers quote homeowners at this price and never see trade prices.
Approved installers. Ex VAT, tier prices by quantity. A reseller quoting an installer uses this price.
Wholesalers. The lowest rate, MOQ/pallet quantities. Only resellers with Wholesale permission can quote it.
Each quote line shows the customer price, the reseller's commission and how much discount they can still give. Your cost and margin are never shown.
e.g. 3–5% of net order value
e.g. 8–10% (the TJ Digital MOU uses 10%)
e.g. 10% + volume bonus
The reseller pays for any discount from their own commission, so your margin is protected automatically. Rates are examples. Set them per reseller in the admin.
A Cloudflare Worker + D1, same as the Hub and Quote Desk. Reseller sign-up, login (2-step), My customers, New quote, My quotes, Commission, Statements, Marketing kit, Leaderboard, plus the Reseller Copilot chat.
Magento price feed per group (14/15/16), pricing.js, packages, 0% VAT rules, PDF writer, Stripe pay links and the webhook. A new reseller role and a reseller_id on each quote, with no second pricing engine.
Reseller leads and quotes appear in the Lead Hub (source = reseller). Paid orders are created in the right store (/uk or /uk_b2b) with a reseller attribute. Commission flows to the Finance agent and the Digital Sales Hub.
Try the 🤝 Reseller sale flow in the autopilot animation to watch it end to end.
Five layers. You keep Magento, your CRM, quote system and rebate engine, and the agents sit on top of them. Nothing gets ripped out.
Small fast models handle triage, tagging and FAQs, mid-tier models do most agent work, and the top model is saved for strategy and complex quotes. This typically cuts AI cost by 60–80% compared with using one big model for everything.
Each system is exposed once as a tool (for example "get stock", "create quote", "book pallet"). Every agent reuses it, so a new agent takes days, not months.
Admins and managers draw their own flows ("if a Gold installer hasn't ordered in 30 days → offer…"), and agents execute them. This is the upgrade path for your existing CRM, marketing and quote systems.
Tap a team to filter. Each agent has a human owner, a measurable target and a defined autonomy level (see Guardrails).
Your leaders also run other businesses (V-TAC, LEDXpress, Terracore, SourceHaus…). Each gets one agent that sees all their businesses, filters the noise and delivers one brief on mobile. Data stays ring-fenced per company, and a leader only sees the businesses they're permitted to see.
A 7:30am brief on the Digital Sales Hub and by email (extending the Lead Hub's existing digest), one-tap approvals on mobile, and the Monday team summary the Quote Desk already sends.
One login with a separate workspace per company. The agent can compare across businesses ("which unit needs me today?") without mixing their data.
Anything over the spend, discount or risk limits lands in the leader's approval queue with a recommendation, the evidence and the expected impact.
Each agent starts at L1 and is promoted only after it proves accuracy. "Fully automated" is earned, not switched on.
Reads data and reports. No actions.
Prepares quotes, emails and changes. A human clicks approve.
Executes alone within set limits (for example ≤5% discount, ≤£200/day ad change).
Runs end to end. Humans review exceptions and weekly samples.
Revenue agents ship first because they pay for everything after. Each phase has a gate: the next phase starts only when the KPIs are met.
One-off build cost and new monthly cost. Pick a scenario. Tools you already pay for (Magento hosting, Pipedrive, CloudTalk, Stripe fees, existing staff) aren't counted. Internal team time is valued at market rates, so your actual cash outlay is lower if your own team builds it with Claude Code, as it built the Hub and Quote Desk.
Spread over months 0–9
From month 2, rising with traffic
One-off + 12 months running
This uses the figures in the ROI calculator below. Change them there and this updates.
Indicative ranges for planning, not quotes. Get vendor quotes for WhatsApp (BSP), the reviews platform and any outside dev before you commit. AI cost scales with conversations: at today's volumes it sits at the low end.
12-month targets based on what conversational commerce and automation typically deliver. Each is tracked on the live dashboard against the Phase 0 baseline.
Guided selling, instant quotes and recovery flows convert visitors who currently leave confused.
Agents handle the repetitive 70%, so the team focuses on installers, partners and big deals.
Leaders see sales, margin, stock and cash daily across every business, so decisions come in hours, not weeks.
Enter your real baseline. It updates instantly, and nothing is sent anywhere.
Defaults roughly match your Marketing Hub data (about 8 real orders a month). Sessions and order value are placeholders, so replace them with your GA4 figures.
Counts conversion uplift and traffic growth. Staff hours saved, trade reorders and reseller sales would add to it.
The full one-off and monthly cost breakdown, with Lean / Recommended / Scale scenarios and payback, is in the Investment section.
The quickest wins cost almost nothing: Workers Paid ($5/month), deploying the chat price fix, installing OrderPing, and the P1 website fixes.
| Area | North-star KPI | Supporting metrics | Owner |
|---|---|---|---|
| Revenue | Online revenue & gross margin | Conversion rate, AOV, quote win rate, cart recovery | Sales Director |
| Marketing | Blended CAC / MER | Organic & AI-search share, ROAS, cost per lead | Marketing Head |
| Trade | Active installers & reorder rate | Tier upgrades, rebate liability, days between orders | Sales Director |
| Ops | On-time-in-full delivery | Stock-outs, dead stock, delivery exceptions | Operations Head |
| Customer | Review score & first response time | Resolution time, % solved by AI, CSAT | Service Lead |
| Agents | Accuracy & cost per task | Approval rate, escalations, AI spend vs budget | Digital Lead |
Fix "Contact Sales", add a reviews platform, show delivery and warranty on product pages, run a GA4/server-side tracking audit, and freeze the KPI baseline.
Product compatibility matrix, price tiers and policies loaded into the knowledge base. Appoint owners for each agent.
The Concierge on site and WhatsApp, and the Quote agent drafting RFQs for sales approval.
Morning brief v1 for each top manager. First review of revenue impact, then a go/no-go on Phase 2.