Eight weeks from an empty repository to a live AI platform.
LeadSignal is a multi-tenant AI setter with an approved Meta app, live payments and paying tenants. We did not commission it, brief it out or buy it in. We built it with the same method we sell, and we run our own pipeline on it every day.
The only case study where we were the client, carried the risk, and cannot quietly edit the brief afterwards.
Every agency says it can build with AI. Almost none has shipped anything.
The honest test of a firm selling AI systems is not a deck. It is whether they have put a real product into production, taken money through it, and kept it running long after the interesting part was over.
LeadSignal is that test, done in public. It handles other businesses’ customer conversations, holds their data, takes their payments and answers to a platform’s app review. There is nowhere in it to hide a shortcut, which is exactly why it is the work we point at.
Not a demo. A product other businesses run their sales on.
A multi-tenant platform
Every business gets its own isolated workspace, data, users, settings and billing. Tenancy is the hardest thing to retrofit and the thing most internal AI projects discover far too late, so it went in first rather than last.
An agent architecture
Separate agents for conversation, support, marketing and outreach, each with its own knowledge, its own voice rules and its own limits. They score, reply, follow up and hand back to a human on rules written down in advance rather than on a hunch.
Live payments and platform approval
Real subscriptions, real invoices, real failed-card handling, and an app that passed review to hold production messaging permissions. The unglamorous half of shipping, and the half that separates a product from a prototype.
The eight weeks, week by week.
This is the sequence we use on client systems, run at full speed on our own. Look at where the weeks actually went: scope and data before anything clever, production traffic early, billing and review before launch rather than after it.
Requirements, before a line of code
What a lead is. What the agent may say, and where it must stop and fetch a human. The data model the whole thing hangs off. The same conversation we charge clients for, because skipping it is what makes month seven expensive.
The spine
Tenancy, authentication, the lead record and the CRM around it. Boring, load-bearing, and the part that decides whether everything after it is cheap or painful to change.
Channels, in production early
Instagram, Facebook, WhatsApp, SMS and X wired in and taking live webhooks against real accounts. Integrations are where these projects actually fail, so they were proven on real traffic rather than on a mock at the end.
The agents and the guardrails
Scoring, drafting in the client’s voice, follow-up timing, opt-out handling, quiet hours and the handback rules. The guardrails were built alongside the capability, not bolted on after it misbehaved.
Billing and entitlements
Plans, subscriptions, trials, and what each tier may switch on. The point at which a build becomes a business, and the point most internal AI projects never reach.
Review, hardening, monitoring
Platform app review, rate limits, failure handling, and alerting that tells us a seam has come apart before a customer does. Then live, with paying tenants on it.
Three things a deck cannot.
We finish things
Approval, payments and production traffic are not opinions. Either an app holds live messaging permissions and takes money, or it does not. Ours does.
We run what we build
We are the ones woken up when a platform changes its API without asking. That is the work a retainer actually buys, and we do it for ourselves before we sell it to you.
Eight weeks is a real number
Not an estimate, a marketing figure or a best case. It is what it took, with the requirements work counted in rather than left out to make the number look better.
It was built to land in whatever CRM you already run.
LeadSignal carries its own CRM and pipeline, and syncs two-way out of the box with HubSpot and GoHighLevel. Where a business runs something else — Salesforce, Pipedrive, Zoho, Dynamics, or a system chosen long before we arrived — we build the integration, which is exactly the work we are hired for on client systems.
We know HubSpot best, which makes data configuration faster and cheaper there. It has never been a condition of working with us. If your CRM has an API we can land scored, deduped, attributed leads in it, and if it does not, that is a finding the audit will hand you in writing rather than a surprise in month three.
What eight weeks did not buy.
It bought a working platform, not a finished one. Everything since has been the unglamorous half: channels added, guardrails tightened, edge cases found by real conversations rather than by testing.
That is the honest shape of every AI system, including yours. The build is the short part. Anyone quoting you a date with nothing after it has quoted for the easy half, and the reason we price Run separately is that we would rather say so on the page than discover it with you in month seven.
We will show you ours before we quote for yours.
Two weeks, from £1,500, and a written answer on whether this is worth building for you at all.