A marketing engine your own team operates, with no outside AI service.

Summary

Most companies smaller than the very largest run marketing as a patchwork. There is an agency on retainer for the work nobody in-house has time for. There is a tool for scheduling posts, a CRM holding the customer records, a screen of website statistics, a design tool and a writing assistant, each holding a different fragment of what the company knows about itself. And increasingly there is a habit — often undeclared — of pasting positioning notes, customer lists and pipeline detail into text generators owned by somebody else.

Runink PULSE: Prescriptive Unified Lead & Social Engine replaces that patchwork with one application that a marketing team operates directly. It audits your website and social presence and scores it against nine measures. It researches your market, finds competitors and partners worth knowing, and prospects the public web for leads. It writes the posts, whitepapers, decks, short videos, podcasts, infographics, reports and courses that carry somebody from first hearing of you to buying from you — what marketers call the funnel. It maps customer journeys, runs structured 30-day follow-up cycles, and reads the numbers back to you.

Two properties separate it from the alternatives.

The first is that every publishing channel is disarmed until you arm it. Every draft — a post, a cold email, a whitepaper, a call script — is staged with an explicit status and an approve and a reject against it, and a channel that has not been armed publishes nothing whatever status a piece carries. Page 18 sets out how those two controls work together.

The second is where the work happens. The reasoning that does the writing and the analysis runs on the machines your plan names: your own servers on Enterprise, your own cloud account on Dedicated, or Runink’s shared machines on Lite. Your positioning, your customer records and your pipeline stay on those machines, and no outside AI service is called. This is a property of how the product is built, not a policy setting you have to trust somebody to honour.

The problem, in your terms

Ask a marketing lead at a company of two hundred people what they are working on and you will hear a version of this:

“I have a website nobody has properly looked at in eighteen months. I have an agency that does good work but takes three weeks to turn around a brief and costs more than a headcount. I have a scheduler with half a calendar in it. I have a CRM the sales team updates when they remember. I have an analytics dashboard I open on Mondays and cannot really act on. And I have a content plan I wrote in January that we abandoned in March.”

None of these is a failure of effort. They are the predictable result of building a marketing function out of parts that were never designed to share what they know.

The tools do not talk to each other. The audit tool knows your site is slow. The content tool does not, so it keeps writing posts that land on a page nobody stays on. The CRM knows a lead went quiet in week three. The scheduler does not, so it keeps posting into a channel that lead never reads. Every tool holds one fragment and none holds the picture.

The work is slow because every step waits for the one before it. A brief goes to an agency. The agency schedules it. It comes back. Somebody reviews it. It goes back for a revision. By the time it publishes, the market moment that prompted it has passed. The turnaround is measured in weeks and the market moves in days.

The output is generic because the inputs are generic. A writer who does not have your customer records, your positioning history or your pipeline writes from your website and their own experience of your sector. That produces competent, forgettable copy. It is not their fault. They were never given the material that would make it specific.

And the effort does not compound. Every campaign starts from a blank page, because nothing learned in the last one was written down where the next one could read it.

What the fragmentation actually costs

The cost shows up in four places.

The time it takes. The gap between deciding to say something and having it published decides whether you can react to a competitor’s announcement, a regulatory change or a seasonal spike at all. At three weeks, the schedule is set in advance and held to whatever the market then does.

Coverage. A company that can produce one piece of material a week produces roughly fifty a year. Spread across a website, a blog, four social channels, an email list and the library of material the sales team hands to prospects, that is thin coverage everywhere and strong coverage nowhere. The channels that get attention are the ones the loudest person in the room prefers, not the ones the audience uses.

Attention on the wrong leads. Without a shared view of fit, a sales team works whatever came in most recently. Time goes to whoever replied fastest rather than whoever is most likely to buy. This is expensive in a way that never appears on an invoice, because the cost is the deal that was not worked.

Institutional memory. When the agency changes account manager, or the marketing lead leaves, the accumulated understanding of what works for your business walks out with them. It lived in their head and in a shared drive, not in a system.

There is a fifth cost that is harder to name and increasingly the one that stops a purchase: nobody can say with confidence where the company’s information has ended up.

Where the data goes, and why that is now a board question

To write good marketing copy, whatever is doing the writing needs to know things about you that you would not put on your website. Your pricing logic. Which customers are unhappy. Which competitor you actually lose to. The positioning you tried last year and abandoned. The names in your pipeline.

That is exactly the material a marketing team pastes into a text generator to get a useful draft. The better the prompt, the more of your commercially sensitive material is in it.

Once that material leaves your network, several things become true at once. You no longer control who can read it, how long it is kept, or what it is used for. If you are asked in a security review, a procurement questionnaire, or a customer audit to describe where your customer data is processed, the honest answer involves a list of third parties and a set of terms you did not write. And if the vendor changes those terms, you find out afterwards.

For a company selling into regulated sectors — finance, healthcare, defence, public sector, critical infrastructure — this is not an abstraction. It is a question on a form that has to be answered before a contract is signed. For companies operating under data-residency requirements, it can be disqualifying.

PULSE answers that question differently. The system that does the reasoning and the writing runs where PULSE runs: a workstation for evaluation, and for daily use your own servers or cloud account, or Runink’s shared machines on Lite. There is no third-party database holding your working data. There is no outside AI service in the path when a whitepaper is drafted or a lead is researched.

This has a commercial consequence beyond compliance, and it is worth stating plainly: because PULSE is priced per seat, the licence does not rise with how much work you put through it. A team that produces ten pieces of material in a month and a team that produces two hundred pay for the same seats.

What PULSE does about it

PULSE is one application, operated by your own people, covering the full arc from finding out where you stand to keeping the customers you win. The work moves through four stages, and each stage feeds the next from a shared understanding of your business rather than from a re-briefing.

Diagnose. Point PULSE at your website and social profiles. It returns a scored read of your position — nine measures — a business diagnosis, and a prescription for each channel. This is the baseline everything else works from.

Attract. Market research surfaces the topics moving in your sector, the competitor channels worth watching, the influencers and the partnership targets worth approaching. Lead prospecting searches the public web for companies matching a niche you describe, and drafts the first contact for each one.

Create. A single brief becomes channel-specific posts, a whitepaper, a set of short videos, a presentation, a podcast, an infographic, a report or a full course. All of it carries the same understanding of your positioning, because all of it draws on the same diagnosis.

Retain. Customer profiles and journey maps show where each account is. Structured 30-day follow-up cycles keep every relationship moving, with an evolution score that makes a quiet account visible before it is a lost one. Metrics and a forward view of traffic show what is working.

Around all four stages sit two constants.

A review queue, and a channel switch that is off by default. Every draft — a LinkedIn post, a cold email, a call script, a whitepaper — arrives with an explicit status and an approve and a reject against it, so at any moment you can see what is waiting on you. The control that actually decides whether anything leaves is separate and blunter: every publishing channel is disarmed unless it has been explicitly armed, by a global switch narrowed by a per-channel one, with the resolved answer for each channel printed when the application starts. A channel that is off cannot publish, whatever status a piece carries. Page 18 sets out how the two controls work together.

You watch the work happen. Research and copy stream into the screen as they are produced, rather than appearing as a wall of text after a wait. If a direction is wrong, you see it while it is being taken.

Diagnose: knowing where you actually stand

Site Audit is the starting point and the single analysis hub. A URL goes in. One result comes back, on a tabbed page, carrying three things at once: a relevance score, a full business diagnosis, and market research findings for the same business.

The score is not a single number with no account of where it came from. It opens out into nine measures, each of which is itself made up of smaller checks you can read:

• Search Visibility — how findable you are in conventional search.
• AI Visibility — how well your material can be read, quoted and cited by the systems that increasingly answer questions on your customers’ behalf.
• Technical Foundation — the mechanics: whether search engines can read your pages and file them, how fast the pages load and stop moving about, whether the phone version shows what the desktop version does.
• Authority & Trust — the signals that establish credibility: who links to you, how often you are mentioned, whether experience and expertise are visible on the page.
• Acquisition — whether the page gives a visitor a reason and a way to act.
• Engagement — whether there is anything to stay for.
• Retention — whether there is a reason to come back.
• Monetization — whether the commercial path is present and readable.
• Demographics — who the material is actually addressed to.

Underneath those sit the specific checks. Whether the instructions your site gives to automated readers let them in at all, and whether the machine-readable summary of each page is there for them to use. Whether page titles and web addresses are structured so a reader can tell what a page is about before opening it. Whether your pages link to each other, and whether any page has been left with nothing pointing at it. Whether your words are in the page as delivered, or only appear once the browser has run a program — because a reader that does not run programs sees an empty page. How long the main content takes to appear, how quickly the page answers a tap or a click, whether the layout jumps about while it loads, and how long your server takes to say its first word. How deeply you cover your subject, how recently, and how many of the specific things, places and ideas your market talks about are addressed anywhere on the site. How many other sites link to you and how often you are mentioned by name. Whether there are sentences worth quoting and questions answered plainly. Whether there is something to click, a form to fill in, a way to buy, and a version in the reader’s own language.

Two design choices matter here. The weightings are settings rather than fixed rules, so they can be tuned to what your business actually cares about. And they are re-balanced around whichever sources of data are available for a given site, so a score always reflects what was measured rather than being dragged down by something that could not be.

The audit ends in ranked recommendations, and those recommendations can be applied from the same screen — not exported to a ticket that somebody opens in a fortnight.

Alongside the audit, a multichannel diagnostic reads your website, LinkedIn, Instagram and TikTok together and produces a written business analysis you can export as a PDF and put in front of a board.

Attract: finding the people worth talking to

Two things happen in this stage: understanding the market, and finding specific companies in it.

Market research. PULSE surfaces trending posts in your sector, discovers the hashtags and topics carrying attention, and suggests profiles worth following. It discovers competitor channels — where your competitors are publishing and what is working for them. It finds guest and appearance opportunities: podcasts, publications and events where your subject matter belongs. And it drafts partnership outreach for the organisations worth approaching.

Research draws on two places. It reads the live public web through a browser that extracts the readable substance of each page rather than the navigation and the banners. And it reads your own material — your documents, your audit results, your prior work — through a search that combines exact keyword matching with meaning-based matching, so a question phrased one way still finds material phrased another.

When you mark a result as useful or not, that judgement carries into the next round of research. The system gets more useful to you specifically, and that improvement stays with you.

Lead prospecting. Describe a niche — the kind of company, the market, the region — and PULSE searches the public web for organisations matching it, and returns them as prospects.

For each one, it drafts three pieces: a personalised cold email, a cold-call script, and a LinkedIn direct message. Each is written from what is known about that specific company and from the diagnosis of their business, not from a template with the company name dropped in.

Leads move through a pipeline you work as a board: new, contacted, qualified, then won or lost. Leads synchronise with HubSpot, so the sales team continues to work where they already work.

Voice. A voice sales agent handles conversation directly. In the console it is press-to-talk: you speak, it listens, the sales specialist responds, and the reply is spoken back — all on the machines PULSE runs on, with the same pair of speech engines every Runink console speaks and listens with: in Runink TIDE: Trusted Intelligence for Developer & Data Experience, and in Runink FACE: Fulfilment Autonomous Claims Engine, as much as here. For outbound calling, PULSE connects to a telephone exchange you host yourself, so calls run through infrastructure you own rather than a per-minute service.

WhatsApp. A lead can also be reached on WhatsApp, through your own Twilio connection. It is optional: without one configured, the channel is simply off, and nothing else in PULSE depends on it.

Create: one brief, every channel

The production stage takes a single input and produces the range of material a company needs at every stage, from a first glance to a signed deal.

One brief, opened out into every channel. The ring is the review; the schedule past it only reaches a channel that has been armed. One sheet on the left — a single brief, not a stack — opens out into a field of pieces whose shapes differ: long bars for written pieces, paired squares for pictures, tall narrow blocks for short video, broken runs for email. Every piece carries one status mark. A vertical line runs down the picture with a single ring in it, a tick, and only the pieces whose status mark is filled reach that ring; the ones that do not are drawn hollow and their strands stop short of the line. Past the ring the strands open out again into a dated grid of slots, some filled and some empty, which is the schedule of what publishes and when.

Channel content. Posts are written for the channel they are going to, in the shape and the tone that channel expects. The supported set covers LinkedIn, Instagram, TikTok, blog, email, X, YouTube, Facebook, whitepaper, book outline, webinar, course and podcast.

Studio. Six long-form generators sit in Studio:

• Whitepapers — long, structured documents for people who have only just heard of you, and for the sales team to hand over.
• Podcasts — episode material, produced as audio.
• Shorts — short-form video.
• Presentations — decks.
• Infographics — visual explanations of a position or a dataset.
• Reports — periodic written analysis.

Alongside them, Content Strategy produces a channel-by-channel plan, and the strategy planner produces a dated 30-day schedule of work — what publishes, on which channel, in which week.

Courses. Teaching sequences for customer education and for the programmes that establish your expertise in a subject. These are the pieces that turn a customer into someone who recommends you, and they are usually the first thing a stretched team stops producing.

Creatives. Pictures are produced on the same machines and stored against the campaign they belong to. Short videos are assembled frame by frame from pictures produced the same way.

The approval stages. Every piece of content has one stated status: draft, waiting for review, approved, rejected, published, archived. Those stages are not decoration. They are how you can tell, at any moment, what is waiting on you, what has been rejected and why, and what actually went out.

Campaigns hold the work together, carry a budget, and move through draft, active, paused, completed and archived.

Publishing schedules approved posts and shows what is queued.

Generation streams as it happens. You watch a whitepaper take shape rather than waiting for one to appear, which means you catch a wrong angle in the second paragraph instead of on page nine.

Retain: keeping what you fought to win

Winning customers gets the attention and keeping them pays the bills. PULSE treats the second as a proper part of the work rather than an afterthought bolted onto a customer-record system.

Customer 360 generates a living profile for a customer or a segment: what they care about, what they respond to, where they are in their relationship with you. From that profile it maps a journey — the sequence of touches that account has had and the one that should come next.

Persona Profile holds the detail behind each persona, so the people writing to that audience are writing to a specific person rather than a category.

Intelligence Console is where audience analysis is worked: the screen for understanding who is actually paying attention.

Follow-up runs structured 30-day cycles. Each cycle has a weekly planner, task states you move, saved check-ins that record what was discussed, and an evolution score with its history. The score is the useful part: it makes a cooling relationship visible as a trend rather than as a surprise at renewal.

Influencers are listed, saved and tracked as a working roster rather than a spreadsheet somebody maintains.

Metrics and Insights. Metrics reports impressions, clicks, conversions and return, with a day-by-day view of how those move and a view of how they split across channels, and a written reading of what the numbers mean. Analytics pulls in your Google Analytics figures and produces both a trend and a forward view, so a seasonal peak is visible while there is still time to prepare for it.

Content Gaps compares what you publish against what your competitors publish, and names the subjects your market is asking about that you have not addressed. Impact Analytics shows which of your material is doing the work.

The console, screen by screen

The console is organised around what you are trying to do, not around how the software is put together. Five groups.

Overview — the hub, the Marketing Console, Site Audit, Insights, Impact Analytics and Content Gaps. This is where you go to find out how things stand.

Audience — Influencers, Customer 360, Persona Profile, Intelligence Console, Leads and Follow-up. This is where you go to work on people: who to watch, who to partner with, who to sell to, and how to keep them.

Content — Campaigns, Creatives, Approval and Schedule, plus Courses. This is where work is framed, reviewed and queued.

Studio — Content Strategy, Strategy Roadmap, Whitepapers, Podcasts, Shorts, Presentations, Infographics and Reports. This is where the material itself is produced.

System — Integrations, AI Team, Settings and Billing.

Integrations connects PULSE to the accounts you already run: LinkedIn, Instagram, TikTok, Facebook, X, YouTube, Google Ads, Google Analytics, HubSpot, Mailchimp, Spotify and GitHub. Each connection is configured in the console and tested before you rely on it, and can be disconnected from the same screen.

AI Team is a roster of named specialists, each with a defined role, that the in-console assistant hands work to. You can add a specialist, see the roster, and activate or deactivate any of them — which means the division of labour is something you configure rather than something you infer.

Billing shows which plan you are on and how many people are licensed.

The console runs in a browser, as an Android application from Google Play, and as a Linux desktop application. It keeps a copy of results on the machine itself, so a review session in a meeting room with a poor connection keeps working.

The interface is available in English, Spanish, French and Portuguese, with a language picker in Settings.

What it is built on, and why that matters commercially

PULSE is built to run on the machines your plan names, with no outside AI service. That has five commercial consequences.

It runs where you put it. The same application runs directly on a workstation for evaluation, and for daily use on one machine or several in your own estate, in your own cloud account, or on Runink’s shared machines on Lite. Evaluating it does not require moving anything anywhere. Production does not require a different product.

Your working data stays where PULSE runs. It is kept in a database that lives inside the application itself, copied continuously into file storage on those same machines, and restored from that copy when the application starts. There is no third-party database holding your customer records, your positioning or your pipeline.

The reasoning stays there too. The system that writes the copy and performs the analysis runs on the same machines. Your positioning, your customer lists and your competitive material are processed there, never by an outside AI service. When a procurement questionnaire asks where your data is processed, you can name the machines.

The licence does not rise with how much you use it. PULSE is priced per seat, so producing more material does not produce a larger licence bill for producing it. A team that runs one campaign and a team that runs twenty pay for the same seats. This changes what is worth attempting: trying a second angle on a campaign adds nothing to the licence, so second angles get tried.

It runs on ordinary hardware. The reasoning runs on ordinary processors where no specialist graphics chip is present, and production installations run on ordinary processors alone. This is not a specialist installation requiring specialist procurement.

Controls. The console and the server behind it answer at the same web address, reached over an encrypted connection whose certificates Runink’s own platform issues rather than an outside authority. People sign in with an email address and a password, or — where a deployment is set up that way — only with a Google account, and only if that address is on a named list. What a person can see and do follows from three things at once: the role they hold, the attributes recorded against them, and what they are related to, such as the accounts they own. Personal details in a request are handled by the platform’s own protection layer as the request passes through. What the specialists are allowed to be asked is checked against fixed written rules — the same input always gets the same answer, and each specialist has its own set — and those rules are mapped to the OWASP Top 10 for large language model applications, the published industry list of the ten commonest ways systems of this kind are attacked. And no channel carries anything out of the building until somebody arms it: drafts wait in a review queue with a status against them, and a channel nobody has armed sends nothing, whatever status a draft carries.

Runink runs its own marketing on PULSE

The most direct evidence available for a marketing product is whether its vendor uses it. What follows is our own use of PULSE, offered as that rather than as a customer’s result.

PULSE does the content work behind runink.org: the site and search audits, the writing, and the social and lead-generation work, run by PULSE’s own specialists on Runink’s own hardware against the model that runs there. It is worth being exact about the boundary, because the obvious reading of that sentence is wrong: runink.org is a static site kept in git and published to GitHub Pages by a build job. PULSE produces the material; it does not host the result, and we are not going to describe a public marketing site as something it keeps for us.

The observable state of that presence, read from the build rather than estimated: the use-case pages, the pricing page and the homepage exist in all four languages; the long-form articles, the papers and the industry pages are in English, with one article also in Portuguese and one industry page also in French.

One thing follows from that: the nine audit measures are the scores Runink watches on its own pages, which is the reason they are nine rather than a rounder number.

Who owns it, who sponsors it, and who signs it off

Three different people are involved in a purchase like this one, and they are almost never the same person. Naming them separately is not organisational theory. It is the difference between a conversation that progresses and one that goes round twice and stops.

The person who feels it is the marketing lead. They can describe the problem in one breath — three weeks to turn a brief around, a content plan abandoned in March, an agency that costs more than a headcount — and they usually cannot sign for the amount involved.

The person who sponsors it carries the budget. In a company of this size that is the chief executive or the commercial director as often as it is a chief marketing officer, because marketing spend and sales pipeline are the same conversation at two hundred people.

The person who signs it off is the one most often left until last, and that is the mistake. Where company material is processed is now a question with a named owner — a security lead, a data-protection officer, sometimes the finance director wearing that hat in a smaller company. They can stop the purchase and cannot start it.

The pattern worth naming

In four of those six rows the sponsor is not the marketing lead. The budget sits with whoever owns revenue, and the case has to reach them in revenue’s units rather than in production units. “We could publish four times as much” is a marketing sentence. “The response time to a competitor’s announcement drops from three weeks to a day, and here is what that is worth” is the same fact in the sponsor’s language.

And in three of the six rows the person who signs it off is asking about data, not about marketing. That conversation has one answer, it is on page 6 and page 13, and it is worth having early. Bringing a security lead in at the end converts a short conversation into a long one held under deadline pressure.

Who should be in the first meeting

The marketing lead, the person who owns the revenue number, and the security lead. Three people, one sitting. If the security lead cannot come, send them page 6 first rather than last.

Who PULSE is for

PULSE fits a recognisable set of situations. It fits less well where marketing is already a large, specialised department with its own production capacity — that is a different problem.

Companies between twenty and five hundred people with one to five people in marketing. This is the core case. There is enough to say and not enough hands to say it. The constraint is production capacity, and that is exactly the constraint PULSE removes.

Companies currently paying an agency retainer for content and social. The relevant comparison is not quality against a good agency’s best work. It is the arc: brief in the morning, draft the same day, published the same week, with the accumulated understanding staying in your system rather than in the agency’s. For companies where the retainer buys volume of work rather than a specific creative skill, that arc is the argument.

Companies selling into regulated sectors. If your customers ask where their data is processed — finance, healthcare, defence, public sector, critical infrastructure — the processing-location question is not a preference. Running the work on your own servers or in your own cloud account, on the Enterprise or Dedicated licence, gives you an answer you can put in writing.

Companies operating under data-residency requirements. Where information must remain within a jurisdiction, an application that runs on servers you site, on the Enterprise licence, is the straightforward way to satisfy that.

Companies with a seasonal business. Where a small number of periods carry most of the year’s revenue, the forward view from your own analytics data and the ability to produce a full campaign in days rather than weeks change what preparation is possible.

Companies whose sales team needs specific outreach at volume. Where the pipeline is built by outbound and the difference between a reply and silence is whether the message addresses that particular company’s situation, per-lead drafting drawn from a diagnosis of that lead’s business is directly commercial.

Companies selling to more than one language market. The production cost of a second and third language is the reason most companies address one. When production is not the constraint, that reasoning changes.

The questions a buyer asks

These are the questions that come up, in roughly the order they come up in. Each answer describes how the product behaves rather than offering an assurance, because a description can be checked in an afternoon on a workstation of your own and an assurance cannot.

Does it need training on our business first?

No, and there is nothing for you to label or upload in advance.

The writing and the analysis are done by a model held as weight files on the machine PULSE runs on. Those files are the same on your first day and your five hundredth, and on your own servers you can compare them and confirm it.

What makes the output yours is not training but reading at the moment of the question. Your site audit, your positioning documents, your prior material and your customer records are indexed on that machine; when a brief is written, the relevant passages are retrieved and placed into the question, with the source of each travelling alongside. That is the mechanism behind the claim on page 5 that generic inputs produce generic copy: the inputs here are specific because they are your own records, read at the moment of writing.

It is also why deleting a document removes its influence entirely. Nothing is left behind in a set of weights.

How does it get better over time?

Four ways, none of which changes the model.

Your judgements carry forward. When you mark a research result as useful or not, that judgement is kept and shapes the next round. Approvals and rejections in the content queue are recorded the same way. The improvement accumulates against your account rather than walking out with whoever made it, which is exactly what does not happen when the knowledge lives with an agency account manager.

The material it can reach grows. Every audit run, every piece produced and every document indexed widens what the next brief can be written from.

The scoring is yours to tune. The nine measures and the checks beneath them carry weightings held as settings rather than fixed rules, so the balance can be set to what your business actually competes on. They are also re-balanced around whichever data sources were available for a given site, so a score always reflects what was measured.

The quantitative work is chosen by testing. Where a forward view is produced from your analytics history, candidate methods are fitted to the earlier part of that history and asked to predict the part held back; the one that missed by less is used, and how far each missed travels with the answer.

What happens when it cannot work something out?

A source that could not be read is set aside, not scored zero. If a measure depends on data that was unavailable, that measure is skipped and the remaining ones are re-weighted around it. Your score is never quietly depressed by something that was not measured, which is the single commonest way audit tools mislead.

A source that failed is reported as failed, distinctly from one that was simply not applicable, because the two need different responses from you.

A disarmed channel publishes nothing, and an unset setting is never consent. Every draft carries one stated status — draft, waiting for review, approved, rejected, published, archived — so at any moment you can see what is waiting on you and what actually went out. A channel that has been switched off stays off even for approved material: an approval is consent to the content, and switching a channel off is a statement about the channel. Content held back that way is recorded as held back — its own terminal status, distinct both from published and from failed — because “the upload failed” sends somebody to check tokens and quota for something that was never attempted, and leaving the row merely due would load a spring behind the kill switch that fires the whole backlog the moment anybody flips it back. Each such row also carries a provenance for its own explanation: recorded, nothing-to-explain, or missing — so a held-back row with no stated reason reads as a defect rather than as silence. And where a publishing setting is absent or unrecognised it resolves to the most restrictive value available, under a test whose stated rule is that absence of configuration is not consent to publish.

Arming a channel is the decision you are making. Where a channel is armed, the gate on publication is the channel switch itself: nothing publishes on a channel you have not armed, and the switch for each channel is printed when the application starts.

The reasoning is watchable while it happens. Research and copy stream onto the screen as they are produced. A wrong angle is caught in the second paragraph rather than on page nine, which is a more practical form of quality control than any confidence score.

How does it work with the systems we already run?

PULSE connects to the accounts you already have — your social channels, your advertising and analytics accounts, your customer-record system, your mailing platform — configured in the console and tested there before you rely on any of them, and disconnected from the same screen.

Leads synchronise into your customer-record system, so the sales team keeps working where they already work rather than being asked to move. That direction matters: PULSE is not trying to become your system of record, and a marketing tool that asks to be one is asking for a migration you did not budget for.

Web research reads the public web through a browser the platform drives itself, extracting the readable substance of a page rather than the navigation and the banners.

For outbound calling, PULSE connects to a telephone exchange you host yourself, so calls run over infrastructure you own rather than a service billed by the minute.

Where does our material live, and is it used to train anything?

It lives where PULSE runs. Working data is kept in a database inside the application itself, on the machines your plan names, copied continuously into file storage on those same machines and restored from that copy when the application starts. There is no third-party database holding your positioning, your customer records or your pipeline.

It is not used to train anything. No material is sent out to be trained on, and there is no account with an outside model provider for it to be sent to. That is enforced mechanically as well as stated: a published list of outside model libraries and their network addresses is checked against the software before any change is accepted, and a change that introduced one would be refused rather than reviewed.

Confidential material is protected in transit and in the record. The console and the server behind it answer over an encrypted connection whose certificates the platform issues itself. Sign-in is by company identity or by password, and where a deployment is configured for company identity only, an address that is not on the named list cannot sign in — a list that has been configured but left empty admits nobody rather than everybody. What a person can see follows from the role they hold, the attributes recorded against them, and what they are related to. Personal details in a request are handled by the platform’s own protection layer as the request passes through.

What the specialists may be asked is checked against fixed written rules — the same input always produces the same answer, each specialist has its own set, and the rules are mapped to the published industry list of the ten commonest ways systems of this kind are attacked.

Does it need special hardware?

No. The reasoning runs on ordinary processors where no specialist graphics chip is present, and production installations run on ordinary processors alone. An evaluation wants a well-specified developer workstation — a reasonable amount of memory, a decent number of processor cores and around thirty gigabytes of free storage — rather than a server purchase.

This is not a specialist installation requiring specialist procurement, and the practical consequence is that an evaluation can start this week rather than after a hardware conversation.

How long does it take, and what do you need from us?

Steps one and two cost an afternoon each and move no data anywhere, which usually means an evaluation can begin without a procurement conversation about data handling.

What we need from you: a licence that fits where the work should run, and on Dedicated or Enterprise, room in your own cloud account or on your own servers; your existing account credentials for the channels you want connected; a sign-in arrangement and the list of addresses permitted to use it; and one named person who will own the approval queue. What we do not need: a data migration, a change to your customer-record system, or a period of setup before anything useful comes out.

Can a small marketing team run this without technical help?

Yes, and the core case on page 16 assumes exactly that: one to five people in marketing.

The operator is a marketing person, not an engineer. Connections are configured in the console. Briefs are written in ordinary language. The roster of specialists the assistant hands work to is something you configure from a screen rather than something you infer, so the division of labour is deliberate.

The honest framing, repeated from page 19, is that the technical part is the small part. What determines whether adoption succeeds is a change in working habit — the team’s centre of gravity moving from producing material to deciding about it — and that is a management question rather than a technical one.

Why should we trust what it produces?

You should not, on trust. You should run the two-step test on page 23 and look at the output with your own name notionally on it, because that is the only judgement that matters for material you will publish.

What the product does to earn that look is five things. Every audit score opens out into the measures beneath it and then into the individual checks, so a number is always traceable to what was actually examined. Generation streams as it happens, so a wrong direction is visible while it is being taken. Every piece carries an explicit status and an explicit approve or reject, and every publishing channel is disarmed until somebody arms it — with how those two controls work together set out on page 18. The weightings behind the scoring are settings you can read, which means you can disagree with them specifically rather than in general. And a diagnostic in progress carries a reasoning trace — the real actions it took and the sources behind them, alongside the streaming text — so a finding is something you can open up rather than something you take on faith.

What adopting PULSE involves

Adoption has three phases. The honest framing is that the technical part is the small part.

Phase one: evaluation, on a workstation. PULSE runs natively on a single machine. That machine wants a reasonable amount of memory, a decent number of processor cores and around thirty gigabytes of free storage — a well-specified developer workstation, not a server purchase. No data moves anywhere to run an evaluation, which usually means an evaluation can begin without a procurement conversation about data handling.

The first useful output is available in the first session: run Site Audit against your own website and read the nine scores and the ranked recommendations. That result is worth having whether or not you go further.

Phase two: putting it into use. For daily use, PULSE runs where your licence puts it: on Runink’s shared machines on Lite, in your own cloud account on Dedicated, or on your own servers on Enterprise — one of them or several working together. The Runink platform does the installing. Working data is kept on those machines and copied into file storage there. The console is reached over an encrypted connection at a web address you choose. The reasoning runs on ordinary processors, so this does not mean buying specialist graphics hardware.

Then connect the accounts you already run: your social channels, Google Ads and Analytics, HubSpot, Mailchimp and the rest. Each connection is configured and tested in the console before you rely on it.

Phase three: working practice. This is the part that determines whether adoption succeeds, and it is a change in habit rather than in technology.

The team’s centre of gravity moves from producing material to deciding about it. That is a different working day. The Approval queue becomes the main surface — the place where the lead spends time, exercising judgement on complete drafts rather than assembling them.

Three habits make the difference. Give feedback on research results, because that judgement carries forward. Keep the 30-day cycle running rather than treating it as a launch exercise, because the evolution score only tells you something when it has a history. And configure the AI Team roster to match how your team actually divides work, so the division of labour is deliberate.

A reasonable expectation: a first audit and a first anchor piece within the first week. A first full 30-day cycle completed within the first month. A settled working rhythm by the end of the second.

Commercial model

Licensing is per seat — one seat for each person who uses the software — and the seat covers the work that person puts through it. PULSE is sold on its own subscription, apart from Runink’s other products. Current pricing is published on runink.org. Three levels are offered, told apart by where the work runs.

Lite licence — for teams of one to nine seats, running on Runink’s shared machines, where your work takes its turn alongside other customers’. Each seat gets 10 tasks a day at full speed; after that, tasks keep running at a slower pace until midnight in the person’s own time zone, when the count starts again. It is available on monthly or annual commitment. This is the entry point for a small marketing function, and the monthly option means an evaluation does not require an annual contract.

Dedicated licence — for ten seats and above, on annual commitment, running on runners in your own cloud account (a Google Cloud project, or a Databricks or Snowflake account). Your cloud bills you for that compute directly.

Enterprise — quoted with you. It includes deployments you host yourself on your own premises, including those with no connection to the outside world, with a fuller record of who did what and terms set to your requirements. This is the level for regulated industries and for organisations whose deployment must sit entirely inside their own boundary.

What you pay is visible in the console rather than arriving as a surprise. The Billing screen shows which plan you are on and how many people are licensed.

The economics are what changes behaviour. The licence is a function of how many people use it, not of how much work they put through it. Under a per-word or per-generation arrangement, every additional draft is a decision with a price attached, and teams ration accordingly — one angle per campaign, one language, one format. When an additional draft does not change the seat price, the rationing stops, and the second angle, the third language and the fourth format become ordinary.

What it is worth, computed on your own numbers

There is no return figure here, because the figure depends on what you currently pay for material, how much of it you produce, how long you take to respond to something, and what a deal is worth to you. All four are yours. What follows is the arithmetic instead: every input named, every one read from an invoice, a calendar or a system you already have, and no value anywhere in it. Run it on your figures and the result is yours.

Be exact about what actually moves

PULSE does not make your market larger and it does not make a bad offer good. What it moves is the cost and the elapsed time of producing a piece of material — and, through those two, how much you produce, how quickly you can respond, and how many things it becomes reasonable to try. Everything below measures those, and a calculation that credits the software with more than that falls apart the first time a finance director reads it.

Six inputs, and where each one is read

One — what a piece of material costs you today. Take the last twelve months. Add the agency retainer and project fees, the internal salary cost of the hours spent briefing, reviewing and chasing, and any per-seat tool costs for writing, design and scheduling. Divide by the number of pieces actually published. That quotient is your cost per piece, and it is almost always higher than anybody in the building expects, because the denominator is what was published rather than what was commissioned.

Two — what you commissioned but did not publish. Count it. Briefs that died in revision, pieces abandoned when the moment passed. This is pure cost with no output against it and it belongs in input one’s numerator, which is why input one is calculated the way it is.

Three — elapsed time from decision to publication. Take five recent pieces. For each, the date somebody decided to say the thing and the date it went out. Take the median, not the mean. This is the number that decides whether you can respond to a competitor, a regulatory change or a seasonal spike at all — and a capability you do not have does not appear as a cost anywhere in your accounts.

Four — coverage against the channels you have decided matter. List the channels and formats you have agreed are worth being present in. For each, how many pieces went out in the last twelve months. The rows near zero are the decision that production capacity made on your behalf, and it is worth seeing it written down as a decision, because that is what it was.

Five — how much rationing is happening. For your last three campaigns: how many angles were considered, and how many were actually produced. How many languages the market has, and how many you addressed. The ratio is what the current cost per piece is buying you, expressed as things not attempted.

Six — what attention on the wrong leads costs. Harder, and worth attempting. From your customer-record system: the share of sales time spent on leads that never qualified, and the value of the deals worked versus the deals that were in the pipeline but not worked. This is the input the sponsor cares about most and the one marketing least often produces.

How they combine

Sum the four, subtract the fifth, divide the annual result by the monthly cost, and you have a payback period in months. This paper does not state one, because every term belongs to you.

The caution on the second line, stated plainly

More material is not automatically worth more. Doubling output doubles value only if the additional pieces reach somebody and do something. If your constraint was never production — if you already publish everything you have to say — then the second line is zero and PULSE is solving a problem you do not have. Page 16 says the same thing from the other direction.

Price the additional pieces at what the existing pieces measurably produce, read from your own analytics rather than from a benchmark. If you cannot measure what an existing piece produces, set the second line to zero and make the case on the other three. It usually still holds, and a case that holds without its weakest term is the stronger one to take into a room.

Four more ways the answer comes out wrong

Counting the retainer as a saving while still paying it. If the agency relationship continues for the work it is genuinely better at, only the replaced portion is a saving. Count that portion, not the invoice.

Forgetting the review time that replaces the production time. The team’s day changes rather than empties. Reviewing complete drafts takes real hours. Measure them in your first month and put them on the cost side.

Comparing across a period when something else changed. A quarter that also carried a rebrand, a new product or a new hire is not a clean comparison. Choose a period where this is the change.

Attributing pipeline to marketing that sales would have found anyway. The fourth line should count deals that were genuinely not being worked, not deals that were worked later.

Record the baseline before you start

The commonest reason a marketing function cannot state what something returned is that nobody wrote down the starting position while it was still true.

Five numbers, in the first week: cost per published piece, count published in the last twelve months, median decision-to-publication days, the coverage table by channel, and the current share of sales time spent on unqualified leads.

All five become unrecoverable once the working rhythm changes, because the thing that would tell you is now the thing that changed. Twenty minutes in week one is the difference between a defensible figure at the end of the quarter and an argument about whether it felt better.

The next step

The most useful first move costs you an afternoon and moves no data anywhere.

Run an audit of your own site. PULSE runs on a single workstation. Point Site Audit at your own website and read what comes back: the nine scores, the ranked recommendations, the business diagnosis, the market research for your sector. That is a concrete document about your own business, produced in a session, and it is worth having regardless of what you decide next.

Then run one anchor piece end to end. Take the largest gap the audit names, brief a whitepaper on it in Studio, watch it stream in, edit it, approve it, and fan it out into the posts, the short videos and the email sequence. That exercise tells you the two things you actually need to know: whether the output is good enough for your name to go on it, and whether the working rhythm suits your team.

Then decide. If it does, choosing where it runs and connecting your existing accounts is the next conversation, and it is a short one.

To arrange an evaluation, a walkthrough of the console, or a conversation about a deployment on your own servers or cloud account:

paes@runink.org

runink.org

Runink PULSE — Prescriptive Unified Lead & Social Engine.

One application. No outside AI service. Every channel off until you turn it on.