B2B marketing automation: workflows that generate pipeline

A B2B company that installs HubSpot takes on average a few weeks to build its first workflows, and several months or even years to realize they are useless. The emails go out. Open and click rates are fine. The monthly report is green. And the pipeline does not move by a single euro.

The problem is almost never the tool. It lies in what we ask automation to do. Most of the setups we audit were designed to send messages, not to move opportunities forward.
These are two different objectives, which produce two different architectures, and only one of them shows up in the forecast.

Marketing automation in B2B is not about sending emails

Let us start with the definition that circulates: marketing automation means automating repetitive marketing actions, mainly sending communications, based on contacts' behavior. That is accurate. It also explains a good share of useless setups.

Because if you start from “automating sends”, you build a sending machine. You get clean, segmented, personalized sequences that run in the background and that nobody can say what they have brought in.

The definition that produces better setups is narrower: marketing automation is the mechanism that moves a contact from one stage of maturity to the next, without involving a sales rep, up to the point where involving one becomes profitable. Everything else (email, SMS, internal notification, property update) is only a means.

This wording has an immediate consequence: a workflow that moves nobody forward has no reason to exist, even if it shows a 75% open rate. And a workflow whose exit is not defined cannot move anyone forward, since it has no destination.

In B2B, three specific features make this discipline more demanding than in B2C. The cycle is long, so nurturing is counted in months, not days. The decision is collective, so you work an account through several contacts who have neither the same role nor the same level of maturity. And volume is low: a few hundred target accounts, sometimes fewer. A mass conversion logic makes no sense on a base of 800 accounts. Precision does.

This last feature is the one most often ignored. Recipes designed for e-commerce volumes are imported into databases where every account counts.

The four families of workflows that produce pipeline

In the field, most of the contribution to pipeline comes from four families. They are not glamorous, and that is precisely why they are underused.

Maturation nurturing

This is the best-known family and the most poorly built. The principle: a contact shows interest (download, visit to a product page, webinar registration) but is not ready to talk to a sales rep. The workflow maintains the relationship by providing, at spaced intervals, elements that address the objections of their actual stage.

The classic mistake is to send the same five-email sequence to everyone, whatever content triggered the entry. A contact who downloads a tool comparison is not in the same place as a contact reading a definition article. The first is comparing, the second is discovering. Sending them the same thing means losing the first and overwhelming the second.

The principle to remember here: one sequence per detected intention, not one sequence per persona.

Qualification and routing

This family sends no external emails, and it is often the most profitable. The workflow observes signals, updates a score, moves a lifecycle stage when the criteria are met, assigns an owner, creates a task, notifies the relevant sales rep.

A concrete example of what this looks like in HubSpot: a contact from a target account visits the pricing page three times in one week. The workflow increments the engagement score, checks that the fit score exceeds the ICP threshold, moves the contact to MQL, assigns the account owner, creates a “priority call” task due within 24 hours, and posts a notification in the team's channel. No email to the prospect. A lot of pipeline.

The condition, of course, is that the scoring holds up. The principle that matters: never merge fit and engagement into a single score. A very curious intern and a discreet purchasing director would then get the same score, and your sales reps learn to ignore your alerts.

Reactivation

Your database contains contacts who came in eighteen months ago, never requalified, never contacted again. Some have changed jobs. Some have a project today. You know nothing about it.

A reactivation workflow reaches out to them on a new signal (a product change, content that addresses their topic, an industry event) and above all observes the response to sort them: re-engaged, permanently inactive, or to be removed from the database. Sorting is worth as much as re-engagement. A clean database costs less and lies less in reports.

This is the family teams postpone the most, because it requires accepting that the result will be partly negative. Let us say it plainly: a reactivation campaign that unsubscribes 30% of its target is a success, not a failure.

Expansion and retention

The last family is almost always missing from the setups we audit, even though it has the best effort-to-revenue ratio. Your existing customers are accounts whose context, usage and contacts you know. Onboarding workflows, upsell signal detection, alerts on falling usage or approaching renewal work on far more qualified material than any inbound lead.

An automation setup that stops at signature leaves half of the revenue potential untouched.

A concrete example in insurance 

This example illustrates the second family better than any diagram. We support an insurer specializing in professional risks, whose advisors receive far more quote requests than they can call back. The starting observation was harsh: a significant share of prospects was never contacted again, for lack of time, and went elsewhere to take out a policy.

The setup we built does not try to replace the advisors. It works precisely on the segment they do not reach: quote requests with no sales activity, and prospects who cannot be reached after several call attempts. A short sequence, with a direct tone, sent from a named address, with a subject line that repeats the initial request.

The channel does not behave like an emailing channel but like a sales conversation: among enrolled contacts, direct replies outnumber clicks. Prospects reply to the message as they would reply to a sales rep. Nobody had anticipated this when designing the scenario, and it changed the way the following emails were written.

Finally, measurement: about one enrolled contact in five becomes a customer if they opened at least one email, one in twelve if only those who clicked or replied are counted. The gap between these two figures is the attribution's margin of uncertainty, and it should be accepted rather than eliminated.

Triggers, exit conditions, safeguards: the mechanics that make the difference

Two setups can contain the same emails and produce opposite results. The difference lies in three settings that teams deal with last, when they should be decided first.

The trigger. A workflow triggered by a static attribute (industry, size, job function) segments. A workflow triggered by a dated behavior (visited, downloaded, has not opened since) reacts. The second produces pipeline, the first produces lists. The question to ask for each scenario: what just happened for this contact, and why is now a better moment than yesterday?

The exit condition. This is the most neglected and most costly setting. A workflow must know when it has succeeded, and remove the contact immediately. Otherwise you get the situation every HubSpot practitioner has already seen: a prospect who has just signed keeps receiving the “discover our solution” sequence, and the sales rep hears about it from the customer. Every workflow needs at least three written exits: success (the contact has progressed), failure (no reaction after N steps, remove and re-sort), and exclusion (they have become a customer, or an opportunity is open on their account).

Safeguards. A global frequency limit per contact, independent of workflows, to prevent three simultaneous scenarios from producing eleven emails in one week. A permanent exclusion list (sensitive customers, accounts in negotiation, contacts flagged by sales). And a precedence rule when two scenarios claim the same contact.

In this screenshot, the exit is set on a workflow goal: as soon as the contact moves to “Opportunity” or “Customer”, they are unenrolled. Three clicks, and the “discover our solution” scenario no longer chases a customer who has just signed.

A remark in passing, because it comes up in almost all our audits: the lead status property is the junction between automation and actual sales work, and it is very often the least maintained. A workflow that triggers on a lead status nobody updates will never trigger, and you will look for the bug elsewhere for weeks.

Measuring: what counts and what does not

Automation reporting is where useless setups survive the longest, because they are measured on the right figures of the wrong objective.

Email metrics (deliverability, opens, clicks) are for technical diagnosis. They say nothing about the contribution to revenue. A falling open rate signals a deliverability problem or a subject line promise problem, and it should be looked at. It neither justifies nor condemns a scenario.

Three indicators tell the truth:

• Stage progression rate. Among contacts who entered the workflow, what share moved to the next stage within the expected window, compared with eligible contacts who stayed outside? This measures the scenario's own effect.
• Time to progress. At an equal rate, a scenario that moves a contact from lead to MQL in three weeks is worth more than one that takes four months. Time is a pipeline variable.
• Influenced pipeline. The amount of opportunities opened on contacts who went through the scenario, over the reference period. Imperfect, disputed, and by far the most useful for deciding on a budget.

We take a position on this point: seeking perfect attribution for B2B marketing automation costs more than it brings. Between buying committees, long cycles and unidentified contacts, no model will capture reality. A simple influence indicator, understood and accepted by marketing and sales, makes it possible to decide. A sophisticated model that only its author can read is mainly useful for justifying oneself.

And the operational corollary, which is also the hardest decision to make: a scenario with no measurable effect after a full sales cycle must be stopped. Not improved, not reworked a third time. Stopped. Workflow debt is technical debt like any other, and the portals we take over often contain dozens of such workflows.

Scoping before automating

Automation amplifies what already exists. This sentence seems obvious, yet it is the primary cause of failed setups, because people read it without drawing the order of the work from it.

Three prerequisites, in this order.

A written ICP. Without an operational definition of the accounts you want to win, fit scoring has no criteria, so routing has no threshold, so sales alerts go out at random. An ICP that excludes nobody is useless.

Lifecycle stages with written criteria applied by workflow. If the move from lead to MQL is done by hand, by judgment, by different people from one week to the next, no progression indicator can be interpreted. The criteria must be written, then applied automatically.

Usable data. Duplicates, empty fields, free values where there should be closed lists: each data entry defect becomes a targeting error at scale. A workflow that segments on an “industry” field filled in as free text segments on typos. This is the work covered by our SalesOps & Data offer, and it is the one that pays off the most, without ever appearing in the slides.

A fourth, non-technical prerequisite is missing, and it often decides the fate of the setup: an explicit agreement with sales reps on what they commit to doing when an alert arrives. A perfect routing workflow that pours out tasks nobody handles produces exactly zero pipeline, and destroys marketing's credibility for the following two years along the way.

AI in workflows: calculated properties and account knowledge

Once the foundation is in place, AI opens up a capability that classic automation did not have. A traditional workflow branches on what the contact has done with you. With an AI layer, it also branches on what the account is, what it is looking for, and what has just changed on its side. A change of nature, not a productivity gain.

This is currently the most underused part of the setups we take over. Teams have activated assisted writing features, which save time without changing anything in the pipeline, and left aside the two uses that really move the needle.

Calculated properties: let AI do the synthesis, keep the decision

Your workflows personalize poorly because they have almost nothing reliable to personalize on. An “industry” field filled in on a form, a job function declared by the contact themselves, and that is about it. Yet customer knowledge exists: it is in call transcripts, email threads, tickets, sales notes, browsing history. In a form no workflow can read.

The solution is to have AI produce structured properties from this unstructured material, then to branch workflows on these properties rather than on free text. The properties we set up most often: the dominant topic of interest deduced from pages viewed and content downloaded, the expressed need summarized in one sentence from conversations, the industry normalized from the company name, the project's maturity stage, the actual role in the buying committee.

The effect on personalization is direct. You move from one sequence per persona, maintained by hand and never up to date, to a single sequence whose content blocks vary according to the calculated topic of interest. Fewer scenarios to maintain, more relevance per message. The same mechanism serves routing: a one-sentence summary of the need in the sales rep's task is worth more than twelve fields they will not read.

Three rules we apply without exception:

• Separate the calculated field from the human field. A property written by AI never overwrites the field sales reps edit by hand. Two separate properties, otherwise no audit is possible and nobody knows who wrote what anymore.
• Date and recalculate. Every calculated property carries a calculation date and a refresh rule. A field calculated only once when the contact was created becomes a lie after six months.
• Plan the default path. Every branch on a calculated property has a branch for empty or doubtful values. They will come.

And a position that sparks debate in workshops: an AI-calculated property that no workflow branches on and no sales rep relies on is not a capability, it is a line on an invoice. The question to ask before creating one is not “can AI produce it” but “which decision changes when it is filled in”.

External enrichment: the account knowledge you will never have internally

Your CRM knows what your contacts do with you. It knows nothing about what they are in their market. This is exactly the data the fit score lacks, and no amount of behavioral tracking will replace it.

External enrichment fills this gap: actual headcount and trajectory, technologies used, company news, intent signals, and above all people's job moves. HubSpot addresses this layer with Breeze, its built-in AI, natively connected to CRM objects, and the ecosystem includes specialized sources that can be connected by API when the need is industry-specific. We detailed the mechanics and pitfalls in an article dedicated to CRM data enrichment with Breeze.

The most profitable trigger in this family is also the most neglected: a contact changing jobs. Two scenarios open at the same time, and most teams play neither. The account loses its internal champion and needs to be secured before renewal. And this champion arrives at another account, with a preference for your solution already established. One signal, two pipelines.

Cost discipline is required, because enrichment is billed per credit or per record. You enrich on entry into a segment that matters, not the entire database. Enriching 200,000 contacts to feed a scenario that will reach 400 is spending with no return, and it is a mistake we regularly see in budget committees.

AI repairs nothing

The promise is real, and so is the condition. AI amplifies what exists, including its flaws. Shaky targeting automated by AI remains shaky targeting, delivered faster and at larger scale.

One level finer, and this is the trap specific to content generation: generated personalization is not relevant personalization. If the model writes a flawless paragraph on a false premise, you have reproduced the error at scale and accelerated its spread. The reader sees a well-written message about something that does not concern them.

The order of the work is therefore not negotiable: data, process, then AI. And measurement does not change either. A calculated property, an enrichment or an agent is judged on stage progression and pipeline, not on the number of enriched records.

The choice of tool matters less than you think

Tool comparisons take up a disproportionate place in automation projects. Six months of shortlisting, two days of thinking about the data model. The order should be reversed, because the tool question is settled quickly once the architecture is known.

The criterion that really matters in B2B is a single one: do the marketing tool and the CRM share the same database, or does a synchronization between the two have to be maintained? Everything else (the number of connectors, the sophistication of the email editor, the names of AI features) can be made up for. A synchronization to maintain cannot. It produces discrepancies, duplicates, workflows that trigger on outdated data, and a permanent maintenance load.

This is the structural reason why most of our clients operate on HubSpot Marketing Hub: marketing, sales and service work natively on the same objects, with no bridge to monitor. For teams coming from a Pardot and Salesforce combination, we have documented the switchover points in an article dedicated to replacing Pardot with HubSpot Marketing Hub.

This criterion also applies to the AI layer, and it is the point comparison grids miss. AI natively connected to CRM objects sees the account's full history: calls, tickets, emails, browsing. AI placed alongside only sees what someone has taken the trouble to export to it. It will produce poorer calculated properties, on older data, and you will spend the time saved maintaining the bridge.

Building a first setup in six weeks

There is no need to wait for a full program. The sequence we apply with our clients to get a setup that contributes, rather than one that simply runs, fits into three stages.

Weeks 1 and 2: decide and clean. Write the ICP, set the stage transition criteria, fix the fields that will be used for targeting. Then inventory existing workflows and turn off those that produced no progression over the last cycle. This cut improves readability before anything is added.

Weeks 3 and 4: one qualification and routing scenario. No nurturing to start with. Routing gives a visible result in a few weeks, requires no content production, and creates the agreement with sales on which everything else will depend. Write the three exits before building.

Weeks 5 and 6: one nurturing scenario, only one. On the most frequent intention in your database, with a quantified progression objective and a review date. One measured scenario is worth more than six assumed ones.

From there, the logic is that of a portfolio: you add a scenario, measure it over a cycle, then keep it or cut it.

Along the way, our marketing automation agency works on a flexible basis, from scoping to deployment and then ongoing support for MarketingOps teams.

Markentive customer case

Context. La Poste Solutions Business, the business services branch of the La Poste group, with more than 1,000 employees. A brand known to all its targets, which guarantees nothing about the ability to turn this awareness into pipeline.

Challenge. A digital strategy to develop, tools that did not talk to each other and data that was therefore unusable for targeting. Leads existed, but nothing showed which ones deserved a call. The classic symptom: lots of contacts, little qualification, sales reps who doubt what marketing sends them.

Solution. The work started with the foundation, not with the scenarios. Strategic scoping to settle the objectives and identify the data that was really useful, then definition of an end-to-end data management strategy: collection, structuring, use. The technical integration of HubSpot with existing internal tools followed, so that data would stop living in silos.

Once this foundation was in place, the activation setup: a form strategy designed to qualify rather than collect, a scoring logic to assess contacts' maturity, and automated journeys connected to this scoring. The order of operations matches exactly the one described above, and this is no coincidence: it is the only sequence that produces measurable results.

Results.

• x4 in the volume of new contacts generated
• x3.5 in leads passed on to the sales teams
• +82% in opportunities generated by marketing

The third figure is the one that counts. Multiplying contacts by four without increasing opportunities would simply have meant more noise. It is the gap between the three indicators that shows that qualification worked, not just acquisition.

The project details are on the La Poste Solutions Business customer case page.

Sort out your workflows

If you already have a setup in place, the most profitable exercise for the coming weeks is not adding a scenario. It is taking the list of your active workflows, looking for each one's measurable effect on contact progression, and cutting those that have none.

Learn more

• Marketing automation agency: scoping, deployment and long-term team support.
• HubSpot Marketing Hub: the platform on which most of our setups run, and why a shared database makes the difference.
• CRM data enrichment with Breeze: the mechanics of external enrichment and the governance to put in place before activating it.
• SalesOps & Data: the data quality work without which none of the above holds up.