Why Your Users Aren't Building Workflows (And How AI Fixes It)

Updated February 05, 2026

TL;DR: The average feature adoption rate across B2B SaaS is just 24.5% — and for complex features like visual workflow builders, it's far worse. Graph-based canvases (campaign flows, IVR trees, automation sequences) demand abstract logic that most users don't have. Traditional onboarding tools like tooltips and product tours can't solve this because the problem isn't "finding the feature" — it's "knowing what to build." Tandem takes a different approach: an AI agent embedded in your product that lets users describe their use case in plain language, then generates the graph directly in the interface. Qonto used this approach to guide 375,000 users through complex workflows, and Aircall saw a 20% activation lift on self-serve accounts.

Your most powerful feature is gathering dust

You spent months building it. The visual workflow builder — the canvas where users drag blocks, draw connections, set conditions, and create logic that makes your product truly powerful. Maybe it's a campaign automation flow. Maybe it's an IVR phone tree. Maybe it's a lead scoring sequence or a multi-step approval chain.

It's the feature your power users love and your new users never touch. According to Pendo's benchmarking data, the median feature adoption rate across products is just 6.4%, with 80% of click volume concentrated in a small fraction of features. For features that require users to construct abstract logic — to think in graphs, branches, and conditions — the adoption gap is even more painful.

This isn't a documentation problem. It isn't a discoverability problem. Your users know the workflow builder exists. They've clicked into it, stared at the blank canvas, and clicked away. The problem is structural: you're asking business users to think like engineers.

The canvas problem: why visual builders fail most users

Visual workflow builders were designed to make logic accessible. The theory was sound: replace code with drag-and-drop, replace syntax with blocks and arrows, and anyone could build complex automations. But the theory missed something fundamental about how most people think.

The first flawed assumption is that visual equals intuitive. A canvas full of blocks and connectors looks intuitive to someone who already understands the underlying logic. To a marketing manager trying to set up a drip campaign, or a support lead configuring a voicemail routing tree, the blank canvas creates the same paralysis as a blank code editor. The abstraction shifted from text to visuals, but the cognitive load stayed the same.

The second flawed assumption is that showing the components teaches the logic. Traditional onboarding approaches — tooltips explaining what each block does, product tours walking through the sidebar — address the wrong layer of the problem. Users don't struggle because they can't find the "delay" block or the "condition" node. They struggle because they don't know what combination of blocks, in what sequence, with what conditions, will produce the outcome they need. Research on business process management implementations shows a 60 to 80 percent failure rate, with only 15% of firms satisfied with the results — and a major driver is that complex workflows are too rigid and abstract for real work situations.

The third flawed assumption is that power users represent your user base. The users who build intricate 30-node workflows with nested conditions are your top 5%. They'd figure out the interface with or without help. The other 95% — the users who signed up because they need one specific workflow to solve one specific pain — are the ones who abandon the builder and open a support ticket instead.

The real adoption gap: intent versus interface

Here's a scenario that plays out thousands of times daily across B2B SaaS. A customer support manager at a mid-size company needs to configure voicemail routing: if the caller is an enterprise client, route to the dedicated team; if it's after hours, play a message and offer a callback; if the queue is full, deflect to self-service.

This person knows exactly what they want. They could explain it in two sentences to a colleague. But when they open the visual flow builder, they face: which block type do I start with? How do I represent "enterprise client" as a condition? Where does the "after hours" branch connect? What happens if two conditions are true simultaneously?

The gap isn't knowledge about the business logic. It's the translation layer between "what I want" and "how the interface represents it." Every traditional onboarding tool — tours, tooltips, checklists, help docs, even video walkthroughs — tries to teach the user to bridge that gap themselves. For features with inherent logical complexity, that approach has a ceiling. Interactive walkthroughs drive higher adoption at around 31% compared to 16.5% for traditional documentation, but that still means nearly 70% of users aren't adopting.

The alternative is to eliminate the translation layer entirely.

From "learn the tool" to "describe what you need"

This is the shift that AI-powered agents enable for complex workflow features. Instead of teaching users to think in graphs, you let them speak in outcomes — and the AI constructs the graph for them.

The approach works like a conversation. A user opens the workflow builder and, instead of staring at a blank canvas, they interact with an AI agent embedded directly on the page. They describe their use case: "I want to send a welcome email when someone signs up, wait three days, check if they've completed onboarding, and if not, send a reminder with a link to the getting-started guide." The agent asks clarifying questions — "What should happen if they complete onboarding during the three-day wait?" — and then generates the complete workflow graph in the canvas. Blocks placed, conditions configured, connections drawn.

The graph still exists. Power users can still inspect it, modify individual nodes, add edge cases. But the entry point has shifted from "construct the logic yourself" to "describe what you need and refine from there." The visual canvas becomes a verification layer and an editing surface, not the primary authoring interface.

This is the approach Tandem enables. Tandem is an AI agent that operates directly inside your product's interface — not in a chatbot sidebar, not in a help center popup, but on the actual page where the user is trying to get work done. When deployed on a workflow builder screen, the agent can interpret the user's natural language description, interact with the interface elements, and construct the graph by manipulating the same UI the user would.

How AI-generated workflows change user behavior

The shift from manual graph construction to AI-assisted generation changes several dynamics simultaneously.

Time-to-first-value collapses

Instead of spending 45 minutes learning the builder, failing, watching a tutorial, and trying again, a user can have a working workflow in under five minutes. Companies with self-serve revenue score 18.3% higher on time-to-value delivery than those without — and reducing time-to-first-value on complex features is one of the highest-leverage ways to improve self-serve conversion.

Tandem's deployment at Qonto demonstrated this concretely: 40% faster time-to-first-value across their platform, with feature adoption tripling in the first month for features where Tandem was active. At Aircall, self-serve accounts saw a 20% activation lift because advanced features that previously required human onboarding became genuinely self-serve.

Support ticket volume drops

When users can't build the workflow they need, they contact support. They describe what they want, a support agent builds it for them (or walks them through it over a 30-minute call), and the ticket resolves. But the user hasn't learned anything — next time they need a different workflow, they open another ticket.

AI-powered workflow generation intercepts this loop. The user describes their need to the agent the same way they'd describe it to a support rep — but they get an immediate result, and they can see how the AI translated their request into the graph. Over time, this creates a learning loop where users start understanding the visual representation because they're seeing their own intent reflected in it.

Feature adoption expands beyond power users

The 24.5% average feature adoption rate across B2B SaaS exists partly because complex features self-select for technical users. When the entry barrier drops from "understand graph logic" to "describe what you want," the addressable user base for your workflow builder expands dramatically. Sellsy saw an 18% activation lift after deploying Tandem, driven largely by users who had previously abandoned complex features during onboarding.

Where traditional DAPs fall short on workflow builders

Digital adoption platforms like WalkMe, Pendo, and Appcues were designed for a different class of onboarding problems. They excel at: showing users where features are, guiding them through linear sequences of clicks, and surfacing contextual tooltips when users hover over unfamiliar UI elements. These are real and valuable capabilities for straightforward features.

But workflow builders aren't linear. There's no single "correct" click sequence because every user's desired workflow is different. A product tour that walks through "here's how to add a condition block" doesn't help the user who doesn't know whether they need a condition block or a delay block or a branch block for their specific use case. The fundamental limitation is that traditional DAPs show and explain, but they can't do. They can point to the canvas and describe the tools, but they can't pick up the tools and build something on the user's behalf.

Tandem operates on a different model — what the company calls the explain/guide/execute framework. The agent can explain how a feature works, guide users through a process step by step, or execute actions directly in the interface. For workflow builders, the "execute" capability is transformative: the AI doesn't just tell users what to do, it does the work of constructing the graph while the user watches, then hands back control for refinement.

At-a-glance: approaches to workflow builder adoption

What Tandem does differently on complex features

Tandem deploys as a single JavaScript snippet — technical setup takes under an hour with no backend changes required. The AI agent appears on whichever pages you configure, and the no-code configuration interface lets product or ops teams set it up without engineering support.

For workflow builder screens specifically, Tandem's agent operates in a conversational loop. The user describes their goal in natural language. The agent asks targeted clarifying questions (what triggers the workflow? what conditions matter? what should happen at each branch?). Then the agent generates the complete graph by interacting with the builder's UI — placing nodes, configuring properties, drawing connections. The user sees the result, can ask follow-up questions ("what happens if the customer doesn't respond within 48 hours?"), and can request modifications ("add a fallback to voicemail on the second branch").

The self-healing architecture means that when your product team updates the workflow builder's UI — renames a block type, moves a sidebar, changes a dropdown — Tandem adapts automatically rather than requiring manual flow updates. This is a meaningful operational difference compared to traditional DAPs where every UI change can break authored walkthroughs.

Proof points from production deployments validate the approach at scale. Qonto deployed Tandem across 100,000+ users to activate paid features including insurance and card upgrades, and guided 375,000 users through a full interface redesign. Their account aggregation feature saw adoption jump from 8% to 16%, and feature adoption tripled in the first month. Over 10,000 users engaged with insurance and premium card features within two months of deployment.

Honest limitations

Tandem is a young company, founded in 2024, and the product has real boundaries you should evaluate. It's web-only — native mobile support is not yet available. It does not include deep product analytics in the way tools like Pendo or Amplitude do; if you need cohort analysis, funnel visualization, or feature-level engagement tracking, you'll want to pair Tandem with a dedicated analytics platform. Pricing is custom with no public tiers, which makes comparison shopping harder. And like every digital adoption tool, content work is ongoing — you still need to write messages, refine targeting, and update experiences. Tandem reduces technical maintenance through self-healing, but the strategic work of designing good onboarding content is universal.

The ROI case: what workflow builder adoption is worth

Improving adoption of complex features has compounding effects on retention, expansion revenue, and support costs. Here's a framework to estimate the value for your specific product.

Fill in your numbers

Start with these inputs:

• A: Number of users who open your workflow builder monthly

• B: Current completion rate (% who actually create and save a workflow)

• C: Average revenue per user per month

• D: Retention rate difference between users who adopt vs. don't adopt complex features (industry data suggests users engaging with 70%+ of core features are twice as likely to retain)

• E: Monthly support tickets related to workflow builder help

• F: Average cost per support ticket ($15-35 is the typical range)

Worked example

Consider a B2B SaaS product with 5,000 monthly users visiting the workflow builder:

• Current completion rate: 15% (750 users create a workflow)

• After AI agent deployment (conservative 2x improvement): 30% (1,500 users)

• Additional 750 activated users per month

• If these users retain at even 20% higher rates (modest, given the 2x benchmark), and average contract value is $200/month, that's: 750 users x 20% retention improvement x $200 = $30,000/month in prevented churn

• Workflow-related support tickets: 400/month at $25/ticket = $10,000/month. If AI agent deflects 40% of those: $4,000/month in support savings

The Qonto case provides grounding for these estimates: account aggregation moved from 8% to 16% adoption — a full doubling — and Aircall's 20% activation lift on self-serve accounts translated directly into reduced onboarding costs and higher feature engagement.

Decision framework: when each approach fits

Use tooltips and in-app guides when your workflow builder is relatively simple (under 5 block types, linear or near-linear flow), your users are technical, and the main barrier is feature discoverability rather than logic construction.

Use a traditional DAP (WalkMe, Pendo, Appcues) when your primary adoption challenge is across many features, not concentrated in one complex builder. When you need deep product analytics alongside onboarding. When your workflows have a small number of common patterns that can be templated into guided tours.

Use an AI agent like Tandem when your workflow builder involves branching logic, conditions, and non-linear construction. When each user's desired workflow is different enough that templated tours don't cover it. When your users are business operators, not engineers. When you're seeing high open-rates on the builder but low completion rates. When support tickets disproportionately involve "help me build this workflow" requests.

Combine approaches when you want Tandem for complex features and a traditional DAP for everything else. Or pair Tandem with Amplitude/Mixpanel for analytics and Tandem for execution. These tools aren't mutually exclusive — they operate at different layers of the adoption stack.

Specific SaaS categories where this matters most

The "blank canvas" problem hits hardest in products where users must construct logic:

Marketing automation — campaign flow builders (Customer.io, HubSpot, ActiveCampaign, Brevo) where users build multi-step sequences with delays, conditions, and branches. New users frequently default to simple single-send emails because the visual flow builder is too intimidating.

Telephony and contact center — IVR flow builders (Aircall, Dialpad, Five9, Genesys) where users must design call routing trees with DTMF inputs, time-of-day conditions, queue overflow handling, and voicemail fallbacks. These are some of the most logic-heavy canvas interfaces in SaaS.

Workflow automation — integration platforms (Zapier, Make, n8n, Workato) where users connect multiple apps through multi-step sequences with filters and conditions. Even "no-code" platforms have a learning curve that blocks non-technical users.

CRM and sales operations — lead scoring and routing workflows, deal stage automation, and approval chains. Sellsy's 18% activation lift came partly from making these kinds of complex CRM workflows accessible to small business users who lacked technical teams.

DevOps and CI/CD — pipeline builders where even technical users spend significant time constructing build, test, and deploy sequences. The canvas metaphor is so prevalent in this space that "pipeline as code" emerged partly as a reaction to the UX limitations of visual builders.

Measuring success: the right metrics for workflow builder adoption

If you deploy an AI agent to improve workflow builder adoption, track these metrics specifically:

Workflow creation rate — the percentage of users who open the builder and actually create, save, and activate a workflow. This is your primary success metric. Benchmark your current rate before deployment so you can measure lift accurately.

Time to first workflow — how long from first opening the builder to saving a complete workflow. AI-assisted generation should compress this dramatically. Tandem's Qonto deployment showed 40% faster time-to-first-value across their platform.

Workflow complexity distribution — are users creating more sophisticated workflows (more nodes, more conditions, more branches) after AI assistance? If the distribution shifts rightward, users are getting more value from the feature.

Support ticket displacement — track tickets tagged to workflow builder categories before and after deployment. Look at both volume and resolution time for tickets that still come through.

Workflow reuse and iteration — are users coming back to modify their workflows or create new ones? This indicates whether the AI assistance created genuine understanding or just one-time results.

Getting started: a 20-minute demo on your most complex workflow

If your product has a visual workflow builder, a campaign canvas, an IVR flow designer, or any feature where users construct logic through a graph interface — and you're seeing the adoption gap described in this article — Tandem is worth evaluating.

Book a 20-minute demo where the team will deploy Tandem on your most complex workflow builder screen and show it generating a real workflow from a natural language description. You can also request reference calls with leaders at Aircall and Qonto who've deployed Tandem at scale across hundreds of thousands of users.

FAQ

How long does it take to deploy Tandem on a workflow builder screen?

Technical setup is a single JavaScript snippet that takes under an hour to install, with no backend changes. Configuring the AI agent for your specific workflow builder — defining the context, testing interactions, and refining responses — typically takes a few days of iteration through Tandem's no-code configuration interface.

Will the AI-generated workflows be as good as manually built ones?

The AI generates workflows using the same UI components and logic as a human user would. The output is identical in structure — same blocks, same connections, same conditions. The difference is the input: natural language instead of manual drag-and-drop. Power users can (and should) review and refine AI-generated workflows, especially for high-stakes automations.

Does Tandem work with any visual builder, or only specific platforms?

Tandem operates at the UI layer — it interacts with whatever interface your web application presents. This means it works with custom-built workflow builders, not just specific third-party platforms. If your users can build workflows through a browser-based interface, Tandem can assist them.

How does Tandem handle workflow builder UI updates?

Tandem's self-healing architecture adapts to UI changes automatically. When your product team renames blocks, moves panels, or changes dropdown options, the agent adjusts its behavior without requiring manual flow updates. This contrasts with traditional DAPs where every UI change risks breaking authored walkthroughs.

What about mobile users?

Tandem is currently web-only. Mobile support is on the roadmap but not yet available. If your workflow builder has a meaningful mobile user base, this is a real limitation to evaluate. That said, most visual canvas interfaces are desktop-oriented by nature.

How quickly can I expect to see ROI?

Based on existing deployments, initial results appear within the first month. Qonto's feature adoption tripled in the first month, and over 10,000 users engaged with new features within two months. The timeline depends on your traffic volume and how central the workflow builder is to your product's value proposition. Results vary by implementation, so treating the first 30-60 days as a measurement period is wise.

Can Tandem replace our existing analytics tool?

No — and it shouldn't try to. Tandem provides session-level and conversational analytics (what users ask, where they get stuck, completion rates), but it doesn't offer the depth of product analytics platforms like Amplitude, Mixpanel, or Pendo. The recommended approach is to pair Tandem for execution and adoption with your existing analytics tool for measurement and insight.

Glossary

Activation rate — The percentage of new users who reach a defined value milestone in your product. For SaaS products, the average activation rate is 36% with a median of 30%, according to Lenny Rachitsky's benchmarking survey of 500+ products. Top performers exceed 50%.

Feature adoption rate — The percentage of your user base actively using a specific feature. The average across B2B SaaS is 24.5% according to Userpilot benchmarks, though this varies significantly by feature complexity and industry.

Visual workflow builder — A canvas-based interface where users construct automated processes by dragging, dropping, and connecting blocks that represent actions, conditions, delays, and triggers. Common in marketing automation, telephony, CRM, and integration platforms.

Explain/guide/execute framework — Tandem's three-tier approach to in-app AI assistance. "Explain" answers questions about features; "guide" walks users through processes step by step; "execute" performs actions directly in the interface on the user's behalf.

Digital adoption platform (DAP) — Software layered on top of existing applications to help users learn and use features through tooltips, product tours, checklists, and contextual help. Examples include WalkMe, Pendo, Whatfix, and Appcues.

Time-to-first-value (TTFV) — The duration between a user's first interaction with a product (or feature) and the moment they experience meaningful value. For workflow builders, this is the time from opening the canvas to having a working automation running.

Self-healing architecture — A system design where the application automatically adapts to changes in the underlying interface without requiring manual updates. In the context of onboarding tools, this means the tool continues to function correctly even when the product UI is updated.

Graph interface — A UI pattern where users create logic by connecting nodes (representing actions or conditions) with edges (representing the flow between them). Also called canvas interfaces, flow builders, or visual programming environments.