SaaS Integration: Multi-Step Workflows & The Category’s Future

Retrieving and remixing data from web APIs and SaaS through two or more steps is a decade-old idea, starting with Yahoo Pipes in 2007 if not earlier (screenshot above from Derek Banas). While the concept is promising, the execution has often fallen short because of a variety of hurdles:

  • The broader the targeted user base, the harder it is to provide a user experience that less technically-inclined people can figure out. It’s one thing to serve developers and tech-savvy business analysts, but it takes more UI chops to be within reach of your average marketer who’s likely challenged by, say, simple pivot tables.
  • APIs keep changing all the time, so reliable SaaS middleware has to do a lot behind the scenes to limit breakage and confusion for its end users. Otherwise your workflow works, and then it doesn’t.
  • What’s the business model? Yahoo Pipes obviously didn’t have any. On the other end, enterprise tools tend to limit themselves to the “if you have to ask, you can’t afford it” upper end of the market. It’s only in the last couple of years that affordable but sustainable SaaS integration (to do more than basic 2-step workflows) emerged. That means relatively cheap subscriptions, because there’s no way ads or one-off sales can support this sort of product over the long haul.

In a previous entry I tracked the players and events in the SaaS integration market. Here I’ll start by comparing what kind of UI is delivered by market participants to manage nontrivial workflows, and I’ll conclude with a few strategic considerations.

1. Multistep Workflow UX Review

I’ll focus this review on the desktop experience. I don’t believe mobile is terribly relevant here, except maybe to quickly edit existing workflows. On the other hand, some mobile app design trends based on tactile input may be relevant even on web apps intended for desktop use.

1.1. Zapier

Zapier introduced multistep workflows as a paid feature in February 2016. This expanded the tool from 2 (Trigger-Action) to (I believe, but haven’t tested) 25 steps. They published a teardown of the redesigned editor, which highlights the UX perils coming with extra complexity:

That experience could be quite stressful or intimidating with even just 2 steps. Add in the complexity of multiple steps with multiple actions, and things get interesting.


In reality, there weren’t more steps [in the V1 redesign], but users thought there were. Certain steps seemed like they were dragging, and they introduced frustration to the experience. Our initial redesign added power, but the effortless feeling was gone. Shipping something sluggish wasn’t an option.

This editor indeed offers a straightforward user experience because it’s a simple serial/linear cascade with a source trigger followed by filters and actions, which is why Zapier was dubbed the IFTTT (if this then that) for business:

Serial zaps

Serial Zapier

There’s a significant shortcoming if you create longer flows: you can’t move actions or filters once you’ve created them. You can however add them within the chain, so if you just realized you needed a step upstream, you can add it after the fact:


Insert Filter Here, Save $$

This has cost implications because each successfully completed action in a workflow consumes a task against your monthly plan, so you obviously want to place filters as high as possible in the stream.

Also, order execution matters because fields from earlier steps are available to be inserted downstream:


Trickle Down Webonomics

Another big improvement phased in by Zapier at the same time was Search Actions, which allow you to access existing content, as opposed to only pipe the flow of new content coming from your apps. OneLogin announced federated cloud search in 2012, but I don’t think they ever delivered it. There’s the seed of something bigger there.

This type of feature intersects with the “account based marketing” trend du jour, where businesses try to build a more customer-centric and integrated perspective, as opposed to react to a flow of separate inbound streams. See for instance how UnifiedVU pitches their product to build a single customer view:

Zapier’s content aggregation has also recently been improved, with the RSS trigger allowing up to 10 sources, and the new Digest action. This is welcome as it is way too easy to turn an app such as Slack into a firehose filled with micro notifications that’s way too noisy, which is funny given one impetus behind such teamwork apps was to reduce email noise.

The ability to add JS or Python code between actions lets you massage the data and shows Zapier doesn’t intend to lock itself out of more sophisticated users and use cases.

Overall, good job from Zapier. A few final observations:

  • Zapier doesn’t have loops, and search actions stop at the first match. This can be a dealbreaker that rules out Zapier for a whole bunch of desired workflows.
  • A tip for users: copy existing workflows that you want to edit or expand, because any significant change will suspend execution until editing is done and the Zap is reactivated.
  • In October 2018 Zapier added Paths to support conditional logic at the end of workflows.

1.2. Microsoft Flow

Microsoft introduced a preview of Flow in April 2016 and expanded to general availability at the end of October. Its lead PM is Stephen Siciliano who comes from MetricsHub, a SaaS vendor acquired by Microsoft in 2013. Here’s an hour-long video where he presents the tool:

Microsoft also launched PowerApps at about the same date. This service targets the creation of more complex apps beyond Trigger/Filter/Action chains, a la

Both tools integrate, or at least they’re supposed to, because I’m getting an error trying to see my Flows from PowerApps (while being logged in both, from the same user account):

Good thing I'm familiar with JSON /s

Good thing I’m familiar with JSON /s

Also, connections with 3rd parties are not shared, so even if you authorized, say, Google Drive in Flow, you have to do it again in PowerApps. Looks like Microsoft still didn’t figure out authentication, it’s always more complicated than it needs to be.


Selectable in Zapier

Back to Flow, it started from the get-go with multi-step support (up to 64 as of October 2016) and includes branching conditions and looping that you won’t find in Zapier. That power comes at the cost of simplicity though.

It’s not necessarily that Flow is more complicated per se, but beyond what you can seen in screenshots, Flow doesn’t feel as seamless, responsive to your requests and easy to use. It uses constraining side boxes to set up actions and search through available fields/values where Zapier makes better use of available space. You can select steps with Ctrl+arrow then cut/paste them in Zapier, you can’t in Flow. These details add up. The sort of UI affordance that Google has been aiming for with their Material Design seems largely absent in Flow. Refinements such as CSS animations when boxes are opened and closed would help make the app feel less abrupt.

Also, it definitely looks drab and IT-ish with its grey background and square angles, in comparison with Zapier’s use of round shapes, white space, bright orange (#ff5c1a, what happened to good old FF4A00?) and green buttons. The whole thing needs polishing.


1.3. Cloudpipes

This product feels like a British Zapier clone, with its linear multistep pipes. Remember they drive on the left in the UK:

Chain chain chain

Chain chain chain

I’m not entirely fair in calling them a clone. Cloudpipes has the edge with its support of drag-and-drop reordering, though this feature’s discovery could be improved by changing the cursor to the Move arrow-cross cursor.

Drag and Drop

Drag and Drop

I like how the output of each step is summarized so that you can easily see the object and its properties that the next step will work with.

Output object properties

Output object properties

Conditional If/Then/Else is well done, as is access to the properties of previous steps:


Among blueprints already built by Cloudpipes, there’s one that monitors tweets, translates them, analyzes their sentiment, then post to Slack those that are deemed negative. It’s a good one, and both Zapier and Flow may want to build libraries/templates of more complex workflows that users can derive inspiration from, or simply tweak and make their own.

As we’re about to see, there’s a more powerful service elsewhere, but Cloudpipes delivers a solid alternative in terms of making serial workflows easy to use.


tray-draggableTray launched its flows in March 2016. They’re a significant step up from Zapier or Flow with the drag-and-drop interface and core features spanning the likes of Boolean conditions, branching, data mappers, and helpers specialized in handling text, dates, lists and more. Unlike with Flow or Zapier, you can rejigger workflows after the fact by moving actions around. I’m not sure how many steps can be used, but a blog post introducing an editor upgrade in March 2016 mentioned 50+ steps.

The UI feels somewhat like Flow done with a Zapier flair. Where Tray is most limited is in the number of supported apps. If the apps you’re using are listed and you need to build complex workflows, then I’d give it a close look.

Looks good, mostly in a wider window

Looks good, mostly in a wider window

This reminds me of a defunct ancestor...

This reminds me of a defunct ancestor…

James Gill of GoSquared had this to say on Product Hunt [emphasis mine]:

“We’ve been using Tray for a while and love the power it gives us to integrate with tons of other tools, and do far more complex operations than other platforms in the space. It seems like they’ve nailed the balance between simplicity of setup and powerful dev-friendly advanced functionality.”

1.5. Workato

Workato’s recipe designer is less visual than its competitors listed above, and like Flow it definitely could use more design flair. On the other hand I like the options it offers when you add a step past the initial trigger/action:

Add a Step in Workato

The ability to chain recipes together is a really nice touch, they call this modularity “Callable recipes“:

“Let’s say that we wish to build the following recipe: Salesforce closed-won opportunity creates a Quickbooks Online invoice (and creates a customer if the customer doesn’t already exist)

We can turn steps 3-4 into a callable recipe so that we can reuse the (Search QBO customer, create if doesn’t exist) logic in any recipe we want. Hence we won’t have to rebuild these 3 steps and retest them and their mappings over and over again.”

Moreover, you can attribute a REST endpoint to these callable recipes, turning them into mini APIs.

Workato also adds value to its workflows with features such as error monitoring, commenting, versioning, and comments.

Workato versioning


It’s interesting to see how differentiated its Flow Express (added in October 2016) and Flow Enterprise offerings. The first one targets “business users and citizen integrators” with a point-and-click sequential wizard, while the latter offers a drag-and-drop UI for “power users, IT admins and developers.” Express allows multi-step flows but remains linear, while Enterprise adds branching and recursive flows.’s conclusion seems to be that “regular people” don’t get loops and needs to be literally funneled into linear sequences. So far this seems to be Zapier’s implicit conclusion so far. But what I’ve seen happen is these “regular people” fumbling with a bunch of half-done Zaps that they either give up on, or they call someone like me to sort out. Be as simple as you need to be, but not simpler…’s Parallel Workflows: Not for Dummies

1.7. Enterprise iPaaS

Companies such as SnapLogic come at the upper end of the Integration Platform as a Service (iPaaS) market, i.e. the type of company Gartner and Forrester will feature on their fancy charts. They definitely have their place because you won’t be able to connect a Redshift datawarehouse to a Tableau server with the small business/non-tech products.

That said, history teaches us that it’s easier for vendors to move upmarket than downmarket, as per trends such as the consumerization of enterprise software, the rising weight of CMOs vs. CIOs when it comes to buying software, Bring Your Own Device / Bring Your Own Application etc. You know how it goes: IBM laughed at Microsoft, Microsoft scoffed at Google, and now Google is ignoring your puny startup.

The videos below show the UI used by SnapLogic, Informatica, Mulesoft, Dell Boomi, Jitterbit – there are only so many ways to represent the same underlying concepts:

Enterprise iPaaS not only connects to enterprise software, supports more authentication/security/auditing, and scales to higher transactional volumes, but typically it also supports multi multi-step workflows, as shown in the Jitterbit screenshot below.

Except for Workato, Zapier and its ilk do not allow you to take the output of a workflow and make that the input of another one, let alone do that on the same screen. And it’s not obvious that they should, unless they figure out how not to overwhelm their core non-technical users.

Deathstar Laser Cooling Subsystem, Section II

Deathstar Laser Cooling Subsystem, Section II

1.8. Pipefy

Pipefy is not a direct competitor to the products above, but it does focus on multistep workflows via a Kanban-like UI (think Trello) and Zapier-powered integrations. It’s something you’d use downstream of an API integration tool, once you’ve retrieved and massaged data from your source apps.

UI-wise, it’s another good example of drag-and-drop at work:


File name retrieved from Google Sheets via Zapier

2. Where Else to Find Inspiration?

I’m sure all these competitors look at each other like I just did above. I think there’s also something to be gained from dissecting how building/sandbox/puzzle games execute their UI with the assumption that nobody will read the manual, and at best gamers will fly through a tutorial. For instance, I’d look at Minecraft, SpaceChem, Infinifactory, or Portal, but also match-3 games that are so popular on Facebook. I think gamification can be overdone in a business context, so I’m not referring to that trend (so 2014!) but rather how interactivity works through sensory cues in games.

Think about it, games are software too! The best ones let you guess intuitively how building blocks snap into place, interact with each other, and whether success or failure conditions are met. Would 3D work for business workflow automation? Probably not, so don’t be too literal if you do look at game UI design. You’d want to keep it subtle, but glowing halos, magnet snap-ins, or sound effects, could all help keep users work through more complex workflow setups. A lot of that stuff can be done with CSS in modern browsers.

3. What’s Next?

Microsoft, while not as scary as it was 15 years ago, remains a powerful behemoth. Their arrival in this market must have been met by the “incumbent startups” (hah) with a mix of trepidation (“they’ll spend a gazillion dollars and integrate with everything”), disdain (“meh, they’ll make this a Sharepoint feature that nobody uses by next year”), bullishness (“this validates our market!”), denial (“nobody will want to integrate with Microsoft”), and confidence (“by version 3 they’ll have our 2015 product”).

Flow has clearly taken inspiration from Zapier and its competitors, and somewhat of a feature arms race can be expected. Product designers will have to remember that ease-of-use, discoverability or speed are features too. Just in the past couple of months, Zapier added the Push Chrome extension, while Microsoft launched button flows (mobile only, but browser support is in their backlog). More importantly Microsoft has some catch up to do with regards to the number of supported integrations, and how deep these integrations go.

Right now it feels like, while these products make the same promise, they go after different market segments:

  • Zapier’s primary user is explicitly a savvy but not very technical person, most likely a marketer. They’re stepping up their tech abilities, but they’re not engineers.
  • Flow will succeed mostly with Microsoft’s core of Office cloud users and Azure developers. A big market, but not eating the world. That said, watch the Power BI + PowerApps + Flow combo, because Power BI is a force to behold.
  • Tray and Workato will be what Zapier or Flow more advanced users upgrade to when they need to build complex workflows, provided their app stack is supported.
  • Salesforce and Google would be obvious newcomers in this space, given they introduced products both in the self-service BI and in the low-code app maker worlds. Amazon may get there too. These guys tend to get into the same spaces,and iPaaS is adjacent to (almost always) multi-source BI/analytics, where SFDC has Wave, Google offers GA, Data Studio, and Big Query, and Amazon is present with QuickSight.
  • Most of the other guys offering similar services don’t seem to have the same sort of momentum or strengths to bring to the table. They may stick around for years with a slowly dwindling user base, breaking even but not really in a position to grow.
  • Do the tools positioned for the enterprise really offer features and integrations that justify their pricepoint? Are they easy enough to use that they don’t become shelfware? These are key questions for their survival.

A more fundamental question, whatever the market segment:

  • Will SaaS players ramp up their own integration efforts so much that middleware will become less useful?
  • Will dashboard/single view building tools reduce the need for middleware plumbing because they let people jump right to the end results they want?

The ideal for SaaS customers would be to able to build their customized stack from best-of-breed applications while delivering to their internal users a seamless experience that feels like an integrated suite. Middleware, whether in the form of API glue or front-end scotch tape, should remain valuable to help minimize the power of rent-seeking vendors would let their products become obsolete and/or ask for too much money for too little just because of their dominant position.

4. See also

Leave a Reply

Your email address will not be published. Required fields are marked *