Skip to content
All insights

Workflow automation

Five automation tools is five integration projects.

The first workflow is a project. The fifth is an architecture decision, and by then it has usually been made by accident.

Published April 28, 20266 min read

What this covers

  • Point tools each bring their own connectors, permissions model, audit trail and vendor. The cost compounds with every addition.
  • Workflows are not independent: the confirmation one agent tracks is the document another has to match.
  • One control layer means approval thresholds are set in one place and reviewed as one policy.
  • This is not an argument for a big-bang program. It is an argument for the second workflow costing less than the first.

Automation arrives one workflow at a time, and correctly so. Somebody has a quoting problem, buys a quoting tool, and it works. Six months later invoice processing gets the same treatment from a different vendor. A year after that, customer requests.

Each decision was defensible on its own. The result is three connections into the ERP built three different ways, three sets of permissions nobody reviews together, three audit trails in three formats, and three vendors with three roadmaps. The organization now has a second integration estate that nobody chose to build.

What the fragmentation actually costs

Separate toolsOne foundation
Systems connectionEach tool connects to the ERP its own wayConnected once, used by every agent
PermissionsA different model per tool, reviewed separately if at allOne set of scoped permissions, reviewed as one policy
Approval thresholdsConfigured per tool, in different languageSet once, in business terms, applied consistently
Audit trailThree formats, no way to follow a transaction across themOne log, one transaction followed end to end
Adding the next workflowA new procurement cycle and a new integrationConfiguration on a foundation that is already connected
Upgrading the ERPThree vendors to coordinateOne connection to revalidate

The fifth row is the one that changes the economics. On a shared foundation the second workflow is substantially cheaper than the first, because the expensive part (reaching the systems, agreeing the permissions, establishing the log) was paid for once. With separate tools, every workflow costs roughly what the first one did, forever.

Workflows are not independent

The architectural argument is real but abstract. The operational one is concrete, and it shows up as soon as two workflows touch the same document.

  • The supplier confirmation the purchasing agent tracks is the document the invoice has to be matched against three weeks later. Two tools means reading it twice and reconciling two interpretations.
  • The customer part number a person resolved during quoting is the same alias the order agent needs. In separate systems, that knowledge is learned twice.
  • A customer asking where their order is needs the order record, which is what the order agent created. Without shared context the answer is assembled from scratch every time.
  • A credit hold discovered during order entry is exactly the fact that should stop a quote going out to the same account tomorrow.

A business does not have five workflows. It has one flow of work that five teams touch, and the handoffs are where the time is lost.

What "one foundation" means concretely

The phrase is vague enough to be marketing, so it is worth being specific about what it has to include to mean anything.

  1. 01One connection to each system

    The ERP, CRM, mailboxes and document stores are connected once. A new agent inherits that access, scoped to what its job requires, rather than negotiating its own.

  2. 02One control layer

    Approval thresholds, value limits and escalation rules are defined in business language in one place, and apply to every agent. Reviewing delegation policy is one exercise, not five.

  3. 03One activity log

    Every action by every agent, in one record, so a single transaction can be followed from the email that started it to the invoice that closed it, across whichever agents touched it.

  4. 04Shared knowledge

    Part aliases, customer conventions and resolved exceptions are learned once and available to every agent that needs them.

  5. 05Ready-made agents plus your own

    The common workflows exist before a project starts. The workflows specific to your business are built on the same foundation rather than as a separate system.

The thing this is not

There is an obvious misreading of this argument, and it is worth refusing directly: none of it is a case for a company-wide program that delivers nothing for a year.

The sequencing that works is the opposite. Start with one workflow, chosen because it has volume and a defined outcome. Prove it. Then add the second, which is cheaper because the foundation exists. Each addition remains its own scoped decision with its own justification, and any of them can be the last without stranding what came before.

What the foundation buys is not speed on the first workflow. It is that the fifth one is still worth doing. Choosing your first workflow covers how to pick the one that goes first.

When it stops being optional

For one workflow, a point tool is a reasonable choice and the architecture argument is premature. The threshold is somewhere around the second or third, and there are three reliable signals that you have crossed it.

  1. 01Two tools are reading the same document and reaching different conclusions about it.
  2. 02Nobody can answer what all the automations are collectively permitted to do in the ERP.
  3. 03Following a single customer transaction end to end requires opening more than one audit trail.

Any one of those is a sign the estate is being assembled rather than designed. The AI Operating System describes what running every agent on one foundation looks like in practice.

Questions

Questions we get about this.

Does this mean a long program before anything works?

No, and that sequencing would be a mistake. It means starting with one workflow and making sure the foundation it runs on is the one the next workflow will use, so the second deployment is configuration rather than a new integration project.

What if we only ever automate one workflow?

Then a point tool is a reasonable choice and the architecture argument is premature. The question worth asking early is what you would pay for again when the second workflow arrives, because that answer is much cheaper to get now.

Can our own workflows run on it, or only ready-made ones?

Both. The common workflows exist before a project starts; the ones specific to your business are built on the same foundation, with the same connections, the same control layer and the same log, rather than as a separate system.

What does one control layer actually change?

Approval thresholds and value limits are defined once in business language and apply to every agent, so reviewing what your automations are collectively permitted to commit is a single exercise rather than an audit of several tools with different configuration models.

How does the shared log help?

It lets a single transaction be followed from the email that started it to the record that closed it, across whichever agents touched it on the way. With separate tools that trail breaks at each handoff, which is exactly where questions tend to arise.

See what this would look like in your operation.

A Workflow Review traces one of your real workflows end to end and names the first step an agent could take over. No platform to evaluate first.

Start with one workflow

Fixed scope for the first workflow, agreed before we start. Expansion is your decision once it proves value.

How it works

Security, stated plainly

Where the agents run, what they can and cannot do, and exactly what we do and do not claim.

Security