Pattern Automation
← Blog

Pattern Automation vs traditional automation consulting

A programme office versus a company OS. What Big Four automation practices deliver, what we leave in git, and the RFP questions that sort them.

Discuss this post in AI

Send a pre-filled prompt to ChatGPT, Claude, Gemini, or Perplexity — get a summary, ask follow-ups, or compare ideas from this guide.

Traditional automation consulting sells a programme: discovery, current-state, future-state, RFP, integrator, hypercare, CoE. Pattern Automation sells a company OS and, if you want us, a short implementation that leaves the OS in your git. The deliverable is not a slide that says “30 processes automated by Q4.” The deliverable is agents, skills, and policies you can diff.

This is a comparison, not a takedown. Big consultancies are good at stakeholder maps, regulated-industry programme management, and surviving a 200-person steering committee. They are structurally bad at leaving you a runtime you understand. We are the inverse: strong on the control plane, uninterested in a 18-month transformation office.

If you need both — a bank often does — hire them for the programme and us (or your own team) for the operating layer. Do not hire them as the operating layer.

What traditional automation consulting actually delivers

Accenture, Deloitte, IBM, the regional UiPath partners, the “digital” arms of the Big Four. The pattern has been stable since the first RPA wave:

  • A multi-month discovery that produces a process taxonomy and a business case.
  • Tool selection that often ratifies whatever the firm already staffs (UiPath, ServiceNow, a hyperscaler copilot).
  • Factory implementation: bots and workflows built by people who will not work in your company next year.
  • A CoE charter so the factory has a home after hypercare.
  • A change-request mill for every exception the bot could not handle.

You get governance theatre and, if the partner is good, some real throughput on stable processes. You also get a dependency: the knowledge is in their accelerators, their offshore pod, and a SharePoint the licence lapses on.

AI has not changed the commercial shape. It changed the demo. The same firms now staff “agentic” practices that wrap copilots around the same programme machinery. The invoice still looks like a transformation. The artifact is still not a repo you clone.

What Pattern Automation actually is

We build Neuro OS: an open-source AI management system. Agents, skills, memory, connectors, budgets, review. Work runs on isolated machines. Credentials stay out of the sandbox. You bring any model. You can self-host.

When we sit with a customer, the implementation looks like:

  • Connect the two or three systems the first process needs.
  • Write the skill for that process in your repository.
  • Set Allow / Ask / Block so writes cannot surprise you.
  • Run the first sessions with your operators in the thread.
  • Leave.

We will not run a 40-workshop current-state for a year. We will also not pretend a chatbot is a control environment. The product position is in Introducing Neuro OS and AI transformation needs a company OS.

Side by side

Traditional automation consulting Pattern Automation
Unit of sale Programme / SOW / CoE Platform + optional implementation
Time to first production process Often 6–18 months Weeks, if you pick one process
Where the automation lives Partner tools, orchestrators, slides Git you own
RPA vs agents Still mostly RPA + a copilot slide Agents with gated tools; RPA only where clicks must not improvise
Model lock-in Whatever the alliance is this year Any model, your keys
After they leave Hypercare, then a retainer git clone
Open source No Yes
Self-host Rarely the default First-class
Best customer Large programme, many stakeholders, stable processes Teams that will operate the loop themselves

Where they win

Pick a traditional consultancy when:

  • You must have a Big Four letterhead on the risk committee paper.
  • The blocker is not technology. It is 14 legal entities and a works council.
  • You need hundreds of people coordinated, and the automation is a workstream inside a larger ERP programme.

Those are real jobs. Neuro OS does not staff them.

Where they lose

They lose when the thing you needed was an operating system for agents and you bought a PowerPoint operating model instead. Symptoms:

  • Year two, you still cannot change a prompt without a change request.
  • Nobody in-house can explain which identity the bot uses.
  • Every new process is a new SOW.
  • The “AI CoE” is a reporting line, not a repo.

That is the same failure agencies produce at a smaller invoice.

The test to put in an RFP

Ask every bidder — including us — to answer in writing:

  1. Can we clone the automation on day 90 with no vendor account?
  2. Where do credentials live during a run? If the answer is “in the agent environment,” stop.
  3. What happens on a write? If there is no Ask, you have automated the signature.
  4. Can we change the model without rewriting the processes?
  5. Who is the principal — a shared robot account, or a named agent with a grant list?

Neuro OS is built so the answers are: yes; server-side gateway; the run pauses; yes; a named agent. Traditional consulting can use that stack. Most will try to replace the questions with a maturity model. Maturity models do not post the close.

How to use both without getting captured

Let the consultancy run the programme office if you need one. Require that the runtime is Neuro OS (or an equivalent you can fork), that skills land in your git, and that their accelerators are patches you keep, not a black box they license back. Pay them for the workshops. Do not pay them rent on your own operations.

If you do not need the programme office, skip it. Pick one process — financial reporting is a clean start — put the architecture in place, and grow the skill library. That is automation as an operating system, not as a project that ends.

Explore Neuro OS →

More from Blog