Guide · Process Auditing
AI Readiness Assessment for Operations: Score One Workflow Before You Deploy
How to run an AI readiness assessment on the workflow you plan to automate: the six areas to check, a scorecard, and what each result tells you to do next.
By Ground Truth Ops ·
An AI readiness assessment decides whether an AI agent can run reliably inside a specific piece of work. Most assessments score the whole company: data strategy, infrastructure, skills, governance. For an operations team the deciding unit is smaller: one workflow, as it actually runs today. This guide covers what to check in that workflow, how to gather the evidence, and how to read the result as one of three calls: deploy, repair first, or choose a different workflow.
Company-level readiness vs. workflow-level readiness
Most AI readiness assessments you will find are organisation-wide surveys. They ask whether leadership has an AI strategy, whether there is a data platform, whether teams have the skills, whether governance policies exist. Those questions are worth answering, but a company can pass every one of them and still put an agent into a process that breaks on its first week of real exceptions.
The failure that operations leaders actually live with sits one level down. An agent follows the process it is given. If that process is the one written in the manual rather than the one people run, the agent is faithful to the wrong process, and it is faithful at speed. Automating a broken process makes it fail faster: the colleague who used to catch the missing field or the misrouted request is no longer in the loop.
So run both, in this order:
- Company-level readiness tells you whether the organisation can support AI at all: budget, data access, security review, someone accountable.
- Workflow-level readiness tells you whether this workflow, today, is a place an agent can operate. It is the assessment this guide covers, and it is the one that decides whether a first deployment succeeds.
Step 1: Choose the workflow to assess
Assess one workflow, not a department. A readiness score for "procurement" or "dispatch" averages good and bad sub-processes together and tells you nothing you can act on. Pick a workflow with a clear start and end: a purchase request to a released order, a load tender to a confirmed booking, an inspection request to a closed punch item.
Good candidates share four traits, the same filters set out in Prioritizing Which Process to Redesign First:
- Pain. The organisation already absorbs a recurring, measurable cost from it: delay, rework, missed service levels.
- Data readiness. Its inputs are digital, structured and reachable today, without a migration first.
- Measurability. You can name the number that would prove a redesign worked, and you can measure it now as a baseline.
- Structural soundness. The workflow is not about to be reorganised, outsourced or replaced by a system change.
If no workflow clears all four, that is itself the result of the assessment: the organisation is not ready to deploy, and the next step is to triage the workflows until one does.
Step 2: Gather evidence from the work, not the documentation
A readiness assessment built from the process manual and a workshop will score the declared process. You need the actual one. Collect evidence from four sources:
- Separate walkthroughs. Ask each person who runs the workflow to walk through it on their own, then compare the accounts. Disagreement on sequence, on who decides, or on what "done" means shows where the process was never standardised.
- Observation. Watch the work where it happens: the terminal, the yard, the site office, the inbox. Note every step that does not appear in the documented process, every side spreadsheet and every phone call. That is the shadow work an agent will not know about.
- Exception records. Pull whatever log exists of cases that left the standard path: overrides, manual corrections, escalations, credit notes, change orders. Exception logs show the failure patterns the process map hides. If no record exists, that is a finding.
- System data. Export a recent sample of the records the workflow produces and check them against what people told you. Fields left blank, values typed as free text, and timestamps that do not match the order of steps people describe are all evidence.
Keep the evidence. The scorecard below is only as credible as what sits behind each answer.
Step 3: Assess the six areas
Score the workflow in six areas. Each one maps to a way agents fail in live operations, and each is covered in more depth in Process Audit Red Flags That Signal Automation Will Fail.
1. Process consistency
Can every person who runs the workflow describe it the same way, and do identical inputs produce identical outputs? If two people run two different versions under one name, there is no single process for an agent to follow. It will encode one version and break on the other's cases.
2. Decision rules
List every decision point in the workflow. For each one, is there a rule someone can write down, or does it resolve as "it depends" or "ask someone"? Judgment that lives in one person's head is not something an agent can inherit. Either the rule gets made explicit, or the decision stays with a person by design.
3. Exceptions
How are cases that do not fit the standard path handled, and how often do they occur? An exception path that exists in practice but has never been mapped is where an agent will fail without anyone noticing, because its output will look normal.
4. Data
Are the inputs digital, structured and accessible today? Can each value be traced back to its source? Inspection results in PDFs, approvals in email threads and schedules that record what was planned rather than what happened are all inputs an agent cannot use as given.
5. Integrations and audit trail
Which systems does the workflow read from and write to, and has each connection been tested rather than assumed? After a decision, can you reconstruct why it was made, what action was taken and who reviewed it, from records that reference each other? In regulated or safety-critical work this is a deployment requirement, not a nice-to-have.
6. Ownership and measurement
Who owns the outcome when the agent gets it wrong? Where do the human review steps sit, and are they placed at decisions with irreversible consequences? Is there a baseline for the number the deployment is meant to move, measured before anything changes? Without a baseline you cannot show the return once the agent is live.
The workflow readiness scorecard
Use one row per area. Mark each area ready only when the evidence from Step 2 supports it, not when someone says it is so. Any area marked not ready decides the result on its own.
| Area | Question | Ready | Not ready |
|---|---|---|---|
| Process consistency | Do separate walkthroughs by the people who run it describe the same steps in the same order? | Accounts match; differences are minor and documented. | People describe different sequences, owners or end points. |
| Decision rules | Can every decision point be written as an explicit rule or a named human review? | Rules are written down, or the decision is deliberately kept with a named role. | Decisions resolve as "it depends" or "ask someone". |
| Exceptions | Are exception paths mapped, and is there a record of how often each occurs? | Exceptions are listed, routed and counted. | Exceptions are handled informally and nobody can say how often. |
| Data | Are the inputs digital, structured, traceable and reachable today? | Inputs sit in systems the agent can query, with known sources. | Inputs live in PDFs, email, side spreadsheets or need a migration first. |
| Integrations and audit trail | Have the system connections been tested, and can each decision be reconstructed afterwards? | Connections tested on real data; reasoning, action and review are linked. | Connections are assumed; records sit in systems that do not reference each other. |
| Ownership and measurement | Is there a named owner, a placed human review, and a baseline for the target metric? | Owner, review points and a measured baseline exist before deployment. | No owner for wrong outputs, or no baseline to compare against. |
Step 4: Read the result as one of three calls
The point of the assessment is a decision, not a score. Every workflow lands in one of three places:
- Ready to deploy. All six areas are supported by evidence. Design the agent's role around the process as it actually runs, with human review at the decisions that cannot be undone, and measure against the baseline from day one.
- Repair first. One or more areas are not ready, but the workflow is still the right one to start with. The problem is a repair signal, not an automation signal: fix the inconsistency, write down the decision rule, map the exception, then reassess. Often the repair is a rebuild of the workflow rather than an automation of it.
- Choose a different workflow. The workflow fails on data or structural soundness in a way that cannot be fixed inside a reasonable first project, such as a data migration or a pending system replacement. Move to the next candidate rather than turning the first deployment into a data project.
Write the call down with the evidence for each area. That record is what lets the next workflow be assessed faster, and it is what you will need when someone asks later why this workflow went first.
If you bring in outside help
Firms and vendors offer AI readiness assessments under many names. Before you hire one, ask:
- Will they assess a named workflow, or survey the organisation?
- Will they observe the work and pull exception and system records, or work from interviews and documents only?
- Is the deliverable a deploy, repair or choose-another call per workflow, with the evidence behind it, or a maturity score?
- Are they independent of the software they might recommend afterwards?
Ask for these answers, the timeline and the price in writing. Finding the firm that maps process breaks before any AI decision is made covers what separates the two kinds of engagement.
Frequently asked questions
What is an AI readiness assessment?
A structured check of whether an organisation, or a specific workflow, can support an AI system reliably. For operations teams the useful version assesses one workflow as it actually runs: its consistency, decision rules, exceptions, data, integrations and ownership.
How is it different from a process audit?
A process audit describes how a workflow runs and where it fails. An AI readiness assessment uses that evidence to answer one further question: can an agent operate inside this workflow now, does it need repair first, or should a different workflow go first?
Who should run it?
Someone who will look at the work itself rather than only the documentation, with access to the people who run the workflow and to its system data. The findings should reach an executive outside IT, so operational problems are not filtered into a technology-only view.
How long does an AI readiness assessment take?
It depends on the size of the workflow, how many people run it and how easy its records are to pull. Scope it to one workflow so it produces a decision rather than a report, and ask anyone offering to run it for the timeline and deliverables in writing before you start.
Can we deploy if one area is not ready?
Not into that workflow as it stands. An area that is not ready is a defect the agent will reproduce at scale. Repair it first, or keep that part of the work with a person by design and deploy only where the evidence supports it.
Read next
- What a Pre-AI Process Review Actually Looks Like for Logistics and Construction Operators
Operators must fix broken workflows before deploying AI agents into them.
- Process Audit Red Flags That Signal Automation Will Fail
A pre-automation audit can expose the hidden defects that automation will lock in.
- Repair Signal vs. Automation Signal How Operations Data Tells You Which Category a Process Falls Into
Distinguish repair signals from automation signals before deploying AI.
- Prioritizing Which Process to Redesign First
Choosing the wrong first process to automate sets back your entire program.
- Pre-Automation Audit Checklist for Logistics Operations Leaders
A four-part audit shows which workflows are ready to automate.
- Rebuilding vs. Automating Existing Workflows
Automating a broken process just makes it fail faster.