The RPA Consulting Checklist: 30 Things to Verify in 2026

Buying robotic process automation help is easier when you have a list. This checklist distils the key points from TechCirkle's longer buyer's guide to robotic process automation consulting services into thirty yes-or-no checks, grouped by stage. Tick them off before you sign anything.

Before you talk to vendors

1. We have a written list of candidate processes with a named business owner for each.
2. We know the monthly volume and average handling time for each candidate.
3. We know roughly what share of cases are exceptions, and why.
4. We have checked whether any target system has an API.
5. We know which systems or policies will change in the next 12 months.
6. We have agreed who in IT signs off on bots touching production systems.

Scoping and discovery

1. The proposal includes a real discovery phase (two to six weeks), not just a workshop.
2. Discovery output is a ranked list with volumes, exception rates and payback estimates.
3. The vendor explains its scoring model and shows an anonymised example.
4. The vendor has recommended non-RPA solutions in roughly 20-40% of past discoveries.
5. The scoring model accounts for AI-assisted extraction and exception handling.
6. Low-volume candidates (under ~500 transactions or ~80 hours a month) are challenged.

AI and exception handling

1. The design shows where an LLM step handles exceptions the bot cannot.
2. Confidence thresholds for automatic fixes are written down per exception type.
3. Anything touching money or regulated data requires human approval.
4. Every AI decision is logged with inputs, output, model version and confidence.
5. A sample of automated AI decisions will be reviewed by people regularly.
6. AI failures and timeouts route to a human queue, never silently continue.
7. Model calls are processed in a region that satisfies our data rules.

Pricing and contract

1. We have a three-year total cost: discovery, build, licences, maintenance, run, model usage.
2. Annual maintenance is budgeted (typically 20-40% of build for screen bots).
3. Exceptions covered by fixed-price builds are listed explicitly.
4. Change request rates and process are agreed upfront.
5. The vendor discloses any resale margin or referral fees from RPA platforms.

Ownership and governance

1. We own the bot code, documentation and credentials vault from day one.
2. Each bot has its own service identity with least-privilege access.
3. Bots go through version control, peer review and change control.
4. A knowledge transfer plan with a dated handover milestone is in the contract.

Proof of delivery

1. The vendor shares real maintenance metrics: incidents per bot per month and repair time.
2. The business case assumes a realistic straight-through rate (60-90%), not 100%.

How to use this checklist in practice

Send items 7 to 30 to every shortlisted vendor as a written questionnaire before the final presentations. Ask for evidence, not just a yes: an anonymised discovery report for item 8, a sample exception-handling design for items 13 to 18, and a real metrics export for item 29. Vendors who answer in writing with specifics are usually the ones who will deliver with specifics.

Items 1 to 6 are your homework. If you cannot tick most of them, spend two or three weeks gathering that information first. Vendors will scope better, your quotes will be more comparable, and you will avoid paying a consultant to discover things your own team already knows.

How to score it

• 27-30 ticks: strong proposal; negotiate details.
• 20-26 ticks: workable, but push back on the gaps before signing.
• Under 20: you are likely buying bot builds without the run, governance or AI design to sustain them.

If you want a second pair of eyes on a proposal, TechCirkle's team can review it: contact them here. For the AI side of exception handling, see their LLM integration services.

Frequently Asked Questions

What is the most important item on an RPA checklist?

Real maintenance metrics from the vendor's live bot estates. They reveal whether the firm runs bots successfully or only builds them.

How do I check if a vendor is platform-neutral?

Ask directly whether they receive resale margin or referral fees from any RPA platform, and whether discovery ever recommends alternatives.

What straight-through rate should an RPA business case assume?

Typically between 60% and 90% depending on the process. Adding AI exception handling can push document-heavy processes toward the upper end.

Should AI be allowed to fix exceptions automatically?

Only for low-risk, high-confidence cases defined in writing. Anything involving money or regulated data should go to a person.

Who should own the bots after the project?

You should. Insist that code, documentation and credentials are yours from day one, with a formal knowledge transfer milestone.