Which legacy category is ripe for an AI rewrite?

01.The signal to look for

Ask a simple question: what do people in this system actually do for most of the day? If the answer is judgment, negotiation or creative work, an AI rewrite will help at the edges. If the answer is reading, copying, matching and routing, there is a lot to gain.

You can measure this. Shadow three users for an afternoon each. Count how many actions are copying data from one place to another, filling standard fields or choosing from a short list. That count is your opportunity.

02.Four categories to consider

• ITSM / help desk: repetitive tickets, clear categories. Help desks see the same requests every week: access, passwords, equipment, how-to questions. Categories are well defined, history is stored with outcomes, and many resolutions follow a script. This is often the best first target.
• CRM: manual logging, follow-ups, forecasting. Reps type call notes, update stages and write follow-ups. Agents can handle record updates, draft messages and flag deals that look stalled. Risk is moderate, because messages go to customers, so a rep should approve them.
• ERP back-office: invoice matching, approvals, reconciliations. Finance teams match invoices to purchase orders, chase approvals and reconcile accounts. The rules are clear and the volume high. Risk is higher because money is involved, so approval steps and audit trails are essential.
• HR and onboarding: document collection, policy questions. New joiners must submit documents, sign policies and ask the same questions. Agents can chase documents, check completeness and answer policy questions from the handbook. Data is personal, so access control matters.

03.How to score yours

Rate each candidate workflow on three factors.

• Volume of repetitive tasks. How often does the task occur, and how long does each take?
• Cleanliness of data. Are records complete, consistent and in a system an agent can reach?
• Risk of a wrong action. What happens if the agent gets it wrong? Can it be undone?

High volume, clean data and low risk go first. High risk is not a ban, only a reason to add approval steps and begin later.

04.An example scoring

A mid-sized company compares two candidates. Its help desk handles a large monthly volume with a short list of categories, mostly clean ticket history, and low-risk actions such as password resets. Its invoice matching has high volume too, but supplier data is messy and a wrong payment is costly.

The help desk scores well on all three factors, so it goes first. Invoice matching waits until supplier data has been cleaned, and starts in suggestion mode with human approval.

05.A short plan once you have chosen

• Write down the current workflow, step by step.
• Pull a few months of past cases as a test set.
• Run an agent in shadow mode and compare with human decisions.
• Add actions with approval, then automatic actions for low-risk cases.
• Review errors weekly and keep an audit log.

06.Looking at each category in more depth

ITSM / help desk. The strength here is the history. Years of closed tickets show which requests repeat, how they were solved and how long they took. That history works as a test set for an agent. The weakness is hidden variety. A ticket titled "cannot log in" may be a forgotten password, a locked account or a security incident. Good triage needs the agent to tell them apart, and to escalate when it cannot.

CRM. The strength is the volume of small actions: logging calls, updating stages, sending follow-ups. The weakness is data quality. Duplicate records, missing fields and inconsistent stage names confuse an agent just as they confuse people. Cleaning and merging records is often the first useful step.

ERP back-office. The strength is clear rules: a three-way match between order, receipt and invoice, or an approval limit by role. The weakness is cost of error. A wrong payment is hard to reverse, and audit trails matter. Start with matching and suggestions, and keep approval for payments.

HR and onboarding. The strength is repetition: the same documents and the same questions for every new joiner. The weakness is sensitivity. The data is personal, and a wrong answer on a policy question can cause real trouble. Limit access tightly and have the agent cite the policy it relied on.

07.A decision you will face: wrap or replace

For each candidate, you must decide whether to wrap the old system with an agent layer or replace it.

• Wrap when the system works, holds important history and has usable interfaces. You gain value quickly and keep the safety net.
• Replace when the system is blocking you, has no usable interface, or costs more to maintain than to rebuild.
• Do both in sequence when you can. Wrap first, learn what the agents need, and then replace the parts that hold you back.

Most companies find that wrapping is the right first move. A big-bang replacement has a way of taking longer than planned, and the old system must keep running meanwhile.

08.Measuring the result

Before you start, record a baseline. For a help desk, that might be time to first response and the share of tickets solved without escalation. For a CRM, it might be time spent on admin and the number of duplicate records. For ERP, it might be time to match an invoice. Measure again after each stage. If a number does not move, find out why before you move on.

09.A one-page scoring sheet

• Name the workflow in one line.
• Estimate how many times a week it happens.
• Estimate minutes per occurrence.
• Rate data cleanliness: poor, fair or good.
• Rate the cost of a wrong action: low, medium or high.
• Note whether an old interface exists for the agent to use.
• Rank the workflows. Choose the first one where volume is high, data is good and risk is low.

10.Common mistakes

• Choosing by excitement. The most visible system is not always the best candidate. Choose by score.
• Ignoring data quality. An agent cannot fix records that contradict one another. Clean what matters first.
• Skipping the people who do the work. They know the exceptions that never made it into documentation.
• Aiming for full replacement. Wrapping and augmenting an old system is often safer, and delivers value sooner.
• No baseline. Without a starting measurement, you cannot show that anything improved.

11.What to do next

Make a list of your four or five busiest internal systems. Score each on volume, data cleanliness and risk. Pick the top one and run a two-week shadow study.

Whichever category you choose, plan to keep a person responsible for the outcome. An agent can take over the typing. It cannot take over accountability. Name the person who answers for each workflow before the first agent goes live, and make sure they can see what the agent did and stop it when needed.

At Brandsmashers, we help teams assess and rebuild legacy tools in stages, starting with the workflows that score best. If you would like a second view on your list, we are happy to look at it with you.