Conflict Check Software for Small Law Firms: Where AI Helps

AI can do the searching: it spots near-matches such as "Katherine Smith-Jones" and "Kate Jones", pulls party names out of enquiry emails, and summarises each hit. It cannot decide whether a conflict exists. A partner must confirm the party list is complete, review every hit, sign off in writing, and keep the record.

The failure that matters is the match nobody sees. A search can only find what is in the database, and the vendors' own documentation says so. Smokeball's conflict report warns that it may not represent a complete conflict check, and one case management vendor's checker leaves closed matters out unless you tick a box. Treat any "no conflicts found" as a statement about your records, not about the world.

Follow me on Instagram@sagnikteaches

What a conflict check actually compares

A conflict check compares the people and organisations in a new enquiry against everyone your firm has recorded before: clients, former clients, opposing parties, related individuals, witnesses and anyone who consulted you and was turned away. The comparison is only as good as the record. Clio's own guide to conflict checks lists what to capture for individuals and businesses so that a close match can be cross-referenced. Those are the first two rows below; the last two are my additions, because a conflict can come from the role someone played or from work that has ended.

Connect on LinkedInSagnik Bhattacharya

The practical lesson is that the weakest part of most firms' conflict process is data entry, not searching. A party mentioned in an email but never added as a contact can't be matched by any software, however clever. Records that matter most are the ones people are least likely to type in: the other side's director, a relative who is also involved, the old trading name.

Subscribe on YouTube@sagnikteaches

Exact, normalised and fuzzy matching in plain English

Matching comes in layers. Each layer catches something the one above it misses, and each adds more false alarms for you to review.

Vendors describe their matching in broadly these terms. Clio's guide says its conflict check searches contacts, matters, notes, documents and calendar events across Clio Manage and Clio Grow in one search, runs up to ten checks at once, lets you customise the search depth and keeps PDF reports; ask Clio exactly what each search setting matches, such as nicknames and misspellings. Filevine says its checker searches first, middle, last, prefix, suffix and nickname fields for fuzzy matching. Access Legal's CaseMatters describes fuzzy logic for nicknames, abbreviations and alternative spellings, and scores each result as a probability, with default bands of 75% for high and 50% for medium that an administrator can change.

None of those pages calls the matching AI. They describe fuzzy matching, which is older and more predictable than a language model. That matters for what you can promise a partner: a fuzzy score is the same every time you run it, while a language model's reading of a messy note can vary.

Where AI adds something beyond fuzzy matching

Language models are good at reading messy text and weak at being a source of truth. That points to five jobs around the edges of the search, each of which you can test with an assistant inside your practice management system, or ask your vendor whether it offers.

Pulling the parties out of a messy enquiry

Intake emails and call notes bury the names that matter. An assistant can turn one into a structured party list and, just as usefully, say what is missing. These examples use invented content.

Enquiry (input)
Hi, I run a small print shop and want to take action against my former supplier, Brightline Papers (I think it is now part of a bigger group?). My business partner, [first name], is involved too. The supplier's director signed the contract.

Party list (illustrative AI output)
Prospective client: sender's print shop (legal name not given)
Opposing party: Brightline Papers (possible group parent, unconfirmed)
Related party: business partner [first name] (surname missing)
Opposing director: name not given
Missing before searching: client's legal name, the supplier's legal entity, the partner's surname, the director's name

What you would fix: nothing in the list itself, but the missing items mean the search shouldn't run yet. The risk is an assistant that fills the gaps with a guess, so check that it says "not given" rather than inventing a surname.

Clustering records that look like one person

Three contact records can describe the same person: "K. Jones, 14 Mill Lane", "Kate Jones, 14 Mill Ln" and "Katherine Smith-Jones, 14 Mill Lane". An assistant can propose that they are one individual with a married name, using the shared address as supporting evidence. It should propose, not merge. A human confirms before any records are combined, because a wrong merge can hide a real hit or create a false one.

Reading notes for relationships that never became fields

The relationship that creates a conflict is often in free text: "client's brother-in-law runs the supplier". No contact field holds that, so a name search misses it. An assistant that reads matter notes can surface sentences like that when you search for a surname. Two cautions apply: it only helps if the tool is allowed to read the notes, and it must respect matter permissions. Smokeball's conflict report shows a warning when documents or memos are hidden from the searcher by permissions, which is exactly the behaviour to look for.

Summarising each hit for the partner

A partner reviewing 14 hits needs one line each, not 14 matter files. A useful summary names the matter reference, status, the firm's role and the reason for the hit, as in this invented example.

Hit 3 of 14
Matter 2019-044, closed 2021. The firm acted for Brightline Papers Limited (supplier side) in a lease dispute. Same address as the new opposing party. Name differs: "Brightline Papers Ltd" against "Brightline Paper Co". Likely the same business.

The summary saves time only if the partner can open the original matter in one click and see that the summary is accurate. Treat the summary as a pointer, never as the evidence.

Drafting the memo and the letters

Once a partner has decided, an assistant can draft the conflicts memo for the file, a non-engagement letter, or a request for consent from affected clients. These are routine drafting jobs with a human reading every word. For the engagement-letter side, see the AI-use clause clients read in a law firm engagement letter.

Where AI must not decide

Everything above is retrieval and reading. The decision is law and judgement, and it stays with a person. Your conduct rules and your regulator's guidance govern each of these, and a qualified adviser should be asked when the answer isn't clear.

• Whether a relationship is a conflict at all, including direct adversity and a duty to one client that limits the firm's duty to another.
• Whether a former client's closed matter is close enough to the new one to matter.
• Whether informed consent is possible, what it must say, and whether a screen between lawyers is acceptable.
• Whether a conflict carried by one lawyer, such as a new hire from another firm, applies across the whole firm.
• What duties are owed to a prospective client you turn away, and what you may say to them.

The most dangerous AI output is a fluent "no conflicts found". A reader assumes a search ran across everything. Require every result, including a clear one, to show the parties searched, the settings used, the records covered and the time. Access Legal's article says a PDF report should be generated at the conclusion of every check and that the exported report includes clear checks. It also says a name-only search is a starting point, and that if a high conflict alert appears you shouldn't create the client record.

One enquiry through a six-lawyer firm's records

A six-lawyer firm holds 5,200 contacts and 1,450 matters, about a quarter of them closed. A print-shop owner phones about a dispute with a former supplier, and the intake handler records four parties: the owner's business, the owner, the supplier company and the supplier's director. Every number here is invented.

Three hits are real. The owner is a former client on an unrelated matter, which is noted and raises no issue. The supplier company is a former client under a slightly different trading name, in a closed matter. The supplier's director appeared as a witness in another closed matter. An exact search finds only the first, a normalised search finds the first two, and the third needs fuzzy matching plus the setting that includes closed matters. A firm relying on exact search would have cleared the enquiry having found one hit out of three, and missed the one that matters most: a former client now sitting on the other side.

The tool saves 10 minutes per enquiry against the spreadsheet and AI summaries save 23. At 30 enquiries a month that is 11.5 hours. But watch the review row: fuzzy matching produces false alarms, so review takes longer with a plain search tool than with a spreadsheet (21 minutes against 14), and summaries bring it back down. The risk is fatigue. When 11 of 14 hits are noise, people start skimming, so keep the triage log, not just the survivors, in front of the partner.

What the audit record should hold

A record that survives a complaint says who searched, for what, where, what came back, who decided and why. Several vendors document reports along these lines, which is a good benchmark for whatever you use.

• Smokeball says its saved conflict check report has a cover page with the name of the user who ran the search, the matter the PDF was saved to, a timestamp, any notes added when saving and a linked summary of totals. The remaining pages hold every result shown in the search. If matter permissions hide documents or memos from the searcher, the report carries a warning that results may be incomplete.
• Filevine says exported PDFs include who ran the check, the date and time, the search terms, and each result's roles and project numbers. Once a result is marked undecided, conflict or no conflict, the details of project-related results are locked to capture their status at that moment.
• Clio says its conflict check records outcomes for an audit trail and generates PDF reports; its guide describes re-running searches at any stage of a matter.

Those are the vendors' descriptions, so check them on your own screen. A template that works regardless of software:

Conflicts record
Enquiry reference: [ref]        Date and time: [date, time]
Run by: [name]                  Systems searched: [practice management, intake CRM, email archive]
Parties searched, every spelling: [list]
Settings: [exact or flexible]; closed matters included [yes or no]
Hits reviewed: [number]. For each: [matter ref], [disposition], [reason]
AI used for: [extraction, summaries, none]; output checked by [name]
Decision: [clear / clear with consent / decline]
Signed off by: [partner name] on [date]
Follow-up: [engagement letter / non-engagement letter / consent letters sent]
Re-run triggers: [new party added, new related person, change of sides or scope]

Notice the AI line. If an assistant extracted the parties or summarised the hits, the record should say so and name the person who checked its output. That protects the firm if the assistant ever turns out to have missed something, and it makes clear that the sign-off was human.

Test your own tool in 20 minutes before you trust it

Vendor pages describe matching in general terms, so build a small test set from your own records. Plant 12 variants of names already in your database, run each through the search, and record whether it was found. These sample strings show the idea, and the results column is invented.

In this invented run, the exact search found 1 of 12 and the flexible search found 8 of 12. The four misses were not matching failures. Row 8 needed an alias field nobody had filled in. Row 10 needed a link on address, which the tool didn't support. Row 11 was in a note the search didn't read. Row 12 was a closed matter excluded by default, the very setting one vendor's article warns about. Each miss has a fix that has nothing to do with AI: add alias fields, add a manual note search to the checklist, and switch closed matters on.

Eight questions to ask a vendor about conflict matching

1. Which records does it search? Contacts, matters, notes, documents, emails, calendar entries? Clio's guide lists contacts, matters, notes, documents and calendar events; get the complete list for the plan you would buy.
2. Are closed matters and declined enquiries included by default? One vendor's checker excludes closed matters unless you tick a box, so ask where that switch lives and who can change the default.
3. Which matching methods are used? Exact, normalised, nickname, phonetic, edit distance, and alias fields. Ask for a demonstration with your own tricky names, not the vendor's sample data.
4. Can we see and change the score threshold? Access Legal's article describes default bands of 75% and 50% that administrators can change. A threshold you can't see is a miss rate you can't manage.
5. Is every search saved automatically? With the user, time, search terms and results, including clear results, and in a form you can attach to the matter.
6. How do permissions affect results? If a matter is hidden from the searcher, does the search silently skip it or warn that results may be incomplete? Smokeball says its report shows a warning, which is the behaviour to look for.
7. Can several parties be searched together and re-run later? Clio's guide describes running up to ten checks at once and re-running at any stage of a matter.
8. Is AI involved, and where does the data go? Ask whether any step uses a language model, whether it can be switched off, what is sent to model providers, and whether the vendor trains on your data.

One plan note: Clio's pricing page lists conflict check on Core, Signature and Elite and not on Starter, so a firm that wants Clio's checker has to budget for a plan that includes it. For how the main platforms compare more broadly, see Clio, Smokeball and Lawmatics compared.

Fitting the check into intake

The check works only if the party names reach it. A sequence that holds up in a small firm runs like this.

1. Capture the parties at first contact, with the fields from the table near the top required rather than optional.
2. Run the check before a consultation is booked. Clio's guide makes the same point, and it avoids spending a lawyer's hour on an enquiry the firm can't take.
3. Triage the hits: obvious false alarms closed by the person who ran the search, with a reason, and everything else passed to the partner.
4. The partner decides and signs the record.
5. File the record against the matter or the declined enquiry.
6. Re-run when a party or related person is added, when a new lawyer joins, and before the engagement letter goes out.

If intake runs through a chatbot or voice agent, its job is to collect names, not to reassure. It must never tell a caller there is no conflict, because only the completed check and the partner's sign-off can say that. What to compare in legal client intake software and AI client intake for law firms cover the capture side. And before anyone pastes party names into a general chatbot to speed things up, read whether solicitors can use ChatGPT without breaching confidentiality, because the names of clients and opposing parties are confidential information.

Conflict checks and AI: questions firms ask

Can AI replace the partner's sign-off?

No. Software and AI can search records, rank near-matches and summarise what they found, but whether a relationship amounts to a conflict, and whether consent or a screen can cure it, is a judgement under your conduct rules. The sign-off belongs to a named partner or the responsible lawyer, recorded in writing with the date and the hits reviewed.

When should a conflict check be re-run?

Run it before a consultation is booked and again before the engagement letter goes out. Re-run it whenever a new party or related person is added to the matter, when a new lawyer joins who may carry conflicts from a previous firm, and when the matter changes sides or scope. Clio's guide describes re-running searches at any stage of a matter.

Do declined enquiries belong in the conflicts database?

Usually yes. People who consulted you and were turned away may still be owed duties under your conduct rules, and they may appear as opposing parties later. Record the enquiry, the parties and the outcome, then check your regulator's guidance on how long to keep those records and what duties remain.

Is a spreadsheet enough for a tiny firm?

It can work while one person owns it and the list is short. It tends to fail on spelling variants, on closed matters nobody added, and on proving who searched what and when. Practice management software adds the search and the saved report, but only if you record every party in it as you go.

Can I paste client names into a general AI chatbot?

Only if your confidentiality duties, your clients' terms and the tool's data terms allow it. Names of clients and opposing parties are confidential information. The safer route is AI inside your own practice management system or a business plan whose terms exclude training on your data, with the permission and logging controls checked first.

Further reads

• AI Voice Agent for Lawyers: Cost and Safe Handling of Intake Calls — Safe handling of intake calls before names reach the conflicts search.
• Family Law Client Intake: An AI Questionnaire and Urgency Triage — An intake questionnaire that captures parties and urgency.
• AI for Small Law Firms: What to Automate First — Where conflict checking sits among the first automations.
• AI Implementation Plan for a Small Law Firm: The First 90 Days — The first 90 days of adding AI to a small firm.
• AI Software for Solicitors: Data, Supervision and Supplier Checks — Data, supervision and supplier checks before you sign.
• AI App for Lawyers: Score Accuracy, Privacy and Cost in a Short Trial — A short scorecard for trialling any legal AI tool.
• E-Discovery Software for Small Law Firms: Costs, Limits and AI Review — Six e-discovery pricing models, one matter costed from your own data, and a sampling test for AI review, written for a firm with a handful of lawyers.
• Lawmatics Merlin: Is It Worth Adding Qualify, Engage and Copilot? — What Merlin Qualify, Engage and Copilot do for law-firm intake, what each costs, and a four-week test on real leads before you pay.
• Live Receptionist for a Small Law Firm: Cost and When AI Fits — What a live receptionist costs a small law firm against an in-house hire or AI, with plan prices, an intake script and an out-of-hours handover.
• Smokeball Pricing: What Bill, Boost, Grow and Prosper+ Include — Smokeball is quote-only. What Bill, Boost, Grow and Prosper+ include, which small firm each fits, and what to ask about onboarding and contract length.
• Family Law AI Software: What a Small Practice Should Check Before Buying — A scored checklist for a small family practice buying AI software: confidentiality, data location, conduct rules, citation checks and exit terms.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Smokeball support-hub articles on conflict check reports and contacts; Clio's guide to running a conflict check (updated 31 March 2026) and Clio's pricing page (copy dated 20 October 2026); Filevine's conflict checker page; Access Legal's CaseMatters help article on its conflict checker.