How Do You Find Patterns in Communication Records?
Find patterns in communication records by normalizing timestamps, resolving identifiers, establishing a baseline, and comparing changes in contact frequency, timing, and message content. Verify each finding against its source and test alternative explanations. In a communication analysis investigation, a recurring pattern is a lead to examine—not proof of identity, intent, or coordination.
Lantern Intelligence is an AI evidence platform for investigators. This guide explains a repeatable review method and where Lantern can support it without replacing investigator judgment.
What counts as a pattern in communication records?
A communication pattern is a repeated or meaningfully changing relationship among participants, timestamps, channels, and content within a defined set of records.
Useful patterns include repeated contact before an event, unusual changes in call frequency, recurring phrases across conversations, or the same identifier appearing in different evidence sources. Their value depends on context and record completeness.
Separate three layers in your notes:
- Observation: What the records directly show.
- Interpretation: What that observation might mean.
- Verification: What additional review could support or weaken the interpretation.
For example, “Account A contacted Account B six times during the review window” is an observation. “They coordinated an activity” is an interpretation. Message context, account attribution, and comparison periods are needed to evaluate it.
What should you check before comparing communication records?
Start with coverage, not counts. Two sources may describe different parts of the same communication activity, and neither may be complete.
Create a source inventory containing:
- File name and evidence reference.
- Provider, device, or other source.
- Available date range and known gaps.
- Timestamp format, timezone, and precision.
- Identifiers and communication fields present.
- Whether the source contains content, metadata, or both.
- Extraction, OCR, or formatting limitations.
A call record may show a connection attempt without establishing that a conversation occurred. A message export may include content but omit deleted material or activity outside the supplied period. Missing records are not automatically evidence of silence.
Keep originals unchanged and document transformations in a separate working copy. If you standardize dates or telephone numbers, retain the original values so another reviewer can retrace the comparison.
How do you normalize timestamps and resolve identities?
How should you handle time differences?
Use one comparison timezone while preserving each source timestamp. Record whether the source states UTC, a local timezone, an offset, or no timezone at all.
Check daylight-saving transitions and distinguish event types. Sent, received, delivered, read, and export timestamps describe different events. Sorting them into one sequence without labels can create an apparent order the records do not support.
Do not silently assign precision. A record dated only to the day cannot establish a minute-by-minute sequence. Where timezone information is missing, mark the uncertainty and test whether it changes the proposed pattern.
When can you treat two identifiers as one person?
An identifier is not a person. A telephone number, account ID, email address, username, and display name carry different attribution strengths.
Build an identifier worksheet with the identifier, proposed person, supporting source, relevant date range, and unresolved questions. A saved contact name is a lead, not independent proof of ownership. Shared devices, reassigned numbers, and changing usernames can complicate attribution.
Keep uncertain matches separate. Merging accounts too early can manufacture a high-frequency contact or make unrelated conversations appear connected.
Which communication patterns deserve closer review?
Compare each candidate pattern against an appropriate baseline: earlier activity, similar weekdays, or other periods with comparable source coverage. Avoid selecting only the window that makes a hypothesis look strongest.
| Pattern | What to compare | What to verify |
|---|---|---|
| Contact burst | Counts in comparable time windows | Duplicate records, retries, automated notifications |
| Repeated event proximity | Contact before or after independently documented events | Timezone accuracy and ordinary scheduling explanations |
| Channel change | Activity across calls, messages, and email | Whether each source covers the same period |
| Recurring language | Exact phrases and surrounding conversation | Quotations, templates, slang, and OCR errors |
| Apparent intermediary | Separate contacts involving a common identifier | Attribution and evidence of actual information transfer |
| Communication gap | Expected activity versus supplied records | Missing files, collection limits, and service changes |
Frequency alone does not establish significance. A frequently contacted account may be routine, while a single message may contain important context. Review volume and content together where available.
For a complementary approach to evaluating connections, see Link Analysis Law Enforcement Teams Can Use to Map Connections. Connections should remain distinguishable from conclusions about relationships.
What is a repeatable communication analysis investigation workflow?
Use a bounded question rather than starting with an unrestricted search for suspicious activity. For example: “Did contact between these identifiers change around the documented event, compared with otherwise similar days?”
- Define the question and window. Identify the event, comparison period, relevant sources, and limits of the review.
- Inventory and prepare the records. Preserve originals, identify coverage gaps, standardize working fields, and flag parsing or OCR issues.
- Resolve identifiers cautiously. Document confirmed, tentative, and unresolved associations without collapsing them into one category.
- Establish the baseline. Compare counts, timing, direction, and participants across equivalent periods. State what counts as one communication.
- Review candidate patterns in context. Read surrounding messages and inspect source records rather than relying on isolated search results.
- Test competing explanations. Look for duplication, routine schedules, automated traffic, unrelated group activity, and evidence that contradicts the working theory.
- Record the result and next step. Preserve source references, reasoning, uncertainty, and a specific follow-up question.
Define your counting rule before reporting totals. A group message, a multipart message, and several records describing one call can otherwise inflate comparisons. Distinguish raw rows from deduplicated events and explain the difference.
How can Lantern support communication record review?
Lantern accepts any evidence file, including warrant returns from any provider, phone extractions, PDFs, scans, photos, video, spreadsheets, email, and bank and call records. OCR turns scans, images, and handwriting into searchable text. Check important OCR-derived names, numbers, and phrases against the original image before relying on them.
Start in Upload Case, then use the available review areas according to the question. Live tabs include Dashboard, Messages, Persons, Media, IP / Login Activity, AI Insights, Files, Records, Search, and Export. AI Insights provides an intel brief and next steps; the “Ask the evidence” AI chat supports evidence-focused questions.
Useful prompts include “Which records mention this identifier?” and “What evidence contradicts this proposed sequence?” Treat AI output as a review aid. Verify statements and references against the underlying evidence before including them in a finding.
Timeline and Relationship graph are coming soon, not live features. For current workflows, consult how Lantern works and explore the use cases.
Lantern runs in private AWS or on-prem on Google Gemma 4; evidence is never sent to an outside AI service. CJIS, FedRAMP, and SOC 2 are coming soon and are not current certifications. Free sign-up includes 10 credits; see the pricing page for details.
What should a defensible pattern finding contain?
A useful finding lets another reviewer reproduce the observation without accepting your interpretation first. Keep the claim narrower than the evidence, particularly when attribution or source coverage remains uncertain.
Before sharing a finding, check that it includes:
- The precise question and review window.
- Source references and relevant record locations.
- Original identifiers and the basis for attribution.
- Timestamp normalization and counting rules.
- The observed pattern and comparison baseline.
- Supporting and contradictory material.
- Coverage gaps, uncertainty, and alternative explanations.
- A follow-up step tied to an unresolved question.
Prefer “Recorded contact increased during this window” to “The participants coordinated.” If message content supports a stronger interpretation, explain the specific connection rather than allowing a count or visualization to imply it.
Pattern recognition is most useful when it produces a testable question, a traceable observation, and a clear next step—not merely a persuasive story.
Put this to work on your own case
Sign up free and get 10 credits — enough to process a typical warrant return and see the messages, people, AI summary and recommended next steps.