Skip to content
Callmenta
CHANGELOG

Callmenta keeps evolving

Callmenta keeps evolving. See what changed in each release.

CALLAII versions and improvements

v4.11November 2026

No missing text displayed for fully scored headings Bug fix

  • In some calls, a heading would receive a full score while still displaying missing text underneath. Headings that receive a full score or are not measured in that conversation no longer display missing text. The score remains unchanged.

Score in coach comment matches the score on screen Bug fix

  • When the score changed after a company interruption, objection, or re-analysis, the General Score in the coach comment could remain at its old value. It now syncs with the score on screen with every change.

Meaningless lines are filtered out from meeting transcripts Bug fix

  • Meaningless lines, such as language names appearing in silent sections, are no longer written to the transcript.

Ungraded meeting report now displayed together Enhancement

  • In meetings where the sufficient section could not be graded, the report no longer disappears. In the panel and summary post, it now displays that it could not be graded and the reason, instead of a score. Scope information from the integration event is also included in the general criteria report.

Report API Scope Rules Enhancement

  • Report endpoints view the connection as a company administrator: reports generated by the internal team and consultant reports are not visible. Call X-Ray is visible only if it has a single company signature or if its signature exactly matches the connection's company scope. Branch-limited connections cannot access reports, as reports are company-wide. The summary field does not include internal keys or the text engine name. The group scope and company_id field of the meeting query have also been added to the documentation.

Report Center reports can now be retrieved via API New feature

  • You can automatically pull reports generated in Report Center into your own system. Manually generated, subscription-based scheduled, and custom-defined reports are listed, and PDF, Excel, or presentation files can be downloaded. You can also view the scheduling of your custom reports from the same location.
  • The read key you use for analysis queries is also valid here. Reports are still generated from the panel, and ready reports appear in the API. Details are in the "Pulling reports" section on the developer page.

Meeting analyses for all brands in group integration Bug fix

  • In groups with multiple brands, call analyses for all brands could be retrieved with a single integration key, but meeting analyses were only available from the brand where the key was activated. Now, meeting analyses are also retrieved from all selected brands. Each meeting recording also indicates which brand it belongs to.

Queued meeting records no longer time out Bug fix

  • During peak hours, meeting records waiting in the queue for more than six hours could be closed as errors without being processed. The waiting time has been temporarily extended to 24 hours. A permanent fix will be released separately.

Transcription correction is no longer skipped during peak times Bug fix

  • During peak hours, re-listening to suspicious lines in the transcription could be skipped due to time limits. The wait time has been extended, and correction is now performed for every call.

Call analyses complete faster during peak hours Enhancement

  • The number of call analyses processed simultaneously has been increased. Accumulated calls are now analyzed approximately 20% faster per hour.

Warning issued before downloading petitions with missing substantive sections Enhancement

  • If the substantive section in defense and response petitions is not written or left blank, the petition is marked as incomplete and cannot be downloaded without approval. If there is no procedural objection, the absence of a procedural section is not considered a deficiency.
  • For other petition types, the same warning is issued if the section containing the description of events is not written at all.

Paragraph numbering and repetitions are corrected in defense and response petitions Bug fix

  • In petitions written with headings such as procedural objections, substantive defenses, or grounds for appeal or cassation instead of the "Explanations" heading, paragraph numbers could skip, and the same fact could be repeated across different paragraphs.
  • Now, paragraph numbers are sequentially ordered from beginning to end in these petitions, and repeated sentences as well as the redundant recitation of the same legal article are filtered out. This fix covers all types of petitions, including defense, response, appeal, cassation, objection, and similar documents.

Redundant sentences in petitions are being removed Bug fix

  • In petitions containing multiple requests, identical content could be repeated across different paragraphs with minor variations. In petitions combining divorce, custody, alimony, and compensation claims, this unnecessarily lengthened the text.
  • Now, three types of redundancy are removed from petitions. Sentences that restate a previously described fact using different words are removed. Repeated identical evidence sentences at the end of each paragraph are kept only once. If the same legal article is quoted twice, the second quotation is converted into a reference to the article.
  • What was done is shown in the audit findings. Sentences containing new information, different amounts, or legal grounds are left untouched.

Source and Ad Campaign Breakdown in the To-Be-Called Lists Report Enhancement

  • The To-Be-Called Lists report now shows where records originate. Records from Meta ads, live chat, integrations, and file uploads are counted separately. Records from Meta ads are further broken down by campaign. This allows you to see in a single report how many records each campaign contributed to the call list.

Meeting report score not given based on a small portion of the recording Enhancement

  • There is now a minimum coverage rule for meetings scored against your company's rubric. The report provides a score if at least half of the applicable headings in that call could be scored. If less than half could be scored, no score or score range is displayed, and the report states "recording headings were insufficient for evaluation".
  • Previously, when a recording contained only a small portion of the call, scores from a few headings appeared as the score for the entire call. This could lead to both unfairly low and unfairly high scores.
  • Scoreable headings continue to display their scores and justifications. Headings that are not applicable to that call, such as due to the absence of a sales signal or objection, are not included in this calculation.
  • Coverage information showing how much of the call the score is based on has been added to the meeting analysis notification.

Meeting scorecard no longer double-penalizes the same missing item Bug fix

  • In your company's meeting criteria table, "should do" checks linked to a heading no longer deduct points separately. The heading itself measures that behavior, and if it is missing, the heading already loses points.
  • Previously, when a heading received partial credit, the check for the same behavior could also deduct points separately. This caused some call scores to be significantly lower than they should have been.
  • The check result continues to appear on the scorecard. "Should not do" checks, such as promising a definitive outcome or unauthorized redirection, still deduct points as before.
  • The recording notification check has been expanded to recognize phrases like "we are recording".

Scorecard headings in integration output Bug fix

  • If you are transferring meeting analysis to your own system, the output now includes the score received for each heading on the scorecard.
  • If your company has its own criteria table, each heading name, weight, score, contribution to the total score, and rationale are included. If a heading is not applicable to a specific meeting or there is no section to measure in the record, this is explicitly stated. The table name, score range name, and any score deductions are also added.
  • For companies without a defined criteria table, six general criteria are now included with their names. Previously, the scores for these criteria were provided, but it was not specified which criterion they belonged to, and for meetings scored with a company table, the heading list was empty.
  • Existing fields have not changed; only new fields have been added. The examples and descriptions in the developer guide have been updated.

Incomplete meeting recordings are not scored Bug fix

  • Meetings where the majority of the conversation could not be evaluated due to connection issues or very short recordings are no longer scored. Previously, a recording where only one of the six criteria could be evaluated could receive a score of 100 based on that single criterion, which skewed the agent's average.
  • For meetings evaluated using the general criteria, a score is only awarded when at least four of the six criteria can be evaluated. There are no changes for meetings scored using the company's own evaluation table.

Zoom recordings are processed without delays on busy days Bug fix

  • Zoom recordings sent from your own system are now downloaded immediately upon arrival. Previously, when many sessions were sent simultaneously, downloads would queue up, and recordings could not be retrieved because Zoom's download permission expired.
  • Three sessions are processed concurrently. The summary and report for a completed session are prepared before other sessions waiting in the queue. Long sessions are completed without being interrupted.
  • If the download permission expires, the recording immediately transitions to a "download failed" status. When you resend the same recording ID with a new permission, it is processed.

Only tasks achievable in short calls are scored Enhancement

  • Very short calls were previously scored as if they were full sales calls. If a customer declined at the beginning of the call, the agent could receive a very low score for not asking about needs or closing the sale. Three scenarios are now evaluated separately.
  • If the customer stated they did not want the offer at the beginning of the call, needs assessment, presentation, and closing are not included in the score. Introduction, listening, language, professionalism, and asking for the reason for the refusal are scored. The label "Declined at the start of the call" appears on the call screen.
  • If the agent promised a callback and the customer confirmed it, the call is considered postponed and scored according to postponement rules.
  • If an outbound call ended during the introduction phase and the customer did not engage in conversation, the call is not scored. These calls are not included in the agent's average or coaching suggestions. The change applies to new analyses.

Line endings no longer dropped from transcript during re-listening Bug fix

  • Lines that appear suspicious in the transcript are re-listened to from the audio and corrected if necessary. The second listening sometimes only captured the beginning of the line. In this case, the line was replaced with the audible portion, and the rest of the sentence was dropped from the transcript. For example, the agent's recording notification or payment question said after the greeting was lost. Now, if the second listening only captures part of the line, the line is not shortened and remains as is. Genuine corrections are applied as before. This change applies to new conversations and conversations where the transcript is re-generated.

Verified words from re-listening are retained in the transcript Bug fix

  • Lines that appear suspicious in the transcript are re-listened to from the audio. Until now, even if the second listening heard the same words, the line would be displayed as "[unintelligible]" and excluded from evaluation if a neighboring word was mixed in or if the speech recognition confidence was slightly low. As a result, sentences actually spoken by the agent, such as opening introductions, recording notifications, or price explanations, were sometimes dropped from the transcript. Now, lines re-heard during the second listening are retained in the transcript and included in the evaluation. Lines where no audio is heard or where something entirely different is heard are still displayed as "[unintelligible]" as before. Known speech recognition hallucinations are also filtered out separately. This change applies to new conversations and conversations where the transcript is re-generated.

Unspoken words are not displayed in quotes in coaching comments Bug fix

  • The coaching comment could previously quote a sentence from the agent or customer that was not actually said during the call. Now, every word quoted in the "by saying..." format is compared against the call recording. Words that do not have a corresponding match in the recording are not displayed in the comment, while the rest of the sentence is preserved.
  • Suggestion sentences and required notification text are excluded from this check and remain unchanged. Coaching comments from the last month have also been reviewed with the same check.

Misheard words are corrected in the transcript Enhancement

  • Speech recognition sometimes mishears words on noisy lines, leaving nonsensical spellings like "süreşte", "mücredat", or "ücredi" in the transcript. These words are now corrected to their proper forms based on the conversation flow: "süreçte", "müfredat", or "ücreti". The correction works cautiously. A spelling that almost never appears in your conversations is only replaced if it closely resembles a word you frequently use. Names, numbers, amounts, and the positive or negative tone of sentences remain untouched. If uncertain, the transcript is left as is. The initial output of speech recognition is also saved. Changes apply to new conversations and re-evaluated conversations.

Speaker separation improved in single-channel recordings Enhancement

  • In calls where the agent and customer are recorded on the same channel, a few words at the point where the turn-taking changes were sometimes attributed to the wrong party. The ownership of a word is now determined with greater precision based on audio analysis.
  • Measurements conducted on our own calls showed that the number of words attributed to the wrong person decreased by approximately one-fifth. This change applies to newly analyzed calls.

Evidence verification tightened in criteria and coaching quotes Enhancement

  • Evidence indicating a criterion violation and the representative quote in the coaching plan are now more strictly compared against the actual sentence in the conversation. Short responses like "Yes" are no longer considered to validate long quotes. The coaching plan always displays the exact sentence from the conversation as the quote.

General score in coach comment matches the score on the screen Bug fix

  • The general score at the beginning of the coach comment could previously display the score before company scoring rules and notification audit were applied. In new comments, the general score is the same as the score on the call screen.
  • Additionally, the coach comment does not write missing or development area for steps where the conversation did not occur.

Coaching and reports now only consider genuinely measured skills Bug fix

  • When a call was postponed, interrupted, or the wrong person was reached, steps like needs analysis, presentation, and closing were already excluded from scoring. However, coaching plans, skill averages, and quality reports treated these steps as if they had been measured, making an untested skill appear as the agent's weakest area.
  • Now, the coaching focus, target measurement, quality trends, manager summary, and score panel only account for skills actually evaluated in that call. The coaching plan now directs the agent to the areas they genuinely need to improve.

Repetitive confirmation words and times in transcripts are no longer lost Bug fix

  • When preparing call transcripts, the repetition cleanup was removing too much content. If a customer said "Yes" or "Okay" a second time during the call, that word was dropped from the transcript, and repeated times and numbers for confirmation were reduced to a single instance. As a result, the evaluation sometimes could not see the customer's confirmation or the agreed-upon time. Now, only places where speech recognition gets stuck and writes the same word repeatedly are filtered out. Every word actually spoken in the call retains its place in the transcript and summary. This change applies to calls transcribed from now on.

Unnecessary warnings reduced in call summaries Enhancement

  • When a customer explicitly declines an offer or states they have agreed with another institution, the summary no longer displays the "next step not determined" warning. In such a conversation, there is no next step to plan.
  • For postponed, interrupted, or calls where the right person could not be reached, the summary no longer lists steps that were not expected in that specific interaction, such as "fee not discussed, presentation not made." Genuine gaps related to opening and confirmation remain in the summary.

Words starting with capital İ are now correctly recognized in call analysis Bug fix

  • Words such as "İstemiyoruz", "İlgilenmiyoruz", and "İki gün sonra" spoken by customers at the beginning of a sentence, as well as the farewell "Good afternoon", were not reflected in the evaluation in some calls. As a result, rejections or postponements were missed in a small number of conversations, and a normally completed conversation could be counted as interrupted. These words are now recognized as if they were written in lowercase.
  • Sentences like "Şimdi Instagram'da gördüm" and "şimdi İngilizce dersi var" are no longer interpreted as the customer being busy at that moment. Only genuine obstacles such as "şimdi iniyorum" and "şimdi toplantıya giriyorum" are taken into account.
  • In follow-up calls, the sentiment of a customer who politely declines an offer is now shown as "neutral" instead of "positive". Thanking someone does not imply satisfaction.
  • Affected conversations have been re-evaluated.

Polite rejections by parents in follow-up calls are now interpreted more accurately Enhancement

  • In follow-up calls with previously contacted parents, a brief thank-you response to the decision question is no longer counted as a positive answer. The analysis now interprets such responses as the parent not accepting the offer and does not expect a needs question or analysis explanation in this call.

Call scores are now fairly calculated when the line drops Bug fix

  • When a call is interrupted due to a line drop or the other party hanging up mid-conversation, the analysis now recognizes this from the audio. These calls are no longer evaluated as deferred conversations. Agents are not penalized for steps that could not be completed after the interruption. Missing next steps and appointments are also not recorded. The score is awarded based on the greeting and communication performed up to the point of interruption.

New Report: Complaint Risk Report New feature

  • The Complaint Risk Report has been added to the Report Center. It compiles calls identified as complaints in the CALLAII analysis into a single document. The figures in the report match the Complaint Risk card in the Call Analysis Center.
  • The report includes the number and percentage of calls with complaint risk, a comparison with the previous period, calls that need to be escalated to management, the number of affected customers, and the number of customers who have filed multiple complaints. It also features distributions by day or week, agent breakdowns, frequent topics, and a call list sorted by priority. If the complaint module is enabled, you can also see how many records opened from these calls are currently open versus closed.
  • The report can be downloaded as PDF, Excel, or CSV, and can be sent automatically on a weekly or monthly basis. Phone numbers are displayed in masked format. This feature is available for companies with the conversation analysis package enabled.

v4.10November 2026

Search engine records for legal tools fixed Bug fix

  • Page information submitted to search engines by the UDF translator and ten accounting tools has been corrected. Fields flagged as errors by audit tools have been moved to the correct location. The appearance of the pages has not changed.

UDF converter page consolidated into a single source Enhancement

  • The three ways to convert a UDF file to PDF are now explained directly on the UDF converter page. The old blog post on the same topic has been moved to this page, and the old address redirects here. Users can find both the tool and the comparison on a single page.

Meeting scorecard automatically retries on interruption Bug fix

  • If the agent scorecard cannot be generated due to a brief interruption while the meeting summary is being prepared, the scorecard no longer remains empty. The summary is ready as usual, and the scorecard is automatically retried a few minutes later.
  • During the wait, the meeting screen indicates that the scorecard will be retried. When the scorecard arrives, it appears on the screen, and the meeting analysis sent to your connected system is also re-sent along with the scorecard. The summary email is not resent.
  • If it cannot be generated after three attempts, this is indicated on the screen, and you can try again by requesting the summary again.

Every score in the meeting report is verified with a quote from the conversation Enhancement

  • In Zoom meetings scored against your organization's criteria table, every positive score now relies on a direct quote from the conversation. The quote appears under the heading in the report, allowing you to see exactly what the score is based on at a glance.
  • If a quote cannot be found in the conversation transcript, that score is excluded from the total and marked separately in the report. This ensures the report does not come out higher than it should due to a narrative that did not occur in the conversation.
  • Behaviors not performed in the conversation are no longer left as "not evaluated" but are scored as missing. If the recording does not include the beginning of the conversation, behaviors during the opening are excluded from the evaluation.

Agent scorecard is fully scored in long meetings Bug fix

  • In long Zoom meetings, the agent scorecard was generated from a short meeting summary. Behaviors not visible in the summary, such as recording notifications or a used analogy, remained as "not evaluated" in the scorecard. In some meetings, the scorecard could not generate any score at all.
  • The scorecard is now generated using the full meeting transcript. Behaviors that were performed receive a score, while behaviors that were not performed appear as missing in the scorecard.
  • For very long meetings where the full transcript does not fit, the scorecard is still generated from the summary. In this case, behaviors not visible in the summary are not incorrectly counted as errors.

Second Draft of Unverifiable Citations in Legal Research Enhancement

  • If a legal research opinion cites a decision that cannot be found in the library or a provision that has been repealed, the text is rewritten once without those citations. The less problematic and more reliable of the two texts is displayed. This reduces the number of researches deemed unreliable due to unverifiable citations.

Morning operation alerts are now generated daily Bug fix

  • The preparation of morning threshold alerts for high call volume accounts was being interrupted. As a result, some accounts did not receive morning alerts for several days. The calculation has been optimized, and alerts are now prepared every morning. The call analysis summary on the dashboard also opens with fewer resources on large accounts.