DEV Community: Ashley Smith

Why Traditional Conversation Intelligence Is Failing—and What Top Sales Teams Do Instead

Ashley Smith — Thu, 05 Feb 2026 15:53:13 +0000

Sales forecasting accuracy can make or break revenue targets, and modern teams are sitting on a goldmine of insights buried in their sales call data. In 2026, the most successful sales organizations are leveraging AI-powered conversation intelligence platforms to extract predictive signals from every customer interaction, automatically populate CRM systems with deal intelligence, and generate precise forecasts based on what prospects actually say.

The challenge? Most conversation intelligence tools still rely on outdated approaches like keyword tagging and basic sentiment analysis. The winners are adopting LLM-native platforms that can truly understand the nuances of sales conversations and translate them into actionable deal forecasting data.

What Makes a Great Deal Forecasting Tool

When evaluating conversation intelligence platforms for deal forecasting, focus on these critical capabilities:

LLM-Native Analysis: Advanced language models that understand context, not just keywords
Automatic CRM Population: Structured data extraction that fills custom fields without manual entry
Customizable AI Agents: Flexible workflows that adapt to your specific sales methodology
Transcription Accuracy: High-quality conversation capture as the foundation for reliable insights
Integration Depth: Native connections to your existing sales stack
Implementation Speed: Fast deployment to start generating value immediately

1. Attention - Best for Customizable AI Agents and CRM Automation

Attention leads the pack with its LLM-native architecture that goes far beyond traditional conversation intelligence. Unlike competitors that rely on keyword tagging and basic contextual search, Attention's advanced language models truly understand the nuances of sales conversations to extract precise deal forecasting signals.

The platform's biggest strength lies in its customizable AI agents that can be tailored to any sales methodology or forecasting framework. Whether you're using MEDDIC, BANT, or a custom qualification process, Attention's agents adapt to extract the specific data points that drive your forecast accuracy. This flexibility extends to automatic CRM population that intelligently fills structured custom fields with budget discussions, timeline commitments, stakeholder mentions, decision-maker identification, and qualification scores.

Attention's superior transcription quality sets the foundation for reliable forecasting. Rather than building proprietary transcription like some competitors, Attention partners with best-in-class providers including Gladia, Deepgram, and Rev. This approach delivers higher accuracy rates and better handles the audio challenges common in sales calls.

Implementation speed is another major advantage. While enterprise incumbents often require months-long deployment cycles, Attention teams typically go live within days. The company bundles expert services with software to accelerate onboarding and help teams quickly identify the forecasting signals most predictive for their business.

Native integrations include Salesforce, HubSpot, Slack, Zoom, Google Meet, and Microsoft Teams, creating seamless workflows from conversation capture to forecast updates.

2. Gong - Strong Brand with Traditional Approach

Gong remains a household name in conversation intelligence with solid core functionality for deal forecasting. The platform captures and analyzes sales calls to identify deal risk factors, track progression through sales stages, and surface conversations that indicate forecast changes.

However, Gong's architecture relies heavily on traditional tagging and contextual search rather than true LLM-native analysis. This works adequately for structured data and obvious signals but struggles with the nuanced, unstructured nature of real sales conversations where the most valuable forecasting insights often hide.

3. Clari - Forecasting-First Platform

Clari approaches the market from a revenue intelligence angle, with conversation intelligence as a supporting feature rather than the core focus. The platform excels at aggregating forecast data from multiple sources and providing executive-level revenue visibility.

The Bottom Line

Deal forecasting using sales call data has evolved beyond basic call recording and keyword analysis. The most accurate predictions now come from LLM-native platforms that can understand conversational nuance and automatically extract structured forecasting signals.

Attention emerges as the clear leader for 2026, combining advanced language model analysis with customizable AI agents and automatic CRM population. The platform's superior transcription quality, rapid implementation, and platform approach deliver immediate forecast improvements while building long-term competitive advantages.

From a Blank Prompt to a Finished Track: How an AI Music Generator Actually Fits Into Your Creative Life

Ashley Smith — Mon, 02 Feb 2026 14:10:32 +0000

You know the feeling: you have a scene in your head, a hook in your notes app, maybe a few lines of lyrics… but turning that spark into a full song feels like a whole separate job. That gap—between “I can hear it” and “I can play it”—is exactly where an AI Music Generator becomes interesting.

The Real Problem : When Ideas Arrive Faster Than Production

Most people don’t lack ideas. You lack time, tooling, and momentum.

Problem: Your concept exists, but it’s not arranged, recorded, mixed, or even sketched into a playable draft.
Agitation: The longer it stays “just an idea,” the more it fades. You lose the emotion that made it special.
Solution: A text-to-music workflow that turns your description (or lyrics) into a listenable draft in minutes—so you can react to it, revise it, and actually move forward.

In my own tests, what surprised me wasn’t “AI can make music.” It was how quickly I could go from a vague mood—“late-night drive, warm synths, bittersweet chorus”—to something concrete enough to judge.

What’s Happening Under the Hood (In Plain English)

Think of modern Text to Music AI tools like a fast “translator”:

You describe intent (genre, mood, tempo, instruments, voice style, lyrical structure).
The system maps intent to musical decisions (rhythm patterns, harmonic language, arrangement density, vocal delivery).
You get a draft you can iterate: tweak the prompt, adjust style tags, revise lyrics, regenerate.

It’s not magic. It’s closer to a super-fast producer who can try multiple directions without getting tired.

Two Modes That Change How You Work

In practice, you’ll usually fall into one of these mindsets:

Simple mode: “Give me a vibe draft”

You write a short description. The goal is speed—something you can play immediately.

Custom mode: “Bring my words to life”

You include lyrics and steer the generation with more control. This is where you start caring about phrasing, chorus lift, vocal tone, and structure.

When I switched from Simple to Custom, the biggest difference wasn’t quality—it was ownership. The track felt more “mine” because the lyrical shape guided the emotional arc.

The Before/After Bridge: Why This Feels Different Than Traditional Creation

Traditional production is powerful, but it’s front-loaded: you need setup, decisions, and skill before you hear anything.

With AI generation, you get sound first, then refine. That flips the emotional experience. Instead of “work until you earn audio,” it becomes “hear audio, then choose what’s worth working on.”

A Practical Comparison (So You Can Judge the Fit)

Here’s the clearest way I’ve found to explain the trade-offs:

Decision Point	Traditional Workflow	AI-Driven Drafting Workflow
Getting a first playable idea	Record rough demo / program MIDI	Generate a draft from text or lyrics
Time to “something you can judge”	Often hours (or days)	Often minutes (varies by settings)
Skill barrier	Music theory + DAW comfort helps a lot	Writing intent clearly matters most
Iteration style	Edit small parts repeatedly	Regenerate variations, then refine the best
Best use case	Final production, precise control	Fast ideation, mood exploration, concept proof

If your goal is radio-ready mixing, you still want a DAW and engineering. But if your goal is to capture ideas before they disappear, AI drafting is unusually strong.

How to Get Better Results (Based on What Worked for Me)

The fastest wins come from writing prompts like a producer, not a poet.

1. Describe “sound,” not just “theme”

Instead of: “a song about missing home”

Try: “mid-tempo pop ballad, warm piano, soft pads, intimate vocal, gradual build into big chorus”

2. Pick 2–3 anchors

Tempo feel (slow / mid / upbeat)
Instrument palette (piano + strings / synthwave / acoustic band)
Vocal attitude (breathy / confident / soulful)

3. Use structure when you have lyrics

A simple verse–chorus–verse–chorus–bridge–chorus layout often makes the output feel more intentional.

Where It Shines (And Where It Doesn’t)

A believable review includes constraints. Here are the ones I noticed most.

Strengths

Momentum: you can keep moving instead of “stuck at zero.”
Variety: the tool is good at offering multiple creative directions quickly.
Accessibility: you don’t need a studio mindset to hear your concept.

Limitations

Prompt quality matters: vague prompts can produce generic tracks.
You may need multiple generations: the first result isn’t always “the one.”
Vocals can be hit-or-miss depending on style: some genres feel more stable than others.
Consistency across a series: if you need five tracks that match perfectly, you’ll likely do extra iterations.

This is why I treat AI music as a *draft engine*, not a final promise.

A More Grounded Way to Think About “Quality”

It helps to measure “quality” differently:

Not “Is this perfect?”
But “Is this good enough to spark the next decision?”

If a generation gives you one great chorus melody, a usable groove, or a surprising chord movement, it has already paid for itself in creative value.

Closing Thought: The Most Useful Outcome Isn’t a Finished Song

The best outcome is that you stop losing ideas.

When an AI tool can turn your messy notes into something you can actually listen to, you’re no longer guessing what your idea might become—you’re collaborating with it, and iterating with your ears instead of your imagination.

How I Generate Local Business Leads Without Cold Calling

Ashley Smith — Fri, 30 Jan 2026 13:32:37 +0000

Cold calling used to be my least favorite part of running a marketing agency. Hours spent dialing numbers, getting hung up on, leaving voicemails that never got returned. Then I discovered that local business lead generation doesn't have to feel like pulling teeth.

The shift happened when I stopped treating lead generation as a numbers game and started approaching it like a developer approaches a problem: systematically, with the right tools, and with data driving every decision.

Understanding What Local Businesses Actually Need

Local businesses aren't looking for fancy marketing jargon or complex attribution models. They want one thing: customers walking through their door or calling their phone. That's it.

When I started focusing on hyper-local strategies, my conversion rates jumped dramatically. A plumber in Austin doesn't care about national SEO rankings. They care about showing up when someone in their zip code searches "emergency plumber near me" at 2 AM.

This geographic focus changes everything about how you approach lead generation. Instead of casting a wide net, you're using a spear. And that precision requires knowing exactly who you're targeting.

Building Your Prospect List the Smart Way

The foundation of any local lead generation campaign is knowing who you're trying to reach. I used to spend hours manually copying business information from directories and Google searches. It was tedious and error-prone.

Now I automate the research phase entirely. When I need to build a list of potential clients in a specific area, I use this tool to pull business data directly from Google Maps searches. Instead of copying and pasting for hours, I get a spreadsheet with hundreds of prospects, complete with phone numbers, websites, and review counts.

This isn't about spamming businesses. It's about having accurate data so you can segment intelligently. I can identify businesses with no website, businesses with poor reviews who might need reputation management, or newer businesses that might not have established marketing yet.

Verification Before Outreach

Here's a mistake I see constantly: people build a list and immediately start blasting emails. Then they wonder why their domain gets flagged for spam.

The problem is data decay. That email you scraped might be outdated. The phone number might be disconnected. Before reaching out to anyone, verify your data.

I run every email through verification before adding it to my outreach sequence. There are several ways to do this, but I've found this free service works well for spot-checking emails and finding alternative contact methods when the primary one doesn't work.

This extra step might seem tedious, but it protects your sender reputation and ensures you're not wasting time on dead ends.

Multi-Channel Approach That Actually Works

Email alone won't cut it for local businesses. They're busy running their operations, and a cold email often gets lost in the shuffle.

My most successful campaigns combine multiple touchpoints:

Personalized email with specific value proposition
LinkedIn connection request (if they're active there)
Follow-up call referencing the email
Direct mail piece for high-value prospects

The key is personalization at scale. I'm not sending the same generic message to every prospect. I segment by industry, business age, current marketing presence, and specific pain points.

For example, restaurants without online ordering systems got very different messaging during and after the pandemic than established ones with robust digital presence.

Creating Offers They Can't Ignore

Local businesses are skeptical of marketing agencies, and rightfully so. Many have been burned by previous promises that didn't deliver.

Your offer needs to reduce risk. Here are approaches that have worked for me:

Performance-based pricing (you only pay when we deliver results)
Free audit or assessment (show them what they're missing)
Pilot program at reduced cost (prove value before full commitment)
Money-back guarantee for the first month

I typically start with a free local SEO audit. I'll analyze their Google Business Profile, check their citations for consistency, look at their review velocity compared to competitors, and identify quick wins.

This audit becomes the foundation for the proposal. Instead of generic "we can help your business grow" messaging, I'm showing them specific problems and specific solutions.

Choosing Your Tools Wisely

The martech landscape is overwhelming. There are thousands of tools promising to revolutionize your lead generation process.

I've wasted money on plenty of them. The truth is, you don't need a massive stack to get started. You need tools that actually solve specific problems in your workflow.

When evaluating new tools, I check resources like this that break down the actual features and pricing without the marketing fluff. It saved me from several expensive mistakes where the sales demo looked great but the actual product was buggy or missing key features.

Tracking What Actually Matters

Vanity metrics will kill your local lead generation efforts. Impressions and clicks mean nothing if they're not converting to customers for your clients.

For local campaigns, I track:

Phone calls generated (call tracking is essential)
Form submissions with actual contact information
Foot traffic increases (for brick-and-mortar)
Booked appointments or consultations
Actual closed deals (when client shares this data)

Every campaign should have clear attribution. When a plumber tells me they're getting more calls, I need to know if those calls are coming from the Google Business Profile optimization, the local directory listings, or the content strategy.

Scaling Without Losing Quality

The temptation when you find something that works is to immediately scale it 10x. This usually backfires with local lead generation.

Local businesses value relationships and personalized service. If you scale too fast, you lose the quality that made your approach work in the first place.

I've found the sweet spot is onboarding 3-5 new local clients per month. This gives me time to properly customize their strategy, deliver results, and use those results as case studies for the next round of outreach.

Your existing clients become your best lead generation tool. A happy plumber will refer you to other plumbers. A successful restaurant owner will introduce you to the owner of the place next door.

Final Thoughts

Local business lead generation isn't about having the fanciest tools or the biggest budget. It's about understanding your market, being systematic in your approach, and delivering actual results.

Start small, test your processes, and scale what works. The businesses that succeed in this space are the ones that treat lead generation as a repeatable system rather than a one-off campaign.

And most importantly, remember that behind every business listing is a real person trying to build something. Approach them with genuine value, not just another sales pitch.

How Content-Driven Websites Organize Information in Very Different Ways

Ashley Smith — Thu, 29 Jan 2026 14:08:56 +0000

When browsing content-driven websites, especially those focused on technology or online work, it becomes clear that there is no single “correct” way to organize information. Even sites covering similar topics can feel completely different once you spend a bit of time navigating them.

Some websites prioritize explanation. Others prioritize structure. Some feel like evolving notebooks, while others resemble carefully maintained libraries. These differences aren’t always obvious at first glance, but they shape how the content ages and how often people return.

Below are a few patterns I’ve noticed while looking at various content-oriented projects across the web.

Content as a Self-Contained Reference

Some content websites are built around the idea that each page should stand on its own. Articles are long, detailed, and written to minimize the need for additional context. Readers are expected to land on a page, get everything they need, and move on.

This approach works well for one-time searches, but it can make the site feel heavy when viewed as a whole. Navigation becomes secondary because the page itself is the product.

Content as a Connected System

Other sites treat individual articles as components of a larger structure. Pages are shorter, more focused, and heavily connected through internal links. Understanding the topic requires moving between sections rather than staying in one place.

These sites often feel lighter and more flexible. Instead of expanding a single article endlessly, new ideas are added by creating new nodes within the system.

Narrow Scope, Clear Boundaries

Some content-driven projects deliberately limit what each page is responsible for. Topics are separated by intent rather than combined for completeness. Foundational concepts live in one place, while applied or contextual discussions live elsewhere. YLSSEO is an example of this type of structure.

Rather than trying to consolidate every related idea into one expanding article, content appears to be organized around clearly defined boundaries. Pages don’t attempt to answer every possible question, but they remain stable over time because their scope is controlled.

This kind of structure often reflects a long-term mindset, where content is expected to evolve gradually rather than be constantly rewritten.

Narrative-First Content

Some websites feel closer to journals or essays than reference libraries. Structure exists, but it serves the narrative rather than the other way around. Topics may overlap, revisit earlier ideas, or drift slightly over time.

These sites are rarely optimized for fast answers, but they tend to leave a stronger impression. Readers don’t just consume information; they follow the author’s thinking.

Hybrid Approaches

Many content websites don’t fit neatly into a single category. They mix reference pages with opinion pieces, structured guides with informal notes. At first, this can feel inconsistent, but over time it often mirrors how real projects develop.

As a site grows, structure becomes less about perfection and more about maintainability.

Closing Observation

What separates content-driven websites isn’t the topic they cover, but how intentionally they decide what belongs where. Structure quietly communicates priorities—whether the goal is teaching, documenting, reflecting, or simply recording progress.

As more content is produced automatically and optimized for immediate visibility, these structural choices may become one of the clearest signals of how a site is meant to be used over time. Sometimes, content doesn’t need to say more. It just needs to be placed more carefully.

Building a Location-Based Loan Comparison Tool: Lessons from the Fintech Trenches

Ashley Smith — Fri, 09 Jan 2026 15:41:19 +0000

If you've ever tried to build anything in fintech, you know the rabbit hole goes deep. I recently worked on a project that matches users with lenders based on their location and credit profile. Here's what I learned.

The Problem

Most loan comparison sites treat the US as one big market. But lending is hyperlocal. Interest rates, lender availability, and even regulations vary wildly by state and city.

Someone searching for bad credit loans Atlanta has completely different options than someone in rural Montana. Georgia has specific payday lending laws. Local credit unions serve specific zip codes. Online lenders have state-by-state licensing.

Ignoring location means showing users options they can't actually use.

The Tech Stack

For the MVP, I kept it simple:

// Basic structure

const userProfile = {

creditScore: 620,

location: {

city: 'Atlanta',

state: 'GA',

zip: '30301'

loanAmount: 5000,

purpose: 'debt_consolidation'

};

The matching algorithm weights three things:

Lender availability - Does this lender operate in the user's state?
Credit requirements - Does the user meet minimum thresholds?
Loan terms - APR ranges, amounts, repayment periods

Database Design

The lender table got complicated fast:

CREATE TABLE lenders (

id SERIAL PRIMARY KEY,

name VARCHAR(255),

min_credit_score INT,

max_credit_score INT,

states_available TEXT[], -- Array of state codes

min_loan_amount DECIMAL,

max_loan_amount DECIMAL,

apr_min DECIMAL,

apr_max DECIMAL

);

The states_available array was key. Some lenders operate in 50 states. Others only serve 12. A few are single-state only.

The Matching Query

SELECT * FROM lenders

WHERE $1 = ANY(states_available)

AND min_credit_score <= $2

AND min_loan_amount <= $3

AND max_loan_amount >= $3

ORDER BY apr_min ASC;

Simple, but effective. Users see only what they actually qualify for.

Lessons Learned

Compliance is everything. Every state has different disclosure requirements. California alone has multiple regulatory bodies overseeing different loan types.
Data freshness matters. Lender terms change weekly. I built a scraper to monitor partner pages, but manual verification still catches things automation misses.
Mobile-first isn't optional. 73% of our traffic comes from phones. People search for loans during lunch breaks, not at desktop computers.
Trust signals convert. Adding real user reviews and BBB ratings increased click-through by 40%.

What's Next

Currently exploring:

ML model to predict approval likelihood
Real-time rate API integrations
Personalized recommendations based on similar user profiles

The fintech space moves fast, but the fundamentals stay the same: help users find what they actually need, where they actually are.

What fintech projects are you working on? Drop a comment below.