Law Firm CRM Setup: How to Track Leads From First Call to Signed Case
Your law firm may be getting calls, forms, chats, and ad leads, but can you clearly see what happened to…
So you want to know what’s actually inside a working AI system, not the marketing-deck version? Fair. We get this question almost every week from attorneys who’ve seen us talk about Janice, our in-house assistant, and want to understand the ai automation tools stack we use to run her day to day. Most of them aren’t trying to build their own. They just want to know what’s real and what’s vapor before they hire someone.
Janice isn’t a single product. She’s a stitched-together system that researches topics, drafts content, posts to social, pulls data from Google, and flags things for us to review. The pieces aren’t exotic. What matters is how they connect and who’s babysitting the connections.
We’re going to walk through every tool we leaned on in our ai automation tools stack, why we picked it, and what we learned breaking things along the way. If you’re an attorney thinking about automation for your firm, this should give you a clearer picture of what’s involved before someone pitches you a “fully automated marketing system” with no detail behind it.
This post covers the core ai automation tools stack, how each piece fits, what we’d swap out today, and what attorneys should ask before buying into anyone else’s system.
We tried the off-the-shelf route first. Most of the all-in-one AI marketing platforms we tested were either too generic for legal work or boxed us into workflows that didn’t match how we actually run client accounts at Best Law Firm Ads. The output read like every other agency’s output. Not useful when the whole point is helping a firm stand out.
Building Janice gave us control over prompts, data sources, and review steps. When a personal injury client needs a different tone than an estate planning client, we can fork the workflow instead of fighting a SaaS settings panel. That flexibility was worth the build time.
The other reason was cost. Stacking five or six SaaS subscriptions adds up fast, and most of them duplicate features we already pay for through Google Workspace and our developer accounts. A custom ai automation tools stack let us pay for raw capability instead of someone else’s UI.
We’re not saying every firm should build their own Janice. Most shouldn’t. But understanding what’s under the hood of an ai automation tools stack helps you spot the difference between a real system and a thin wrapper around ChatGPT.
Before we go deep on each piece, here’s the short list of what we use:
Each one handles a job the others can’t do as well. Claude doesn’t search the live web reliably. Serper doesn’t write. Sheets isn’t a real database. PowerShell doesn’t think. The whole point of the ai automation tools stack is matching the right tool to the right step in a workflow.
Quick note on philosophy. We try to keep the ai automation tools stack boring. Boring tools have documentation, communities, and predictable failure modes. Trendy tools break, get acquired, or pivot mid-quarter. When you’re running automations that touch client accounts, boring beats clever every time.
Claude does most of the thinking in our system. We use it for content drafting, summarizing research, rewriting copy in a specific firm’s voice, and reviewing outputs from other steps. The reasoning quality on longer, nuanced tasks has been better for legal content than what we got from other models in our testing.
We hit the API directly instead of the chat interface for anything automated. That lets us pass structured prompts, set temperature, and chain outputs into the next step. For attorney-facing work, we keep temperature low so the writing stays grounded and predictable.
A practical example. When Janice drafts a blog for a family law client of ours in the Dallas market, the workflow pulls competitor headlines from Serper, sends them to Claude with a tone guide we built from the client’s intake interviews, and returns a draft we review before anything goes live. No model writes publish-ready legal content on its own. We’ve never seen it, and we wouldn’t trust it if we did. Anthropic publishes usage guidance worth reading before you build production workflows on top of any large language model.
The tradeoff with Claude inside an ai automation tools stack is cost per token on long context windows. We watch that closely and cache prompts where we can.
Serper is a Google search API. We use it to pull SERP data, “People Also Ask” questions, related searches, and competitor titles. It’s the cheapest reliable way we’ve found to get live Google results into an automation without scraping ourselves.
Here’s why this matters. Any AI that drafts content without seeing what’s already ranking is guessing. Serper lets Janice check the current top ten for a target keyword, pick out common themes, and find gaps. The drafts that come out the other side are more grounded because the model isn’t writing in a vacuum.
We also use Serper for monitoring. When a criminal defense client of ours wanted to track how often their firm name showed up in local search for specific intent keywords, we set up a daily pull that logs results to a sheet. Took maybe an hour to wire up. That kind of visibility used to require a paid rank tracker subscription per location.
The limitation is that SERP data is a snapshot. Google personalizes and localizes, so what Serper returns isn’t always what a user in a specific zip code sees. We account for that with location parameters and by treating the data as directional.
Every automation needs a place to store inputs, outputs, and state. We use Google Workspace for most things and Supabase when we outgrow Sheets.
Sheets and Docs handle more than people give them credit for. For most of our internal Janice workflows, a Sheet with tabs for keywords, drafts, status, and review notes is plenty. The Apps Script layer lets us trigger actions on edits, send emails on status changes, and read from external APIs. For a small firm running maybe 10 to 20 automations, you probably never need anything more in your ai automation tools stack.
Sheets is also where non-technical team members can actually see and edit what the system is doing. That matters. A database nobody can read is a database nobody trusts.
Once a workflow has thousands of rows, relational data, or needs to serve a frontend, Sheets falls over. Supabase gives us a Postgres database with an API on top, plus auth and storage if we need them. We reach for it when we’re building anything client-facing, or when we’re tracking enough events that Sheets becomes slow.
Honest answer on which to pick: start with Sheets. Move to Supabase only when you have a real reason. Premature database adoption is one of the most common ways automation projects stall before they ship anything useful.
PowerShell isn’t glamorous. It’s also the tool we reach for more than almost anything else in the ai automation tools stack. We use it for scheduled jobs, file management, calling APIs from a local machine, and stitching together steps that don’t have a clean SaaS connector.
A typical use: every morning at 6 AM, a PowerShell script pulls overnight form submissions from a client’s CRM via API, formats them, drops a CSV into a shared drive, and pings our Slack with a summary. That whole workflow is maybe 80 lines of code and runs on a Windows VM that costs almost nothing per month.
Could we rebuild that in Zapier or Make? Sure. We’ve done both. The downside is per-task pricing at scale, and the upside of a no-code tool evaporates once you have a developer on the team who can just write the script. Microsoft’s official PowerShell documentation is solid if you’re starting out or auditing someone else’s scripts.
For attorneys reading this who don’t have technical staff, the point isn’t that you need PowerShell. The point is that any agency building a real ai automation tools stack for you should have some scripting layer doing the boring work. If it’s all Zapier zaps, ask what happens when one breaks at 2 AM.
The last piece of the ai automation tools stack is the network of API connections that lets everything talk. We connect to Google (Search Console, Business Profile, Ads, Analytics), Meta (Pages, Ads), LinkedIn, and a handful of analytics platforms. Each connection is a small bridge that lets Janice read or write data without a human logging in.
Search Console for ranking and impression data. Business Profile for posting updates and pulling review activity. Analytics for traffic patterns. Ads for campaign performance. Those four are the ones we touch most. Google’s API docs are dense but reliable, and authentication is more painful than it should be. Plan for that.
Meta and LinkedIn APIs are useful for posting and for pulling basic engagement data, though both platforms have tightened access over the past few years. CRM connections depend on what the client uses. We’ve built integrations against Clio, Lawmatics, and a few generic Zapier-style bridges depending on what was already in place.
The rule we follow: never store credentials in a script. Use a secrets manager, rotate keys, and assume the API will break a feature you depend on at the worst possible time. Build your ai automation tools stack with that assumption baked in.
If we were starting over tomorrow, a few things would be different. We’d probably move to Supabase earlier for anything touching more than one client. We’d build proper logging from day one instead of bolting it on after we missed a failure for two days. We’d standardize prompt templates across all Claude calls instead of letting them drift per workflow.
We’d also be more aggressive about deleting automations that don’t earn their keep. It’s easy to build something clever, ship it, and forget it’s running. Every piece of an ai automation tools stack has a maintenance cost, even the ones that work. We do a quarterly review now where we kill anything that isn’t producing measurable value.
One thing we wouldn’t change: keeping humans in the loop for anything client-facing. No matter how good the ai automation tools stack gets, attorney content has compliance and tone implications that need a human pass. The system drafts. We approve. That’s the rule.
If you want to see how this thinking applies to attorney visibility specifically, our work on AI visibility for law firms covers how generative search is changing what shows up when a potential client asks ChatGPT for a lawyer.
You don’t need to build Janice. You need to know what questions to ask whoever is building or selling you automation. A few worth asking about any proposed ai automation tools stack:
If the answers are vague, the system is probably vague. We’ve seen firms pay monthly for “AI marketing automation” that turned out to be one Zapier flow and a Buffer subscription. That’s not automation. That’s a scheduled tweet.
The other thing worth knowing: automation isn’t a replacement for strategy. The ai automation tools stack we built only works because we already knew what we wanted Janice to do. The tools follow the plan, not the other way around. Plenty of solid background on responsible AI use in business contexts is available through Google’s AI principles documentation if you want to dig deeper on the policy side.
If you’re scoping an ai automation tools stack for your firm, our team at Best Law Firm Ads usually starts with a few foundational pieces before any automation goes in. That includes solid law firm SEO, a working analytics setup, and a clear sense of what counts as a qualified lead. Without those, automation just produces more noise, faster.
An ai automation tools stack is the collection of software, APIs, and scripts that work together to run automated workflows. In our case, that means Claude for reasoning, Serper for search data, Google Workspace or Supabase for storage, PowerShell for scripting, and various API connections for social and search platforms.
Probably not at the level we built ours. Most small to mid-sized firms benefit more from one or two well-chosen automations (intake follow-up or review requests, for example) than from a full custom system. The right question is what specific bottleneck you’re solving, not whether you need “AI.”
Varies a lot based on usage. Our API costs across Claude, Serper, and Google services run in the low hundreds per month for our internal use. Development time is the bigger cost. If you’re hiring someone to build an ai automation tools stack, budget for ongoing maintenance, not just the initial build.
Yes, for a while. Zapier, Make, and Airtable can handle a lot before you outgrow them. We use no-code tools too. The reason we eventually moved to scripting and Supabase was per-task pricing and the ceiling on what no-code platforms let you do. Start no-code, graduate when you need to.
In our testing, Claude has been stronger on longer, nuanced drafts where tone and accuracy matter. ChatGPT is faster for short tasks and has better image and voice features. We use both. For legal content specifically, we lean Claude, but we still review every output before it goes anywhere near a client account.
The first working version of Janice took us about six weeks of part-time work. We’ve been refining the ai automation tools stack for over a year since. Most of the improvement came from running it on real client work and fixing things that broke. You can’t shortcut that learning loop with planning.
Automating something that wasn’t working manually. If your intake process is broken, automating it just breaks it faster at scale. We always tell firms to get the underlying workflow solid first, then layer an ai automation tools stack on top. The tool doesn’t fix the strategy.
Ask for specifics. What models, what tools, what does the workflow look like, what’s the review process, how do they handle errors? If the answers are buzzwords without detail, that’s a sign. A real ai automation tools stack has named components and traceable outputs. Vague “AI-powered” claims usually mean a ChatGPT subscription and good marketing copy.
If you’ve read this far, you’re probably either evaluating an agency or thinking about what automation could actually do for your firm. Either way, a short conversation is usually the fastest way to figure out what makes sense for your situation. We’ll tell you straight if an ai automation tools stack is the right next move, or if something more basic should come first.
Book a free consultation and we’ll walk through what your current marketing setup looks like, where automation could realistically save you time, and what kind of ai automation tools stack would actually fit your firm’s size and goals. No pressure, no canned pitch.
