Most clients walk into discovery with 7 to 15 SaaS subscriptions and a feeling that something has to give. The bills add up. The data is split across systems. New employees take three weeks to learn the tools. Generic AI features tacked onto each platform are mostly cosmetic. The owner asks, half-joking, whether they should just throw it all out and start over.

The honest answer is rarely "yes." It is also rarely "no." It is "for each tool, here is whether we connect to it or recreate the feature inside your custom app." That decision happens once per SaaS, on a five-question framework we run on every discovery call.

The Five Questions

Run each SaaS through these five questions. If three or more answers point toward "recreate," the math almost always works. If three or more point toward "connect," the SaaS stays and we integrate.

Question 1: How standardized is the workflow?

If the workflow is identical across every business that uses the SaaS, that is a strong vote for keeping the SaaS. Payment processing is the cleanest example. Stripe processes a credit card the same way for everyone. The compliance, fraud, and global rails are commodity infrastructure. We connect.

If the workflow is highly customized to your business (your matter intake, your project status reporting, your role-specific dashboards), that is a strong vote for recreate. The SaaS is forcing you to bend your process to fit the tool.

Question 2: What is the per-seat cost relative to recreation cost?

This is the math conversation. A 50-person company paying $80 per seat per month for a CRM is spending $48,000 per year. If recreating that CRM as part of a custom app costs $35,000 in build, the payback is under a year and the recurring cost falls dramatically afterward. The arithmetic moves the decision.

For low-cost-per-seat tools, the math points the other way. Twilio at fractions of a cent per message is impossible to beat by rebuilding messaging infrastructure. Connect.

Question 3: Does the SaaS lock in your data?

If exporting your data is hard, slow, or requires the vendor's blessing, that is a structural risk. SaaS vendors get acquired. They raise prices. They sunset features you depend on. Data lock-in is one of the most underweighted reasons to migrate to a system you control. We have had clients discover, mid-migration, that their five-year SaaS history is held hostage by an export format that does not actually round-trip.

14
Typical SaaS subscriptions in a small-to-mid-sized business stack we walk through during a connect-or-recreate discovery.

Question 4: Does AI add disproportionate value if we recreate?

Some categories become dramatically better when AI is in the loop and badly served by the generic AI features bolted onto SaaS. Email automation, customer support, knowledge search, intake triage, and document processing all fall here. Recreate, because the AI-augmented version is a different and better product.

Other categories do not benefit much from AI. Stripe with AI is not meaningfully better than Stripe. EDI infrastructure with AI is still EDI infrastructure. Payment-card issuance with AI is still payment-card issuance. Connect.

Question 5: Is there a deep-domain industry vendor we cannot match?

Some SaaS companies have 15 to 30 years of vertical knowledge baked in. Procore knows construction. Famous Software knows produce distribution. Lawcus knows legal practice management. AppFolio knows property management. NetSuite or SAP knows manufacturing ERP. athenahealth or Epic knows clinical workflows. Shopify knows ecommerce checkout. We are not going to outbuild any of them, and we should not pretend otherwise. Connect, surface their data inside your dashboard, and let AI add a layer on top that they do not ship natively.

A Worked Example: A 14-Tool SMB Stack

Imagine a typical professional services firm walks in with this stack: HubSpot, Asana, Calendly, DocuSign, Mailchimp, Intercom, Loom, Dropbox, JotForm, SurveyMonkey, Harvest, Stripe, QuickBooks, and a separate analytics dashboard. Run each through the five questions.

HubSpot. Generic-purpose, $50 per seat, AI adds disproportionate value, no deep domain lock. Recreate.

Asana. Generic-purpose project management, per-seat pricing, AI adds value through agentic status reporting. Recreate.

Calendly. Single narrow feature, easy to rebuild, branding and routing improvements meaningful. Recreate.

DocuSign. Per-envelope pricing, the audit log and signature capture are commodity, in-portal signing is better UX. Recreate.

Mailchimp. Email automation, AI drafting in your voice is a step-change improvement, per-contact pricing scales painfully. Recreate.

Intercom. Live chat and support, AI scoped to your business is dramatically better than the generic chatbot. Recreate.

Loom. Video record, easy to rebuild with AI transcript baked in. Recreate.

Dropbox. File storage, role-scoped Cloudflare R2 storage is cheaper and integrates with everything else. Recreate.

JotForm. Form builder, in-dashboard forms are better UX with conditional logic and AI summarization. Recreate.

SurveyMonkey. Same logic as JotForm. Recreate.

Harvest. Time tracking, simple to rebuild with project context baked in. Recreate.

Stripe. Commodity payment rails, regulated, AI does not improve it meaningfully. Connect.

QuickBooks. Decades of accounting and tax compliance, AI does not improve it meaningfully, your accountant uses it. Connect.

Analytics dashboard. Replaceable with role-rendered views inside the custom app. Recreate.

Result: 12 of 14 tools recreated, 2 connected, one custom dashboard, and a software bill that drops by roughly 70 percent in year two while the AI capability goes up by a multiple.

When Recreate Becomes Obvious

The recreate decision becomes obvious in two situations. First, when the cumulative SaaS bill is higher than what an engineer would cost to replicate the features, even before factoring in the AI uplift. Second, when AI adds 5 times the value on top, which it usually does for any category involving language, classification, summarization, or judgment.

The break-even point in 2026 is roughly three to four overlapping per-seat tools. Above that, custom usually wins on year-two cost alone, before counting AI capability.

When Connect-To Is the Right Call

Connect is the right call for the deep-domain vendors and the commodity-rails infrastructure. Shopify for ecommerce checkout. Stripe for payments. Procore for construction management. NetSuite or SAP for serious ERP. Lawcus for legal practice management. AppFolio for property management. athenahealth or Epic for clinical workflows. Twilio for messaging infrastructure. AWS, Azure, and Cloudflare for hosting.

The pattern: when the vendor is solving a problem we cannot reasonably outbuild (because of vertical depth, regulatory complexity, or commodity-scale economics), we connect. The custom app becomes the surface where their data lives next to everything else, with an AI layer they will never ship.

Running the Framework on Your Stack

If you want to start before the discovery call, list your SaaS subscriptions in a spreadsheet. For each one, score the five questions on a 1-to-3 scale (3 = strongly favors recreate, 1 = strongly favors connect). Total the columns. Anything above 11 is almost certainly a recreate. Anything below 8 is almost certainly a connect. The ones in the 8-to-11 range are where the conversation gets interesting and where the right call depends on what else is in your stack.

The full architecture pattern is on the Apps & Dashboards page. The blog series goes deeper on the specific features we recreate, the industry-specific replacement stacks, and what real bespoke actually looks like. Two case studies show the framework in action: the structural engineering build and the law firm employee dashboard.

The framework is not magic. It is just the discipline of running the same five questions across every tool in the stack instead of treating each one as a separate "do we keep this" debate. Once you have done that, the decision usually makes itself.