Build our own vertical copilot — or buy from a category vendor?
Sales copilots ($30-50K/year), legal copilots ($100K+/year), support copilots are all sold against the build option. Three years of AI tooling means building from scratch is no longer a 2-year project — and no longer obviously cheaper.
The question
We need a copilot for one of our core operational functions — sales, support, legal, or recruiting. The category has matured: vertical vendors now sell complete solutions at $30-100K/year. Build-from-scratch on our existing AI stack is feasible in 2-3 engineer-months. Do we buy the vertical product, build on our own stack, or hybrid (vendor + customization)?
The premise
- Team
- ~50 engineers, ~10 actively building AI features, single MLOps engineer. AI work pulls from feature-shipping capacity — any new commitment has to trade against the roadmap. A 'build' option requires 2-3 engineer-months upfront + ~0.5 FTE ongoing for the first year. Pulls from feature delivery.
- Compliance
- SOC2 Type II in scope. EU customer data subjects us to GDPR plus the EU AI Act's August 2026 GPAI-deployer obligations. Vertical copilots often process customer data — vendor DPA + GDPR + SOC2 alignment is a hard prerequisite, not nice-to-have.
- Stack
- Our internal AI stack: OpenAI Embeddings, LangGraph for orchestration, Postgres for vector storage, custom retrieval. We could reuse ~60% of this for a sales/support copilot but the workflow logic + integrations (CRM, ticketing, calendar) are net-new — roughly 3 engineer-months of work.
- Budget
- Monthly AI spend ~$30K with quarterly board visibility. Approvals required for sustained jumps >20%. Cost-per-outcome metrics in place; finance asks for unit economics by use case. Vertical vendor list price ~$30-100K/year. Build = ~$50K of engineering time year 1, ~$20-30K/year ongoing.
How mature are the vertical vendors really?
Sales (Apollo, Outreach, Clay) and support (Decagon, Sierra) are 18-24 months ahead of legal (Harvey, EvenUp) and recruiting (Paradox, Mercor). For mature categories, vendor is hard to beat on feature completeness. For newer categories, build-on-our-stack closes the gap because vendor features are still thin and our domain logic is the differentiator.
What's the lock-in cost if we go vendor and they pivot or raise prices?
Moderate. Workflow data is exportable from most reputable vendors; integration cost (CRM connectors, custom rules, training data) is the real lock-in. Realistic switching cost: 4-6 engineer-weeks if data is portable + a quarter of degraded performance during ramp. The fear of price-raises is mostly priced in — their leverage is your switching cost, not the price tag they wave at you.
When is build clearly the right call?
When (a) the use case is so specific to our domain that vendors will never prioritize it, OR (b) we already have 60%+ of the underlying primitives (retrieval, embeddings, workflow engine), OR (c) the data we'd send to a vendor crosses a compliance threshold we won't accept. Otherwise vendor wins on time-to-value, especially in mature categories.
Counsel's position
Build the vertical copilot on your existing LangGraph stack to avoid the severe vendor lock-in and usage-based price hikes of vertical SaaS, but instrument query-level cost attribution from day one to satisfy finance's unit economics requirements.
Verdict
The verdict: Design your copilot for capability expansion rather than mere acceleration.
Design your copilot for capability expansion rather than mere acceleration
Given your decision to deploy a copilot for your operations team, focus on redesigning the workflow to change when information becomes available rather than just speeding up existing tasks.
Instrument query-level cost attribution before scaling your custom copilot
Given your strict $30K/month AI budget and finance's demand for unit economics, implement per-tenant usage tracking from day one to prevent massive budget overruns.
Read another verdict
- Build agents that own workflows — or workflows that own LLM calls?
- Set our LLM data retention policy now, or wait for an incident to force it?
- Roll AI coding tools across the whole engineering org — and how do we measure it?
- Standardize the team on one agent framework, or let each pod pick?
- Kill every AI pilot that can't show ROI in 90 days?
- Use AI to flatten middle management this year?
- Stand up a FinOps practice for tokens and GPUs now?
- Replace customer support with AI — or avoid the Klarna outcome?