Why standards‑based AI is an R&D opportunity for SaaS founders (and a strategic talking point for sell‑side advisers)

From Screen‑Scraping Bots to Model Context

I argued last week that founders face a strategic fork: realise the value already built or reinvest for compound advantage.

I thought it would be useful to talk about how Ishikawa Technologies pivots product strategy into practical R&D, moving from brittle screen automation to protocol‑driven SaaS workflows that are measurable, governable and priced on outcomes.

Why this is an R&D moment, not a plug‑in moment

The sudden rise of large‑language models makes it feel as though every workflow is one prompt away from full automation. In practice, turning ‘reconcile yesterday’s bank feed’ into a dependable service involves experimentation, testing and governance. This is classic research and development. The upside is huge: software that does the work, rather than merely guides it. But the path is iterative, not turnkey.

The mirage of consumer AI

Public demos usually show a model driving a web browser: clicking menus, copying figures, emailing results. Entertaining, yes, but one stylesheet change or pop‑up and the sequence collapses. That fragility is fine for hobby searches; it is intolerable when statutory accounts must balance to the penny.

“Agents look brilliant... until a cookie banner hides the button they need.”— Our Product Samurai, over coffee last week

Meet the Model Context Protocol (MCP)

Think of MCP as USB for AI: a single open specification that lets models discover data, invoke tools and keep context without screen‑scraping. In MCP there are two roles:

• MCP Server – exposes data and safe‑to‑run‑again actions (eg bank‑reconcile, raise‑invoice).

• MCP Client – the AI assistant that consumes those actions.

Because everything travels in a standard JSON envelope, you wire a connector once and use it in every model that speaks the protocol. Security, state and capability negotiation are baked in. The result: fewer brittle “glue” scripts and a clearer audit trail.

Turning workflows into safe‑to‑run‑again functions

For an AI assistant to handle real work it needs three things:

1. A callable action – exposed through a documented API or MCP capability.

2. Domain rules encoded in code – so the model executes policy rather than inventing it.

3. Human checkpoints – approvals and rollbacks for the edge‑cases it cannot settle.

Those steps involve schemas, test suites and log reviews—hence the R&D label. No‑code tools rarely provide the depth of control required.

A hypothetical example: daily bank reconciliation in Big Red Cloud

Big Red Cloud already provides a Bank Reconciliation wizard that guides users through five repeatable steps:

1. Post all receipts and payments for the period you’re reconciling.

2. Clear any outstanding lodgements or cheques carried over from the previous period.

3. Confirm the statement end date and closing balance against the physical (or online) bank statement.

4. Run the wizard, which auto‑matches ledger entries to bank lines and flags anything it can’t pair.

5. Generate the Reconciliation Report and lock the period once every item is accounted for.

   
   

Suppose we wrapped that flow in an MCP Server:

• Capability: bank.reconcile

• Required context: bank_account ID, statement_end_date, statement_end_balance, confidence_threshold (how certain the AI must be before auto‑matching)

• Response: list of matched transactions, list of exceptions for manual review, reconciliation_report URL, success_state

An MCP Client could trigger reconciliation at 06:00 each morning, present the unmatched lines to the bookkeeper at 09:00, and flag success when every entry balances and the report is archived. No screen‑scraping, no brittle selectors... just deterministic automation built on the exact steps Big Red Cloud documents.

R&D and positioning: a shared checklist

Founders – your R&D essentials

• Catalogue the jobs‑to‑be‑done that consume the most customer time.

• Expose each as a safe‑to‑run‑again endpoint (if a call fails midway it can be run again without side‑effects).

• Create a sandbox MCP connector and measure success and failure rates weekly.

• Frame pricing in plain terms: customers pay only when the day's bank feed reconciles cleanly, not per user licence.

Advisers – your valuation lens

• Quantify the uplift. Map how outcome‑based pricing and daily automation expand total addressable market and boost lifetime value.

• Craft the story. Position MCP‑enabled workflows as a proprietary capability that merits multiple expansion when pitching to strategic or financial buyers.

• Evidence the moat. Gather sandbox metrics (success rates, exception ratios, user hours saved) to satisfy technical due‑diligence queries.

• Highlight the platform angle. Show how the same protocol can connect adjacent products, creating synergy upside for acquirers.

• Quote the numbers. Our early sandbox shows 99% auto‑match, saving SMEs hours - an easy headline for buyer decks.

  
  

Strategic implications

• Outcome‑based pricing becomes feasible: think “£10 per successful auto‑reconciliation” rather than “£30 per seat”.

• Lower churn and deeper moat: customers who let the software do the work are far stickier than those who merely use it.

• Portfolio synergy: once your products speak MCP, one assistant can hop from bookkeeping to payroll to analytics without rewiring.

In short, screen‑scraping bots make flashy videos, but protocol‑driven automations make defensible businesses. Approach the opportunity with the diligence of an R&D programme, and you will be ready when ‘reconcile today’s bank feed’ becomes as routine as pressing Send/Receive in your e‑mail client or not even having to press it at all.

A quiet word for time‑poor founders. If your roadmap already feels oversubscribed and the prospect of mastering another protocol is one leap too many, you are not alone. Ishikawa couples investment with a seasoned engineering guild, allowing ambitious teams to capture the AI upside without hitting pause on core growth. When you are curious about what that could look like, let’s simply compare notes—no slide decks, no hard sell. sourcing@ishikawatech.com